Identification of DNA repair dysfunction

Abstract

The invention provides methods of determining whether a tumour has a high or low likelihood of being PRR-deficient, based on the values of a plurality of mutational signature metrics for a DNA sample from the tumour. The plurality of mutational signature metrics comprises: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post- replication repair (PRR)-deficiency; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)-deficiency. A sample that has a high likelihood of being PRR-deficient has a high likelihood of having mismatch repair deficiency (MMRd), polymerase dysfunction (pol-dys), or combined mismatch repair deficiency and polymerase dysfunction (MMRd / pol-dys). Related methods, systems and products are also described.

Description

[0001] IDENTIFICATION OF DNA REPAIR DYSFUNCTION

[0002] FIELD OF INVENTION

[0003] The present invention relates to a method for characterising the properties of cancer based on a DNA sample from a tumour. It is particularly, but not exclusively, concerned with a method for identifying whether the tumour is deficient for post-replicative repair (PRR), and methods for identifying a treatment or prognosis accordingly.

[0004] BACKGROUND TO THE INVENTION

[0005] Somatic mutations are a hallmark of cancer and can arise through both endogenous and exogenous processes. Endogenous processes that have been shown to give rise to DNA lesions include endogenous biochemical activities such as hydrolysis and oxidation, and errors at replication. Fortuitously, our cells are equipped with DNA repair pathways that constantly mitigate this endogenous damage. These include the DNA mismatch repair (MMR) pathway and replicative polymerase proofreading, together referred to as post-replicative repair. Biological abnormalities leading to post-replicative repair deficiency (PRRd) often display exquisite sensitivity to immune checkpoint inhibition (Cristescu et al. 2018; Wang et al. 2019; Das et al. 2022).

[0006] However, there is a lack of methods to accurately identify the full spectrum of patients that have PRRd.

[0007] STATEMENTS OF INVENTION

[0008] Algorithms to classify MMR-deficiency tumours have been developed using massively-parallel sequencing data (Huang et al., 2013; Zou et al. 2021). Most existing classifiers depend on detecting elevated tumour mutational burdens (TMB) or microsatellite instability (MSI). These are relatively crude metrics of genomic instability that are common manifestations of MMR deficiency but fail to identify many patients that are in fact MMR deficient. A notable exception to this is MMRDetect (Zou et al. 2021). This classifier predicts the MMRd status of a sample using one or more mutational signature metrics selected from: exposure of one or more mutational signatures of mismatch repair (in particular, RefSig MMR1 and RefSig MMR2), similarity between the substitution profile of the sample (i.e. the number or proportions of SBS in each of 96 categories as described further below) and that of one or more MMR gene knockouts, the number of repeat mediated indels in the mutational profile of the sample, and the similarity between the repeat mediated deletion profile of the sample and that of one or more MMR gene knockouts. However, this approach does not have the ability to identify PRRd more broadly, or to distinguish between different types of PRRd, such as PRRd due to pol-dys, MMRd or combined MMRd / pol-dys. The present inventors postulated that improved detection of PRRd could be obtained by combining information from both SBS signatures and indel signatures. Thus, according to a first aspect, there is provided a computer-implemented method of determining whether a subject has a tumour that is deficient for post-replicative repair, the method including the steps of, for a DNA sample obtained from said tumour: determining the values of a plurality of mutational signature metrics for the sample, wherein the plurality of mutational signature metrics comprise: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)-deficiency; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)-deficiency; and based on said values of said mutational signature metrics, determining whether said sample has a high or low likelihood of being PRR-deficient, wherein a sample that has a high likelihood of being PRR-deficient has a high likelihood of having mismatch repair deficiency (MMRd), polymerase dysfunction (pol-dys), or combined mismatch repair deficiency and polymerase dysfunction (MMRd / pol-dys).

[0009] Thus, the methods according to the present aspect may also be described as methods of characterising a DNA sample obtained from a tumour, the method including the steps of: determining the values of a plurality of mutational signature metrics for the sample, wherein the plurality of mutational signature metrics comprise: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)- deficiency; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)-deficiency; and based on said values of said mutational signature metrics, determining whether said sample has a high or low likelihood of being PRR-deficient, wherein a sample that has a high likelihood of being PRR- deficient has a high likelihood of having mismatch repair deficiency (MMRd), polymerase dysfunction (pol-dys), or combined mismatch repair deficiency and polymerase dysfunction (MMRd / pol-dys).

[0010] Due to the use of both SBS signatures of PRR deficiency and indel signatures of PRR-deficiency, the present methods are able to identify samples that have any of the above types of PRRd. This was previously not possible, as at most MMRd could be reliably identified.

[0011] The one or more indel signatures of PRR-deficiency may be selected from: one or more indel signatures associated with mismatch repair deficiency (MMRd), one or more indel signatures associated with polymerase dysfunction (pol-dys), and one or more indel signatures associated with combined MMRd and pol-dys (MMRD / pol-dys). The one or more indel signatures of PRR- deficiency may comprise one or more indel signatures associated with combined MMRd and pol- dys (MMRD / pol-dys). The one or more SBS signatures of PRR-deficiency may be selected from: one or more SBS signatures associated with mismatch repair deficiency (MMRd), one or more SBS signatures associated with polymerase dysfunction (pol-dys), and one or more SBS signatures associated with combined MMRd and pol-dys (MMRD / pol-dys). The one or more SBS signatures of PRR-deficiency may comprise one or more SBS signatures associated with pol-dys. The one or more indel signatures of PRR-deficiency may comprise one or more indel signatures associated with MMRd. The one or more SBS signatures of PRR-deficiency may comprise one or more SBS signatures associated with MMRd. Thus, the plurality of mutational signature metrics may comprise: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)-deficiency, wherein the one or more indel signatures of PRR-deficiency comprise one or more indel signatures associated with combined MMRd and pol-dys (MMRD / pol-dys) and optionally one or more indel signatures associated with MMRd; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)-deficiency, wherein the one or more SBS signatures of PRR-deficiency comprise one or more SBS signatures associated with pol-dys, and optionally one or more SBS signatures associated with MMRd. The one or more signatures may comprise at least one of (i) one or more indel signatures associated with MMRd, and (ii) one or more SBS signatures associated with MMRd.

[0012] The subject may be a human subject. The subject may have a cancer selected from: colorectal cancer, stomach cancer, uterine cancer, lung cancer, bladder cancer, and central nervous system cancer.

[0013] The plurality of mutational signature metrics may comprise: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)- deficiency, wherein the one or more indel signatures of PRR-deficiency comprise one or more indel signatures associated with combined MMRd and pol-dys (MMRD / pol-dys), one or more indel signatures associated with MMRd, and optionally one or more indel signatures associated with pol-dys; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)-deficiency, wherein the one or more SBS signatures of PRR-deficiency comprise one or more SBS signatures associated with pol-dys, one or more SBS signatures associated with MMRd, and optionally one or more SBS signatures associated with combined MMRd and pol-dys.

[0014] The method may further comprise determining the ratio of the number of indels to the number of SBS in the sample. Determining whether said sample has a high or low likelihood of being PRR- deficient may be based on said ratio or a value derived therefrom and said mutational signature metrics. Determining whether said sample has a high or low likelihood of being PRR-deficient may be based on said mutational signature metrics and a value derived from said ratio by applying one or more transformations selected from: log transformation, normalisation and adding a fixed number to the numerator and denominator. The plurality of mutational signature metrics may comprise: the proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)-deficiency; and the proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)-deficiency.

[0015] Determining whether said sample has a high or low likelihood of being PRR-deficient may comprise classifying the sample between a plurality of classes comprising: a class associated with a high likelihood of MMRd, a class associated with a high likelihood of pol-dys, a class associated with a high likelihood of combined MMRd / pol-dys, and a class associated with a high likelihood of PRR- proficiency. Determining whether said sample has a high or low likelihood of being PRR-deficient may comprise using a trained machine learning model that takes as input the values of said mutational signature metrics and produces as output one or more probabilities and / or classification labels indicative of whether the sample has a high or low likelihood of being PRR-deficient. The machine learning model may have been trained using training data comprising the values of said mutational signature metrics fora plurality of training samples with known PRR status. The plurality of training samples may each be associated with a label selected from: PRR proficient, MMRd, pol-dys, combined MMRd / pol-dys. The machine learning model may further take as input the ratio of the number of indels to the number of SBS in the sample or a value derived therefrom. The training data may further comprise, for each training sample, the ratio of the number of indels to the number of SBS in the sample or a value derived therefrom. The plurality of training samples may comprise samples from subjects with a tumour selected from colorectal, stomach and uterine cancer. The plurality of training samples may comprise at least 50, at least 100, at least 200, at least 300, at least 400 or at least 500 samples. The machine learning model may be a multinomial classification or regression model, or a regularised multinomial classification or regression model. The machine learning model may be a multinomial regression model trained with elastic net regularisation. Determining whether said sample has a high or low likelihood of being PRR- deficient may comprise: generating, by said machine learning model using said values of said one or more mutational signature metrics, one or more probabilistic scores; and based on said one or more probabilistic scores, determining whether said sample has a high or low likelihood of being PRR-deficient. This may be performed by comparing said one or more probabilistic scores with respective predetermined thresholds, wherein the probabilistic scores comprise at least a first probabilistic score that is such that the sample has a high likelihood of being PRR-deficient when the probabilistic score is at or below a threshold and a low likelihood of being PRR-deficient when the probabilistic score is above the threshold. The first probabilistic score may be a probability of the sample belonging to a class associated with a high likelihood of PRR-proficiency, and the method may comprise determining that the sample has a low likelihood of being PRR-deficient when said first probabilistic score for said sample is above the threshold, optionally wherein the threshold is 0.5. The machine learning model may produce as output: a first probabilistic score associated with a first class associated with a high likelihood of PRR-proficiency, a second probabilistic score associated with a second class associated with a high likelihood of MMRd, a third probabilistic score associated with a third a class associated with a high likelihood or pol-dys, and a fourth probabilistic score associated with a fourth class associated with a high likelihood of combined MMRd / pol-dys. The method may comprise determining that the sample has a low likelihood of being PRR-deficient when said first probabilistic score for said sample is above a threshold, or determining that the sample has a high likelihood of being PRR-deficient when said first probabilistic score for said sample is at or below the threshold. Determining that the sample has a high likelihood of being PRR-deficient when said first probabilistic score for said sample is at or below the threshold may comprise determining that the sample has the type of PRRd associated with the class of the second, third or fourth class that is associated with the highest probabilistic score.

[0016] Determining whether said sample has a high or low likelihood of being PRR-deficient may not use the tumour mutational burden of the sample. Determining whether said sample has a high or low likelihood of being PRR-deficient may use values derived from said values of said mutational signatures by applying a log transformation and / or a normalisation.

[0017] The mutational signature metrics may comprise one or more of: the sum of the number of indels or proportion of indels in the sample associated with each of a plurality of indel signatures associated with MMRd; the sum of the number of indels or proportion of indels in the sample associated with each of a plurality of indel signatures associated with pol-dys; the sum of the number of indels or proportion of indels in the sample associated with each of a plurality of indel signatures associated with combined MMRd / pol-dys; the sum of the number of single base substitutions (SBS) or proportion of SBS in the sample associated with each of a plurality of SBS signatures associated with MMRd; the sum of the number of single base substitutions (SBS) or proportion of SBS in the sample associated with each of a plurality of SBS signatures associated with pol-dys; and the sum of the number of single base substitutions (SBS) or proportion of SBS in the sample associated with each of a plurality of SBS signatures associated with combined MMRd / pol-dys.

[0018] The one or more SBS mutational signatures may be selected from: one or more MMRd-associated SBS signatures, one or more Pol-dys-associated SBS signatures, and one or more combined MMRd / Pol-dys SBS signatures. The one or more MMRd-associated SBS signatures may be selected from: an SBS signature (e.g. SBS6) that has higher weights for A[C>T]G, C[C>T]G and G[C>T]G classes of mutations than for any other classes of mutations, an SBS signature (e.g. SBS15) that has higher weights for the G[C>T]G class of mutations than for any other classes of mutations, an SBS signature (e.g. SBS26) that has higher weights for A[T>C]A, A[T>C]G, C[T>C]G, G[T>C]A, G[T>C]G, G[T>C]T, T[T>C]A and T[T>C]T classes of mutations than for any other classes of mutations, an SBS signature (e.g. SBS44) that has higher weights for C[C>A]T, A[C>T]A, A[C>T]G, G[C>T]A, G[C>T]C, G[C>T]G, G[C>T]T G[T>C]A and T[T>C]T classes of mutations than for any other classes of mutations, and an SBS signature (e.g. SBS97) that has higher weights for G[C>T]G and G[T>C]C classes of mutations than for any other classes of mutations. The one or more Pol-dys-associated SBS signatures may be selected from: an SBS signature (e.g. SBS10a) that has higher weights for T[C>A]T, T[T>G]T and T[C>T]G classes of mutations than for any other classes of mutations, and an SBS signature (e.g. SBSI Od) that has higher weights for T[C>A]A and T[C>A]T classes of mutations than for any other classes of mutations. The one or more combined MMRd / Pol-dys SBS signatures may be selected from: an SBS signature (e.g. SBS14) that has higher weights for C[C>A]A, G[C>A]T, T[C>A]T, A[C>T]G, G[C>T]A, G[C>T]C, G[C>T]G, and G[C>T]T classes of mutations than for any other classes of mutations, and an SBS signature (e.g. SBS20) that has higher weights for the C[C>A]T class of mutations than for any other class of mutations. The one or more MM Rd-associated SBS signatures may be signatures selected from: SBS6, 15, 26, 44, 97 as provided in Table 12, or in Table 12b. The one or more Pol-dys-associated SBS signatures may be signatures SBS10a and / or SBSIOd as provided in Table 12 or Table 12b. The one or more combined MMRd / Pol-dys SBS signatures may be SBS14 and / or SBS20 as provided in Table 12 or Table 12b.

[0019] The one or more indel signatures may be selected from: one or more MMRd-associated indel signatures (such as e.g. signatures lnD7 and / or lnD19 as defined in Table 9), optionally wherein the one or more MMRd-associated indel signatures are signatures that are primarily associated with 1 and / or 2 or more bp indels at repeat tracts or 5 or more repeats and / or wherein the or more MMRd-associated indel signatures comprise a signature that is characterised by frequent 2 or more bp insertions at long repeat tracts (e.g. 5 or more repeats - e.g. In D7), and a signature that is characterised by 1 and 2 or more bp deletions at repeat tracts of 5 or more repeats (e.g. I n D19); one or more Pol-dys-associated indel signatures (such as e.g. signatures Indi 4 and / or lnD15 as defined in Table 9), optionally wherein the one or more Pol-dys-associated indel signatures comprise signatures that are primarily associated with 1 bp T insertions, and / or wherein the one or more Pol-dys associated indel signatures comprise a signature that is associated 1 bp T insertions in repeat tracts of 5 to 7 repeats in a 5’A sequence context (e.g. In D14), and a signature that is associated with 1 bp T insertions in a plurality of sequence contexts including AC for all repeat lengths, and AA, CA, CC, GA, GC for repeats of 5 to 7 repeats (e.g. lnD15); and one or more combined MMRd / Pol-dys indel signatures (such as e.g. signatures selected from: lnD16a, In D16b, lnD20 and In D21 as defined in Table 9), optionally wherein the one or more combined MMRd / Pol- dys indel signatures comprise signatures that are primarily associated with 1 bp T deletions at long repeat tracts (e.g. 5 or more repeats), 1 bp deletions, 1 bp insertions and deletions at long repeat tracts (e.g. 5 or more repeats), and 1 bp T deletions in AA contexts, and / or wherein the one or more combined MMRd / Pol-dys indel signatures comp (i) a signature that is associated with 1 bp T deletions in AA contexts in repeat tracts of up to 7 repeats and 1 bp T deletions in AC contexts in repeat tracts of 5 to 7 repeats (e.g. lnD20), (ii) a signature that is associated with 1 bp T insertions in repeat tracts of 5 to 7 repeats in 5’A and CG and GG sequence contexts, 1 bp T insertions in repeat tracts of 8 or more repeats AC sequence contexts, 1 bp T deletions in repeat tracts of 8 or more repeats in AA and CC sequence contexts (e.g. lnD21), (iii) a signature associated with 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context (e.g. lnD16a), and (iv) a signature associated with 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context and 1 bp T insertions are repeat tracts of 5 to 7 repeats in AA or AC sequence contexts (e.g. lnD16b). The one or more MM Rd-associated indel signatures may be signatures lnD7 and / or lnD19 as defined in Table 9. The one or more Pol-dys-associated indel signatures may be signatures Indi 4 and / or lnD15 as defined in Table 9. The one or more combined MMRd / Pol-dys indel signatures may be selected from: lnD16a, lnD16b, lnD20 and lnD21 as defined in Table 9.

[0020] The one or more MMRd-associated indel signatures may be selected from: an indel signature (e.g. lnD7) that has higher weights for categories comprising 2 or more bp deletions at repeat tracts of 5 or more repeats (e.g. Del(2,):U(1 ,2):R(5,9)), 1 bp T deletions at repeat tracts of 8 or more repeats in any sequence context (e.g. A[Del(T):R(8,9)]A, A[Del(T):R(8,9)]C, A[Del(T):R(8,9)]G, C[Del(T):R(8,9)]A, C[Del(T):R(8,9)]C, C[Del(T):R(8,9)]G, G[Del(T):R(8,9)]A, G[Del(T):R(8,9)]C, G[Del(T):R(8,9)]G), 1 bp T deletions at repeat tracts of 5 or more repeats in an AA sequence context (e.g. A[Del(T):R(5,7)]A), and 1 bp C deletions at repeat tracts of 6 or more bp (e.g. Del(C):R(6,9)), than for any other categories of indels; and an indel signature (e.g. lnD19) that is associated with PMS2 deficiency and / or that has a higher weight for categories comprising 2 or more bp insertions at repeat tracts of 5 or more repeats (e.g. lns(2,):R(5,9)), than for any other categories of indels.

[0021] The one or more pol-dys-associated indel signatures may be selected from: an indel signature (e.g. lnD14) that is associated with POLD1 exonuclease mutations and / or that has higher weights for categories comprising 1 bp T insertions in repeat tracts of 5 to 7 repeats in a 5’A sequence context (e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, A[lns(T):R(5,7)]G), than for any other categories of indels; and an indel signature (e.g. I n D 15) that is associated with POLE exonuclease mutations and / or that has higher weights for categories comprising 1 bp T insertions in a plurality of sequence contexts, than for any other categories of indels. The plurality of sequence contexts may include AC (for all repeat lengths), AA, CA, CC, GA, GC (all particularly for repeats of 5 to 7 repeats), such as e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, C[lns(T):R(5,7)]A, C[lns(T):R(5,7)]C,

[0022] G[lns(T):R(5,7)]A, G[lns(T):R(5,7)]C, and / or A[lns(T):R(8,9)]C. The one or more combined MMRd / Pol-dys indel signatures are selected from: an indel signature (e.g. lnD16a) that is associated with concurrent loss of POLE proofreading and MMR and / or that has higher weights for categories comprising 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context (e.g. A[Del(T):R(8,9)]A, A[Del(T):R(5,7)]A), than for any other categories of indels; an indel signature (e.g. lnD16b) that is associated with concurrent loss of POLE proofreading and MMR and / or that has higher weights for categories comprising 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context (e.g. A[Del(T):R(8,9)]A, A[Del(T):R(5,7)]A) and 1 bp T insertions are repeat tracts of 5 to 7 repeats in AA or AC sequence contexts (e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C), than for any other categories of indels; an indel signature (e.g. lnD20) is associated with combined MMRd and POLE dysregulation and / or that has higher weights for categories comprising 1 bp T deletions in AA contexts in repeat tracts of up to 7 repeats and 1 bp T deletions in AC contexts in repeat tracts of 5 to 7 repeats (e.g. A[Del(T):R(1 ,4)]A, A[Del(T):R(5,7)]A, A[Del(T):R(5,7)]C), than for any other categories of indels; and an indel signature (e.g. lnD21) that is associated with concurrent POLE proofreading defect and MMR, and / or that has higher weights for categories comprising 1 bp T insertions in repeat tracts of 5 to 7 repeats in 5’A and CG and GG sequence contexts, 1 bp T insertions in repeat tracts of 8 or more repeats AC sequence contexts, 1 bp T deletions in repeat tracts of 8 or more repeats in AA and CC sequence contexts (e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, A[lns(T):R(5,7)]G,

[0023] C[lns(T):R(5,7)]G, G[lns(T):R(5,7)]G, A[lns(T):R(8,9)]C, A[Del(T):R(8,9)]A, C[Del(T):R(8,9)]C), than for any other categories of indels.

[0024] The one or more indel signatures may be signatures that comprise a plurality of weights for a respective plurality of predetermined categories of indels, wherein the plurality of predetermined categories comprise a plurality of categories comprising insertions of 1 bp, a plurality of categories comprising insertions of more than 1 bp, a plurality of categories comprising deletions of 1 bp, a plurality of categories comprising deletions of more than 1 bp, wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises categories that differ from each other by the 5’ and / or 3’ sequence context of the insertion / deletion or repeat tract comprising the insertion / deletion, and wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises a plurality of categories comprising insertions / deletions in repeat tracts of 6 or more bp, said plurality of categories differing by the length of the repeat tract.

[0025] The plurality of weights for any signature (indel or SBS) may sum to 1 and may represent probabilities that mutations in each of the respective classes are observed in a sample in which the signature is present. According to a second aspect, there is provided a method of predicting whether a subject with cancer is likely to respond to a specific therapy, the method comprising characterising a sample obtained from a tumour in the subject as having a high or low likelihood of being PRR-deficient using a method of any preceding claim, wherein when the sample is characterised as having a high likelihood of being PRR-deficient, the subject is likely to respond to the specific therapy. The specific therapy is a therapy that is indicated for subjects with PRR-deficiency. The specific therapy may be a therapy to which subjects with MMRd and / or pol-dys are known to respond or be more likely to respond than subjects with PRR proficient cancer. The specific therapy may be an immunotherapy. An immunotherapy may be an immune checkpoint inhibitor. The method may further comprise administering the therapy (e.g. immunotherapy), to a subject that has been diagnosed as likely to respond to the therapy (e.g. immunotherapy). The method may comprise recommending a subject that has been diagnosed as likely to respond to the immunotherapy for treatment with the therapy (e.g. immunotherapy). The method may comprise administering an alternative therapy (e.g. a conventional chemotherapy, radiotherapy, etc.) and / or recommending a subject for treatment with an alternative therapy, where the subject has been diagnosed as not likely to respond to the specific therapy (e.g. immunotherapy).

[0026] According to a third aspect, there is provided a method of selecting a subject having cancer for treatment with a specific therapy (e.g. an immunotherapy), the method comprising characterising a sample obtained from a tumour in the subject as having a high or low likelihood of being PRR - deficient using a method of any embodiment of the first aspect, and selecting the subject for treatment with the specific therapy (e.g. an immunotherapy) if the sample is characterised as having a high likelihood of being PRR-deficient. The therapy may have any of the features described in relation to the preceding aspect.

[0027] According to a fourth aspect, there is provided a method of treating cancer in a subject determined to have a tumourwith a high likelihood of being PRR-deficient, wherein the likelihood of the tumour being PRR-deficient is determined by characterising a DNA sample obtained from the tumour using a method according to any embodiment of the first aspect.

[0028] According to any of the preceding aspects, the immunotherapy may be administered (or recommended for administration) in combination with one or more therapies, such as one or more chemotherapies, one or more courses of radiotherapy and / or one or more surgical interventions. According to any of these aspects, the immunotherapy may be administered (or recommended for administration) in combination with a PARP inhibitor or platinum-based therapy if the subject has been determined as having a high likelihood of being HR-deficient and / or having a high-likelihood of responding to a PARP inhibitor or platinum-based therapy. Thus, any such method may further comprise determining whether the subject is likely to respond to a PARP inhibitor or platinumbased therapy and / or characterising a sample obtained from a tumour in the subject as having a high or low likelihood of being HR-deficient. Methods suitable for this purpose are described in WO 2018 / 115452, WO 2017 / 191074, and WO 2017 / 191073, all of which are incorporated herein by reference.

[0029] According to a further aspect, there is provided an immunotherapy for use in a method of treatment of cancer in a subject, the method comprising: (i) determining whether a DNA sample obtained from said subject has a high or low likelihood of being PRR-deficient using a method according to any embodiment of the first aspect; and (ii) administering the immunotherapy to said subject if the DNA sample is determined to have a high likelihood of being PRR-deficient. An immunotherapy may be a checkpoint inhibitor drug, such as a PD-1 or PD-L1 inhibitor.

[0030] According to a further aspect, there is provided a method of predicting whether a subject with cancer is likely to respond to a non-fluorouracil-based chemotherapy, the method comprising characterising a sample obtained from a tumour in the subject as having a high or low likelihood of being PRR-deficient using a method of any preceding claim, wherein if the sample is characterised as having a high likelihood of being PRR-deficient, the subject is likely to respond to the non-fluorouracil-based chemotherapy. According to a further aspect, there is provided a method of predicting whether a subject with cancer is likely to respond to a fluorouracil-based chemotherapy, the method comprising characterising a sample obtained from a tumour in the subject as having a high or low likelihood of being MMR-deficient using a method of any preceding claim, wherein if the sample is characterised as having a high likelihood of being PRR-deficient, the subject is unlikely to respond to the fluorouracil-based chemotherapy. According to any of these aspects, the fluorouracil-based therapy or non-fluorouracil based therapy may be administered (or recommended for administration) in combination with one or more therapies, such as one or more chemotherapies, one or more courses of radiotherapy and / or one or more surgical interventions. According to any of these aspects, the fluorouracil-based therapy or non-fluorouracil based therapy may be administered (or recommended for administration) in combination with a PARP inhibitor or platinum-based therapy if the subject has been determined as having a high likelihood of being HR-deficient and / or having a high-likelihood of responding to a PARP inhibitor or platinum-based therapy. Thus, any such method may further comprise determining whether the subject is likely to respond to a PARP inhibitor or platinum-based therapy and / or characterising a sample obtained from a tumour in the subject as having a high or low likelihood of being HR-deficient. Methods suitable for this purpose are described in WO 2018 / 115452, WO 2017 / 191074, and WO 2017 / 191073.

[0031] According to a further aspect, there is provided a method of providing a prognosis for a subject who has been diagnosed with cancer, the method comprising characterising a sample obtained from a tumour in the subject as having a high or low likelihood of being MMR-deficient using a method of any preceding claim, wherein if the sample is characterised as having a high likelihood of being PRR-deficient, the subject is likely to have a better prognosis than a subject characterised as having a low likelihood of being PRR-deficient.

[0032] According to a further aspect there is provided a chemotherapy for use in a method of treatment of cancer in a subject, the method comprising: (i) determining whether a DNA sample obtained from said subject has a high or low likelihood of being MMR-deficient using a method according to any embodiment of the first aspect; and (ii) administering the chemotherapy to said subject if the DNA sample is determined to have a high likelihood of being MMR-deficient, preferably wherein the chemotherapy is a non-fluorouracil-based therapy. Alternatively, the method may comprise administering the chemotherapy to said subject if the DNA sample is determined to have a low likelihood of being MMR-deficient, preferably wherein the chemotherapy is a fluorouracil-based therapy.

[0033] According to a further aspect, there is provided a method of providing a tool for characterising a DNA sample obtained from a tumour, the method including the steps of: obtaining mutational signature profiles for a plurality of training samples associated with known PRR-deficiency status; obtaining a training data set comprising, for each of a plurality of training samples associated with known PRR-deficiency status, the value of a plurality of mutational signature metrics comprising: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)-deficiency; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of postreplication repair (PRR)-deficiency; and training a machine learning model to predict, based on said values of said one or more mutational signature metrics, whether each training sample has a high or low likelihood of being PRR-deficient, wherein a sample that has a high likelihood of being PRR-deficient has a high likelihood of having mismatch repair deficiency (MMRd), polymerase dysfunction (pol-dys), or combined mismatch repair deficiency and polymerase dysfunction (MMRd / pol-dys). The method of the present aspect may have any of the features described in relation to the first aspect.

[0034] According to a further aspect, there is provided a system comprising: a processor; and a non- transitory computer readable medium comprising instructions that, when executed by the processor, cause the processor to perform the (computer-implemented) steps of the method of any preceding aspect.

[0035] According to a further aspect, there is provided a non-transitory computer readable medium or media comprising instructions that, when executed by at least one processor, cause the at least one processor to perform the method of any embodiment of any aspect described herein. According to a further aspect, there is provided a computer program comprising code which, when the code is executed on a computer, causes the computer to perform the method of any embodiment of any aspect described herein.

[0036] BRIEF DESCRIPTION OF THE FIGURES

[0037] Figure 1 is a flow diagram showing, in schematic form, a method of determining the value of one or more mutational signature metrics.

[0038] Figure 2A is a flow diagram showing, in schematic form, a method of characterising a DNA sample.

[0039] Figure 2B is a flow diagram showing, in schematic form, a method of providing a prognosis, identifying a therapy or treating a subject.

[0040] Figure 3 shows an embodiment of a system for characterising a DNA sample.

[0041] Figure 4 is a diagram illustrating an indel classification scheme as described herein, for A. insertions, and B. deletions and complex indels. Insertions are classified between a plurality of classes for 1 bp insertions and a plurality of classes for insertions of 2 or more bp. 1 bp insertions are separated between C / G insertions and A / T insertions, and then depending on the sequence context (3’ and 5’ nucleotides - not illustrated) and / or the length of the repeat sequence prior to the insertion (homopolymer length, HPL). >2bp insertions are separated between classes depending on the length of the inserted motif (insertion length, IL) and the number of repeats of the motif prior to the insertion (number of repeat units, NRU). Complex indels are indels resulting from both an insertion and a deletion. These are all included in a single class. Deletions are classified between a plurality of classes for insertions of 1 bp and a plurality of classes for insertions of 2 or more bp. 1 bp deletions are separated between C / G deletions and A / T deletions, and then depending on the sequence context (3’ and optionally 5’ nucleotides) and / or the length of the repeat sequence prior to the deletion (homopolymer length, HPL). >2bp deletions are separated between classes where the deletion is in a repeat region (Repeat, the deletion comprises one or more perfect repeats of a sequence - also referred to herein as ‘motif’- immediately adjacent to one or more perfect repeats of the sequence), orwith microhomology at the deletion boundary(ies) (MH, the deletion comprises one or more perfect repeats of a motif and one or more non repeat bases, the deletion between adjacent to one or more perfect repeats of the motif). Repeat deletions are separated between classes based on the length of the deleted motif (deletion length, DL), and the number of repeat units prior to deletion (NRU). Deletions with microhomology are separated between classes based on the length of the deleted sequence (deletion length, DL), and the length of the microhomology (MHL, length of the sequence that is repeated). Figure 5 is a diagram illustrating a process of reducing an exhaustive indel classification into a consolidated indel classification. An exhaustive classification (also referred to herein as “complete classification”) obtained as described in relation to Figure 4 is obtained, comprising 476 channels in the illustrated embodiment. The number of indels in each of these classes is added up across a plurality of samples in each of a plurality of cohorts, each cohort comprising samples from a single cancer type. A. Cumulative number of indels in each class of an exhaustive indel classification for: a bladder cancer cohort (N=347 patients, total number of indels: 580,187), a central nervous system cancer cohort (N=392 patients, total indels=512,385), a colorectal cancer cohort (N=2,146 patients, total indels=60, 166,635), and a lung cancer cohort (N=958, total indel=2,200,268). B. Cumulative number of indels in each class of an exhaustive indel classification for: a skin cancer cohort (N=56, total indels=213,391 ), a stomach cancer cohort (N=181 , total indels=1 ,935,212), a uterus cancer cohort (N=695, total indel=14,090,555) and the combined numbers across all cohorts in A and B (N=4,775, total indels=79,698,633). These demonstrate that some classes (also referred to herein as “channels”) comprise many indels whereas many classes comprise few or no indels. The latter can be combined into summary classes to obtain a consolidated classification (C) comprising 89 channels in the illustrated embodiment. The indel types on the x axis on A and B are as described in Table 6. The consolidated classes of indel types on the x axis in C are as described in Table 2 and Table 6. B. Final, consolidated 89-channel indel profile of a GEL sample.

[0042] Figure 6 shows results of experiments investigating an existing indel classification using isogenic cell lines with known mutational processes. A. Experimental workflow to obtain mutational profiles associated with experimentally validated mismatch repair deficiencies and polymerase mutants in TP53-null hTERT-immortalized retinal epithelial cell (hTERT-RPE1TP53-null, herewith referred to as the control background) B. De novo indel mutation count and average indel fold-increase of CRISPR-edited subclones over controls (n = 2-5 subclones per gene edit). Red dashed line represents the mean indel burden of control subclones. The y axis shows the indel count in log scale. C. UMAP analysis of the mutational profiles of the cell lines in B, with mutations classified according to the COSMIC indel classification (right, 83 channels, described in Table 1). UMAP, uniform manifold approximation and projection for dimension reduction. D. Cosine similarity between 83-channel mutational profiles of a plurality of post-replication repair mutant cell lines. E. Substitution signatures associated with PRRd following background subtraction. F. 83-channel (COSMIC) indel mutational signatures associated with PRRd following background subtraction. G. Key features of COSMIC ID1 , ID2, and ID7. ID1 and ID2 are associated with replication slippage and ID7 is believed to be associated with MMRd. H. Decomposed solution of experimentally generated PPRd indel signatures in Fig. 6G into COSMIC ID83 COSMIC signatures ID1 , ID2, 1 D 14 , ID16 and ID18 are believed to be associated with slippage during DNA replication of the replicated strand (ID1 , ID2), and colibactin exposure (ID18). The etiology of ID 14 and AD16 is unknown. I. Analysis of replication strand asymmetry demonstrating mutational bias of POLD1 and POLE mutants respectively. POLE is believed to play a greater role in correcting errors on the leading replication strand while POLE is believed to play a greater role in proofreading the lagging strand. J. Cosine similarity between mutational profiles (y axis, indels classified according to the COSMIC indel classification) of PPRd cell lines and all COSMIC indel signatures (x axis, proposed etiology at the top). HRD=homologous recombination deficiency. MMRd=mismatch repair deficiency. NHEJ=non-homologous end joining. UV=UV light exposure. TOP-2 mut=mutations in topoisomerase TOP-2. Comparing figures F and G shows that the experimentally derived PPRd signatures do not resemble the COSMIC ID7 MMRd signature, indicating that ID7 is likely to represent an extraction artefact. The mutational profiles in F suggest the existence of previously unknown indel signatures associated with polymerase mutations. More information about these results is provided in Example 1 below. K. Distinguishing 83-channel indel profiles of edited subclones from background control. Light blue lines depict the distribution of cosine similarities between bootstrapped indel profiles of controls and the aggregated control indel profile (see Fig. 6N). The x axis shows the indel count in log scale. L. Cosine similarities between individual subclones and averaged indel profile of background controls. M. Relative proportion of seven indel mutation types of each gene edit. Ins, insertion; Del. deletion; R, repeat; M, microhomology; C, cytosine; T, thymine. N. Aggregated indel profile of hTERT-RPE1TP5- null background control. O. Unsupervised clustering of experimentally generated 83-channel PPRd indel signatures using (1- cosine similarity) as distance metric. Dashed vertical line marks 0.1 cut-off.

[0043] Figure 7 shows an analysis of the cell lines of Figure 6 using an indel classification as described herein (Figure 4). A. Example mutational profile for a MMRd mutant cell line sample (KO of MSH2, sample comprising n=3807 indels). B. Distinguishing 89-channel indel profiles of edited subclones from background control (see Fig. 7K). Light blue lines depict the distribution of cosine similarities between bootstrapped indel profiles of controls and the aggregated control indel profile. The x axis shows the indel count in log scale. Bootstrap data is used as a simulation of background mutagenesis (cells gown in culture acquire mutations which we call “background mutagenesis”). Background mutagenesis is simulated across a range of mutation burdens from very few to > 10,000 mutations to obtain an “expected” distribution, if there were no signature due to the KO / KI mutant. Anything that falls far away from this expected distribution is considered as likely to have a signature. C. tSNE clustering of the mutational profiles of multiple mutant cell lines using the new indel classification to demonstrate that there are detectable differences in mutational profiles. D. Mutational profiles (indels classified as described herein) associated with each PRRd mutation (summarized across replicates and with background removed, see methods Example 2). X axis are categories of indels as described in Table 2, in the order indicated in Table 2. E. Proportions of mutations (combined SNVs and indels) explained by SBS (COSMIC classification, trinucleotide context, de novo extraction), and indel signatures (classification as described herein) for each mutant. F. Proportions of deletions and insertions in each mutant. There are very few complex indels in these particular mutants. G. Cosine similarity between the mutational profiles of a plurality of PPRd mutants, based on mutational profiles obtained using an indel classification as described herein. Comparing Fig. 7G with Fig. 6D shows that the indel classification as described herein has improved ability to distinguish between MMRd, MMRd+POLE and POLE mutants. More information about these results is provided in Example 2 below. H. Cosine similarities of edited subclones and bootstrapped controls in 83-channel profiles against 89-channel profiles. Two-tailed Wilcoxon signed-rank test, p=1.917e-07). I. UMAP clustering of de novo 89-channel indel profiles 946 of all subclones (n = 37). UMAP, uniform manifold approximation and projection for dimension reduction. J. Gini index of experimentally generated indel signatures in 83-channel format vs. 89- channel format. Two-tailed Wilcoxon signed-rank test, P=0.001953. K. Aggregated 89-channel indel profile of hTERT-RPE1TP5- null control subclones. L. Unsupervised clustering of experimentally generated 89-channel PPRd indel signatures using (1 -cosine similarity) as distance metric. Dashed vertical line marks 0.1 cut-off.

[0044] Figure 8 shows the results of analysis of a pan-cancer cohort of samples using an indel classification as described in relation to Fig. 4. More information about these results is provided in Example 3 below. A. Left panel shows the spread of mutation count for each tumour type I Middle panel shows the relative proportions of the different broad classes of indels (even at this scale differences between tumour types are apparent I Right shows for each cohort of sample (cohorts comprising samples from a single cancer type as indicated), the proportion of tumours that show evidence of the presence of the indicated indel signatures (size of the dot), and the median (across the cohort) contribution of the respective signature to the number of mutations per Mb of genome in the samples if the cohort. B-D. Graphical illustrations of indel signatures as described herein, with associated etiology. Numerical representations of these signatures are provided in Table 9. X axis are categories of indels as described in Table 2, in the order indicated in Table 2.

[0045] Figure 9 shows the results of substitution mutational signatures associated with post- replicative repair deficiency. A. De novo substitution mutation count, and average substitution fold-increase of CRISPR-edited subclones over controls (n = 2-5 subclones per gene edit). Red dashed line represents the mean substitution burden of controls. B. Distinguishing substitution profiles of edited subclones from background controls. Light green lines depict the distribution of cosine similarities between bootstrapped substitution profiles of controls and the aggregated control substitution profile. The x axis shows the substitution count in log scale. C. UMAP clustering of de novo substitution profiles of all subclones (n = 37). UMAP, uniform manifold approximation and projection for dimension reduction. D. Relative proportion of six substitution mutation types of each edit genotype. E. Cosine similarities among experimentally generated PPRd substitution signatures in Fig. 6E. F. Single nucleotide variant (SNV) to indel ratio of edited subclones. Bar represents the averaged ratio for each genotype. G. Cosine similarities between experimentally generated PPRd substitution signatures and cancer derived SBS reference signatures (RefSig).

[0046] Figure 10 shows the results of de novo signature extraction of 83-channel vs. 89-channel indel catalogues demonstrates differential signature recovery. A. Signature selection plot of de novo extraction by SigProfilerExtractor of n = 37 experimental samples in 83- or 89- channel indel catalogues. Suggested solutions are shaded in orange or purple for 83-channel and 89-channel catalogues, respectively. B. Suggested de novo solution (two signatures) by SigProfilerExtractor for 83-channel experimental cohort (n = 37). C. Suggested de novo solution (four signatures) by SigProfilerExtractor for 89-channel experimental cohort (n = 37). D. Signature selection plots of de novo extraction by signature. tools. lib, SigProfilerExractor, and MuSiCal of n = 52 ICGC colorectal cancers in 83-channel vs. 89-channel indel catalogues. Suggested solutions are shaded in orange or purple for 83-channel and 89-channel indel catalogues, respectively. E. Selected de novo solution (five signatures) from signature. tools. lib for 83-channel ICGC colorectal cancers. F. Selected de novo solution (eight signatures) from signature. tools. lib for 89-channel ICGC colorectal cancers. G. Suggested de novo solution (five signatures) by SigProfilerExtractor for 83- channel ICGC colorectal cancers. H. Suggested de novo solution (six signatures) by SigProfilerExtractor for 89-channel ICGC colorectal cancers. Signatures from MuSiCal extraction are not shown.

[0047] Figure 11 shows the results of hierarchical clustering and dendrogram partitioning of n = 111 tissue-specific signatures into n = 40 distinct patterns. Hierarchical clustering and dendrogram partitioning of n = 111 tissue-specific signatures into n = 40 distinct patterns using (1 -cosine similarity) as distance metric, with a cut-off of 0.15. The 40 distinct patterns were manually reviewed, revised, and inspected for mixed patterns to produce the final 33 consensus indel signatures (InDs) (see Fig.8).

[0048] Figure 12 shows indel signatures of mutagen exposures and DNA repair / replication gene knockouts in human induced pluripotent stem cells, as well as proposed etiology for lnD9a. A. Indel signatures of exposures to environmental mutagens in human induced pluripotent stem cells (hiPSC). Data reanalyzed from previous study. B. Indel signatures of DNA repair / replication gene knockouts in human induced pluripotent stem cells. Data reanalyzed from previous study. C. Extended sequence context of 1 bp C deletions of lnD9a showed TTCT / TTCA enrichment for APOBEC deamination. APOBEC, apolipoprotein B mRNA-editing enzyme, catalytic polypeptide. D. Proposed etiology for APOBEC-associated lnD9a. APOBEC, apolipoprotein B mRNA-editing enzyme, catalytic polypeptide; TLS, translesion synthesis; UNG, Uracil-N-glycosylase.

[0049] Figure 13 shows a heatmap clustering of n = 26 experimental indel signatures and n= 37 cancer- derived InDs. Heatmap clustering of n = 26 experimentally generated signatures and 37 cancer- derived InDs. Experimentally generated signatures include n = 10 from the current study, n = 10 from the mutagen study38 (Fig. 12), and n = 6 from iPSC knockout study (Fig. 12). Putative sources (left) and etiologies (right) of the signatures are annotated if known (Table 10). HRd, homologous recombination deficiency; PAH, polycyclic aromatic hydrocarbons; N-Slip, nascent strand slippage; T-Slip, template strand slippage; MMRd, mismatch repair deficiency. The heatmap was split between panels A and B for readability.

[0050] Figure 14 shows results investigating a clinical classifier - PRRDetect - that improves detection of tumours with post-replicative repair (PRR) dysfunction. A. Simplified workflow of the development of PRRDetect classifier. (I) Initial training using 571 ground truth samples. (II) Final retraining to produce the final PRRDetect classifier. MMRd, mismatch repair deficiency; Pol-dys, polymerase dysfunction; -ve Ctrl, negative control. B. Distribution of coefficients across eight genomic features (including the intercept) contributing to the final PRRDetect classifier. Ranges of values from ten replicates of training in cross validation are shown. Red dots indicate the final coefficients chosen for each class prediction. D. Receiver operating characteristic (ROC) curves demonstrating the superior performances of PRRDetect on independent cancer cohort (n =1 ,351) against alternative biomarker strategies. P values were calculated using two-sided non-parametric test based on the bootstrap distribution (10,000) of the difference in AUCs. C. Validation and assessment of PRRDetect’s performance on independent cancer cohorts. E. PRRDetect results of n =1 ,335 ICGC and Hartwig cancers, ordered from the lowest to the highest prediction probability across the x axis (left to right) for polymerase-dysfunctional samples (orange), combined mismatch-repair deficient (MMRd) samples (blue), and MMRd samples (purple). Negative samples were ordered by TMB in increasing order from left to right. Results of MSIseq, MMRDetect, cancer gene driver annotation, and cancer tissue origin are labelled at the bottom tracks. Dashed rectangle highlights the extent of false positive overcalling if using TMB > 10 muts / Mb as a cutoff. F. Concordance of calls among TMB-H (> 10 mutations / Mb), positive exposure to SBS signatures that impart hypermutation (SBS2, 4, 7a, 7c, 13), and PRRDetect prediction across n = 1 ,335 combined ICGC and Hartwig cancers. Counts indicate the total number of samples matching the indicated calls. G. Concordance of calls among TMB-H (> 10 mutations / Mb), positive exposure to SBS signatures that impart hypermutation, and PRRDetect prediction across n = 4,775 GEL tumours.

[0051] Figure 15 shows prediction results of PRRdetect (see Fig. 14) on n= 1 ,335 cancers from ICGC / Hartwig selected cohorts and n=4,775 GEL cancers, stratified by tissue types. Fig. 15A shows PRRDetect prediction of ICGC and Hartwig three cancer types (n=1 ,335). Fig. 15B shows PRRDetect prediction of GEL seven cancer types (n=4,775). DETAILED DESCRIPTION

[0052] In describing the present invention, the following terms will be employed, and are intended to be defined as indicated below.

[0053] “and / or” where used herein is to be taken as specific disclosure of each of the two specified features or components with or without the other. For example “A and / or B” is to be taken as specific disclosure of each of (i) A, (ii) B and (iii) A and B, just as if each is set out individually herein.

[0054] A “sample” as used herein may be a cell or tissue sample (e.g. a biopsy), a biological fluid, an extract (e.g. a protein or DNA extract obtained from the subject), from which genomic material can be obtained for genomic analysis, such as genomic sequencing (whole genome sequencing, whole exome sequencing, targeted (also referred to as “panel”) sequencing). In particular, the sample may be a blood sample, or a tumour sample. The sample may be one which has been freshly obtained from a subject or may be one which has been processed and / or stored prior to making a determination (e.g. frozen, fixed or subjected to one or more purification, enrichment or extractions steps). In particular, the sample may be a cell or tissue culture sample. As such, a sample as described herein may refer to any type of sample comprising cells or genomic material derived therefrom, whether from a biological sample obtained from a subject, or from a sample obtained from e.g. a cell line. The sample is preferably from a mammalian (such as e.g. a mammalian cell sample or a sample from a mammalian subject, including in particular a model animal such as mouse, rat, etc.), preferably from a human (such as e.g. a human cell sample or a sample from a human subject). Further, the sample may be transported ad / or stored, and collection may take place at a location remote from the genomic sequence data acquisition (e.g. sequencing) location, and / or the computer-implemented method steps may take place at a location remote from the sample collection location and / or remote from the genomic data acquisition (e.g. sequencing) location (e.g. the computer-implemented method steps may be performed by means of a networked computer, such as by means of a “cloud” provider).

[0055] A “tumour sample” refers to a sample that contains tumour cells or genetic material derived therefrom. The tumour sample may be a cell or tissue sample (e.g. a biopsy) obtained directly from a tumour. A tumour sample may be a sample that comprises tumour cell or genetic material derived therefrom, that has not been obtained directly from a tumour. For example, a tumour sample may be a sample comprising circulating tumour cells or circulating tumour DNA. Thus, a tumour sample may also be a biological fluid (e.g. a liquid biopsy such as a blood, urine, or cerebrospinal fluid biopsy). A sample comprising a mixture of tumour cells and other cells (or material genetic derived therefrom) may be subject to one or more processing steps, whether prior to or subsequent to the acquisition of sequence data, in orderto identify sequence data that is representative of the genetic material from the tumour. For example, a sample comprising cells may be subject to one or more cell purification steps which selectively enrich the sample for tumour cells. Similarly, a sample comprising modified and non-modified cells can be subject to one or more purification or selection steps to enrich the sample for modified cells. Protocols for doing this are known in the art. As another example, a sample of genetic material may be subject to one or more capture and / or size selection steps to selectively enrich the sample for tumour-derived genetic material. Protocols for doing this are known in the art. As another example, sequence data may be subject to one or more filtering steps (e.g. based on fragment length) to enrich the data for information that relates to tumour-derived genetic material. Protocols for doing this are known in the art.

[0056] A “normal sample” (also referred to as “germline sample” or “parent sample”) refers to a sample that contains non-tumour or non-modified cells or genetic material derived therefrom. A normal sample may be matched to a particular tumour or modified sample in the sense that it is obtained from the same biological source (subject or cell line) as the tumour or modified sample. A normal sample may be a cell or tissue sample obtained from a subject, or a sample of biological fluid. A sample comprising a mixture of normal cells and other cells (or material genetic derived therefrom) may be subject to one or more processing steps, whether prior to or subsequent to the acquisition of sequence data, in order to identify sequence data that is representative of the genetic material from the normal cells (as already described above). For example, a sample comprising modified and non-modified cells can be subject to one or more purification or selection steps to enrich the sample for non-modified cells. Similarly, a sample comprising normal and tumour-derived cells can be subject to one or more purification steps which selectively enrich the sample for normal cells.

[0057] The term “sequence data” refers to information that is indicative of the presence and / or amount of genomic material in a sample that has a particular sequence. Such information may be obtained using sequencing technologies, such as e.g. next generation sequencing (NGS, such as e.g. whole exome sequencing (WES), whole genome sequencing (WGS), or sequencing of captured genomic loci (targeted or panel sequencing)), or using array technologies, such as e.g. SNP arrays, or other molecular counting assays. When NGS technologies are used, the sequence data may comprise a count of the number of sequencing reads that have a particular sequence. When non-digital technologies are used such as array technology, the sequence data may comprise a signal (e.g. an intensity value) that is indicative of the number of sequences in the sample that have a particular sequence, for example by comparison to an appropriate control. Sequence data may be mapped to a reference sequence, for example a reference genome, using methods known in the art (such as e.g. Bowtie (Langmead et al., 2009)). Thus, counts of sequencing reads or equivalent non- digital signals may be associated with a particular genomic location. Further, a genomic location may contain a mutation, in which case counts of sequencing reads or equivalent non-digital signals may be associated with each of the possible variants (also referred to as “alleles”) at the particular genomic location. The process of identifying the presence of a mutation at a particular location in a sample is referred to as “variant calling”, and can be performed using methods known in the art (such as e.g. the GATK HaplotypeCaller, gatk.broadinstitute.org / hc / en-us / articles / 360037225632- HaplotypeCaller). For example, sequence data may comprise a count of the number of reads (or an equivalent non-digital signal) which match a germline (also sometimes referred to as “reference”) allele at a particular genomic location, and a count of the number of reads (or an equivalent non-digital signal) which match a mutated (also sometimes referred to as “alternate”) allele at the genomic location.

[0058] The term “mutation” refers to a difference in a nucleotide sequence (e.g. DNA or RNA) in a sample compared to a reference. For example, a mutation may be a single nucleotide variant (SNV), multiple nucleotide variants, a deletion mutation, an insertion mutation, a translocation, a missense mutation, a translocation, a fusion, etc. Mutations may be identified using sequence data. An "indel mutation" (or simply “indel”) refers to an insertion and / or deletion of bases in a nucleotide sequence (e.g. DNA or RNA) of an organism.

[0059] Within the context of the present invention, a mutation is typically a somatic mutation, unless the context indicates otherwise. A “somatic mutation” is a mutation that is present in a tumour or modified cell (or genetic material derived therefrom), but not in a corresponding (matched) normal or non-modified cell.

[0060] A composition as described herein may be a pharmaceutical composition which additionally comprises a pharmaceutically acceptable carrier, diluent or excipient. The pharmaceutical composition may optionally comprise one or more further pharmaceutically active polypeptides and / or compounds. Such a formulation may, for example, be in a form suitable for intravenous infusion.

[0061] As used herein "treatment" refers to reducing, alleviating or eliminating one or more symptoms of the disease which is being treated, relative to the symptoms prior to treatment.

[0062] The systems and methods described herein may be implemented in a computer system, in addition to the structural components and user interactions described. As used herein, the term “computer system” includes the hardware, software and data storage devices for embodying a system or carrying out a method according to the above-described embodiments. For example, a computer system may comprise a processing unit such as a central processing unit (CPU) and / or graphical processing unit (GPU), input means, output means and data storage, which may be embodied as one or more connected computing devices. Preferably the computer system has a display or comprises a computing device that has a display to provide a visual output display. The data storage may comprise RAM, disk drives or other computer readable media. The computer system may include a plurality of computing devices connected by a network and able to communicate with each other over that network. It is explicitly envisaged that computer system may consist of or comprise a cloud computer.

[0063] The methods described herein may be provided as computer programs or as computer program products or computer readable media carrying a computer program which is arranged, when run on a computer, to perform the method(s) described herein. As used herein, the term “computer readable media” includes, without limitation, any non-transitory medium or media which can be read and accessed directly by a computer or computer system. The media can include, but are not limited to, magnetic storage media such as floppy discs, hard disc storage media and magnetic tape; optical storage media such as optical discs or CD-ROMs; electrical storage media such as memory, including RAM, ROM and flash memory; and hybrids and combinations of the above such as magnetic / optical storage media.

[0064] The methods described herein may be computer implemented unless indicated otherwise. Indeed, the complexity associated with obtaining indel mutation profiles from samples, such as e.g. by analysing sequence data (e.g. thousands to hundreds of thousands of reads per sample) or somatic mutations lists derived therefrom (comprising multiple hundreds or thousands of mutations) and classifying each of these in one or 89 classes by analysis of the sequence of the mutation and its context, as well as calculating indel signature exposures using such mutational profiles, is far beyond the capability of the human mind.

[0065] Post-replicative repair dysfunction

[0066] The present disclosure relates to the identification of Post-Replicative Repair dysfunction (PRRd). PRRd encompasses defects in MMR (mismatch repair) and replicative polymerase proofreading (referred to herein as “polymerase dysfunction” or “pol-dys”). The mismatch repair pathway is highly conserved and plays a key role in maintaining genomic stability. In eukaryotes, the pathway is mediated by key proteins collectively referred to as “Mut homologue” proteins. These include MSH2 and MSH6 (together forming the heterodimer MutSa), MSH2 and MSH3 (together forming the heterodimer MSHP), MLH1 and PMS2 (together forming the heterodimer MutLa), MLH1 and PMS1 (together forming the heterodimer MutLP), and MLH1 and MLH3 (together forming the heterodimer MutLy). Mutations in the Mut homologue proteins affect genomic stability, and are known to be associated with genetic conditions such as Lynch syndrome (also known as Hereditary nonpolyposis colorectal cancer (HNPCC)), an autosomal dominant genetic condition that is associated with a high risk of colon cancer as well as endometrial, ovary, stomach, small intestine, hepatobiliary tract, upper urinary tract, brain, and skin cancer. MMR deficiency can result in microsatellite instability (MSI), a condition that manifests in the creation of novel microsatellite fragments (repeated sequences of DNA, with repeats often a few base pairs long). MSI has been associated with many cancers, and is most prevalent in association with colon cancer. Studies have found that patients stratified on the basis of whether they were MSI-High (MSI-H), MSI-low (MSI-L) or microsatellite stable (MSS) had different prognosis, with the MSI-H status associated with better survival. This relationship with cancer prognosis has led to the development of multiple commercial diagnostic assays for the detection of microsatellite instability. However, MSI is only one possible manifestation of impaired DNA mismatch repair. Therefore, testing for MSI is not equivalent to testing for MMR deficiency, which is the true biological difference underlying differences in prognosis and response to therapy. Algorithms to classify MMR-deficiency tumours have been developed using massively-parallel sequencing data (Huang et al., 2013; Zou et al. 2021). Most existing classifiers depend on detecting elevated tumour mutational burdens (TMB) or microsatellite instability (MSI). Thus, they also rely on relatively crude metrics of genomic instability that common manifestations of MMR deficiency. A notable exception to this is MMRDetect (Zou et al. 2021), described in PCT / EP2022 / 057387 (published as W02022 / 200293, which is incorporated herein by reference). This classifier predicts the MMRd status of a sample using one or more mutational signature metrics selected from: exposure of one or more mutational signatures of mismatch repair (in particular, RefSig MMR1 and RefSig MMR2), similarity between the substitution profile of the sample (i.e. the number or proportions of SBS in each of 96 categories as described further below) and that of one or more MMR gene knockouts, the number of repeat mediated indels in the mutational profile of the sample, and the similarity between the repeat mediated deletion profile of the sample and that of one or more MMR gene knockouts. However, this approach does not have the ability to identify PRRd more broadly, or to distinguish between different types of PRRd, such as PRRd due to pol-dys, MMRd or combined MMRd / pol-dys. Further, the classifier does not make use of indel mutational signatures. Replicative polymerase proofreading is another essential post-replicative repair mechanism. It relies on the ability of most polymerases to remove incorrectly incorporated nucleotides thanks to the polymerase’s 3’-5’ exonuclease activity. There is no existing tool to identify polymerase dysfunction other than based on mutations in polymerase genes (driver mutations). PRR deficiency is known to be associated with sensitivity to immune checkpoint inhibition (see e.g. Cristescu et al. 2018; Wang et al. 2019; Das et al. 2022). However, there is currently no method to identify both MMRd and pol-dys (let alone distinguish between the two) that is not limited to driver mutation analysis. Driver mutation analysis has been shown to repeatedly fail to identify clinically actionable cases, because the dysfunction in the pathways can be caused by many more mechanisms than the few driver mutations that we are aware of. Additionally, tools that are based purely on the manifestation of PRRd such as MSI also fail to accurately identify patients that have PRRd.

[0067] Mutational Signatures

[0068] Throughout this document, the term “signature” (or “mutational signature”, which encompasses SBS and indel signatures) refers to a set of weights that characterise the imprint of a mutational process on the mutational profile of cancer genomes in terms of the classes and proportions of mutations that are generated by the mutational process. In other words, a mutational signature (SBS or indel signature) comprises, for each of a plurality of categories of indels or SBS, a proportion. The plurality of categories of indels / SBS is the same as the plurality of categories (also referred to herein as “classes”) of indels / SBS present in a indel / SNS profile associated with a sample or subject. An indel / SBS profile may be obtained by classifying a plurality of indels / SBS identified in a genome (e.g. indels / SBS identified in a sample or subject) between a predetermined plurality of classes (i.e. each indel / SBS of the plurality of indels being assigned a single class of the plurality of classes).

[0069] A mutational signature catalogue can be extracted from a plurality of mutational profiles each associated with respective samples. A mutational signature catalogue can comprise mutational signatures extracted separately from a plurality of respective cohorts of mutational profiles. Mutational signatures can be extracted from cohorts of samples, by identifying characteristic combination of mutation types that best explain the mutational profiles of the samples in the cohort. This process also results in the quantification of “exposures” to each of the signatures, which quantify the extent of the effect of the respective signatures on the respective mutational profiles. Signatures are defined for a specific type of mutation, e.g. SBS or indels. This is because the classification of mutations is inherently linked to the type of mutation considered. For example, SBS mutations are typically classified based on the mutation content (i.e. which base is mutated to what base) and trinucleotide content (i.e. the identity of the 5’ and 3’ nucleotides surrounding the mutated nucleotide). A signature using such a classification would consist of a set of weights, one for each of the classes defined using this content + trinucleotide context approach (typically 96 classes, one for each unique combination of content and trinucleotide context). By contrast, indels can differ from each other in length, sequence context (where the context that is informative may not be limited to a single base), sequence content, repeatedness and the presence or absence of microhomology. Each of these can be taken into account to generate a plurality of classes of indels, and a signature using such a classification would consist of a set of weights, one for each of the classes defined (where the total number of classes depends on how classes based on unique combinations of these features are combined).

[0070] A mutational catalogue summarising somatic substitutions associated with a sample or a group of samples may be referred to as a “mutational profile”. The present disclosure relates in particular to indel mutations and single base substitutions mutations. As such, a mutational profile may also be referred to as an indel profile or SBS profile in the present disclosure. An indel mutational profile comprises the number of mutations present in a sample within each of a plurality of indel categories. An SBS mutational profile comprises the number of mutations present in a sample within each of a plurality of SBS categories. An indel / SBS mutational profile can be seen as a summary of a list of indel / SBS mutations present in a sample, categorised according to a predetermined set of indel / SBS categories (also referred to as “channels”). A set of mutational signatures may be referred to as a “mutational signatures catalogue”. A set of mutational signatures only contains mutational signatures of the same type that are classified using the same classification scheme, i.e. SBS signatures or indel signatures. Thus, an indel mutational signature catalogue is a set of indel mutational signatures, and a SBS mutational signature catalogue is a set of SBS mutational signatures.

[0071] The present disclosure makes use of mutational signatures (single base substation (SBS) signatures and / or indel signatures, together referred to as “mutational signatures” or “signatures”) that are associated with PRRd. MM Rd-associated signatures (whether indel or SBS signatures) are signatures that are present in samples that have mismatch repair deficiency (MMRd) in the absence of polymerase dysfunction. Pol-dys-associated signatures (whether indel or SBS signatures) are signatures that are present in samples that have polymerase dysfunction in the absence of MMRd. Combined MMRd / Pol-dys signatures (whether indel or SBS signatures) are signatures that are present in samples that have both MMR and polymerase dysfunction. The reference to “samples” in this context encompasses subjects (e.g. subjects that have been diagnosed as having or being likely to have cancer, and from whom a sample can be analysed as described herein), cancers (e.g. samples comprising cancer cells or genomic material derived therefrom, which can be analysed as described herein), and genomes (e.g. a cancer genome determined by analysing one or more samples, for example from a subject that has been diagnosed as having or being likely to have cancer). Polymerase dysfunction can occur due to mutations in the POLE or POLD genes (e.g. POLD1). Any mutation that alters the exonuclease function of these polymerases may lead to PRRd by polymerase dysfunction. Thus, pol-dys and combined MMRd / Pol-dys signatures may be signatures that are enriched in genomes that have mutations in a polymerase gene, such as POLE or POLD. MMRd can occur due to mutations in MMR-associated genes or promoter hypermethylation of MMR-associated genes. MMR associated genes include MLH1 , MLH3, MSH2, MSH3, MSH6, PMS2, PMS1 , and SETD2. Any mutation that alters the function of one of these genes may lead to MMRd. Thus, MMRd and combined MMRd / Pol-dys signatures may be signatures that are enriched in genomes that have mutations in one or more MMR-associated genes.

[0072] Combined MMRd / Pol-dys signatures may be signatures that are enriched in genomes that have mutations in a polymerase gene and in one or more MMR-associated genes. In other words, an MMRd-associated signature may be a signature that is such that in a cohort of samples, the subset of samples that are determined to have the MMRd-associated signature present are enriched for samples that are known to have MMRd. These may be for example samples with mutations in the one or more MMR-associated genes. A pol-dys-associated signatures may be a signature that is such that in a cohort of samples, the subset of samples that are determined to have the Pol-dys- associated signature present are enriched for samples that are known to have pol-dys. These may be for example samples with mutations in the one or more polymerase genes. A combined MMRd / Pol-dys-associated signature may be a signature that is such that in a cohort of samples, the subset of samples that are determined to have the combined MMRd / Pol-dys-associated signature present are enriched for samples that are known to have pol-dys and MMRd. These may be for example samples with mutations in the one or more polymerase genes and in one or more MMR-associated genes. An MMRd-associated signature may also be a signature that has been obtained based on the mutation profile (SBS or indel profile, classified as described herein) in one or more samples that have been genetically modified to include a mutation in a MMR-associated gene. This may be a background corrected mutational profile, where background correction can be performed as described in Zou et al. 2021. A pol-dys-associated signature may also be a signature that has been obtained based on the mutation profile (SBS or indel profile, classified as described herein) in one or more samples that have been genetically modified to include a mutation in a polymerase gene. This may be a background corrected mutational profile, where background correction can be performed as described in Zou et al. 2021. A combined MMRd / pol-dys- associated signature may also be a signature that has been obtained based on the mutation profile (SBS or indel profile, classified as described herein) in one or more samples that have been genetically modified to include a mutation in a MMR-associated gene and a mutation in a polymerase gene. This may be a background corrected mutational profile, where background correction can be performed as described in Zou et al. 2021 .

[0073] The methods described herein relate at least in part to determining an indication of whether an indel mutational signature in a catalogue is present in the sample, and / or an indication of whether a SBS mutational signature in a catalogue of SBS mutational signatures is present in the sample. The presence of an indel / SBS signature in the sample can be identified by comparing the proportions in the indel / SBS signature for the same classes, to the numbers or proportions of indels / SBS in the indel / SBS profile of the sample. Proportions may be numbers that sum to 1 over the plurality of categories. Thus, the proportions may indicate the pattern of indel / SBS types that typically arise in a genome when the signature is present. Any individual indel / SBS profile (i.e. any sample or subject) may indicate the presence of one or more indel / SBS signatures. In other words, an indel / SBS profile may represent a combination of indel / SBS signatures. Importantly, the categories in an indel / SBS signature and in an indel / SBS profile to be compared are the same, and are defined using an indel / SBS classification as described herein.

[0074] Determining whether a mutational signature is present in a sample may be based on exposure of the mutational signatures in a mutational signature catalogue (also referred to as “mutational signature metrics”). Methods for determining the exposure to a mutational signature are described in e.g. Alexandrov et al., 2020; Degasperi et al., 2020; Degasperi et al., 2022. In particular, the determination of the exposure to one or more mutational signatures in a set may be performed by identifying the matrix E that satisfies C~PE where C is a mutational catalogue for one or more samples for which exposure is to be determined, P is a signature matrix comprising the one or more mutational signatures for which exposure is to be determined, and E is an exposure matrix. This may be performed using any NMF implementation with either a Kullback-Leibler divergence or Frobenius norm objective function. Examples include the implementation in the R package NNLM (see cran.nexr.com / web / packages / NNLM / vignettes / Fast-And-Versatile-NMF.html). The determination of the exposure to one or more mutational signatures in a first set of mutational signatures (or a combined set comprising a first set and one or more additional signatures from a second set) may be performed as described in Degasperi et al., 2020 (which describes determination of exposure to a set of signatures), or as described in Degasperi et al. 2022 (which describes the same process applied to a set of signatures comprising a first set (e.g. common signatures that are widely found across a plurality of cancer types) and a second set (e.g. rare signatures that are only identified in specific cancer types)). For example, for each sample a plurality of signature fitting processes (i.e. identifying the matrix E that satisfies C~PE) can be performed using bootstrapped data (perturbing the mutational profile by randomised re-sampling with replacement) in order to obtain a distribution of exposures for the sample, where a point estimate for the exposure is obtained as the median of the distribution. Mutational signatures with exposure of fewer than a predetermined number of mutations (e.g. 50 mutations) may be set to 0 to increase signature detection specificity. Instead or in addition to this, mutational signature exposures with a p-value below a threshold can be to 0 to increase specificity. The p-value may be an empirical p-value estimated from the bootstrapped data. For example, if more than 5% of the bootstrapped exposure values (empirical p-value 0.05) are below the threshold of 5% of total mutations in the sample, then the signature exposure may be set to 0.

[0075] The determination of the similarity between two mutational profiles, a mutational profile and a mutational signature, or two mutational signatures may be performed by calculating the cosine similarity between the two indel profiles or indel mutational signatures. The cosine similarity between two profiles or signatures can be calculated as: sim(S,M) = . where S and M are | |S| 111 M 11 equally-sized vectors with nonnegative components being the respective mutational profiles or mutational signatures (e.g. S being that of a sample and / W that of a reference profile such as e.g. a reconstructed profile or signature).

[0076] Thus, a profile (or indel profile, SBS profile, as the case may be) refers to a description of a particular sample I genome in terms of the number of mutations of a particular type (indels or SBS) within each of a plurality of categories (channels), as described herein. A signature (or mutational signature or indel mutational signature) refers to a pattern of occurrence of indels / SBS within each of the plurality of categories (channels) that has been previously observed in a plurality of samples / genomes. A mutational profile can result from a combination of one or more mutational signatures. In practice, a profile can be expressed as the number or proportion of all mutations of a particular type (i.e. indels or SBS) observed in a sample I genome in each of a predetermined plurality of categories as described herein. A signature can be expressed as a plurality of weights (typically summing to 1) for each of a predetermined plurality of categories as described herein. A profile can therefore be expressed as a linear combination of one or more signatures. In other words, one or more signatures of a previously identified set of signatures may be determined to be present in a sample / genome when the one or more signatures are the signatures of the previously identified set that are such that the linear combination of the one or more signatures best recapitulates the profile of the sample / genome. Additional filters may be applied to determine that a signature is present, such as e.g. filters that apply to the exposure or number of mutations attributed to the signature.

[0077] The number of mutations in a sample that are attributed to a particular mutational signature may be determined based on the exposure to the mutational signature and the total number of mutations in the mutational profile for the sample, as described further below.

[0078] Embodiments of the disclosure comprise a characterisation of a DNA sample in terms of the indel mutational signatures and / or SBS mutational signatures present in the sample. In embodiments, this is performed by a computer-implemented method or tool that takes as its inputs sequence data from the sample and an indel mutational signatures catalogue comprising one or more indel mutational signatures and / or an SBS mutational signatures catalogue comprising one or more SBS mutational signatures, and produces as output an indication of whether one or more of the indel mutational signatures and / or SBS signatures in the respective catalogues is / are present in the sample. Such an indication, for a particular signature, may referred to as a “mutational signature metric”. A mutational signature metric for a signature may be an exposure to the signature, and / or a number of mutations of the type catalogued in the signature (i.e. indels or SBS) identified in the sample that are attributed to the signature. The exposure can be obtained from the number of mutations attributed to the signature (and vice-versa) by dividing the latter by the total number of mutations of the type catalogued in the signature (i.e. total number of indels or SBS) identified in the sample.

[0079] In embodiments, the computer-implemented method or tool may take as its inputs a list of somatic indel mutations and / or a list of somatic SBS mutations generated from sequence data associated with a tumour sample (such as e.g. sequencing data obtained from genomic material from fresh- frozen derived DNA, circulating tumour DNA or formalin-fixed paraffin-embedded (FFPE) DNA representative of a suspected or known tumour from a patient). These somatic mutations can then be analysed to determine the value(s) of the one or more mutational signature metrics. The computer-implemented method or tool may take as its inputs sequence data associated with a tumour sample, and may use this data to generate a list of somatic indel mutations using any indel mutation calling process known in the art. The computer-implemented method or tool may take as its inputs sequence data associated with a tumour sample, and may use this data to generate a list of somatic SBS mutations using any SBS mutation calling process known in the art. These somatic indel / SBS mutations can then be analysed to determine the value(s) of the one or more mutational signature metrics. A list of somatic indel / SBS mutations may be obtained by identifying mutations present in sequence data associated with a tumour sample, and removing or otherwise excluding indel / SBS mutations that are present or assumed to be present in a corresponding germline genome. Mutations that are present in a corresponding germline genome may be identified by identifying the mutations present in a germline sample obtained from the same subject (also referred to as a “matched germline” or “matched normal” sample). Thus, the computer- implemented method or tool may further take as input sequence data associated with a matched germline sample. Mutations that are assumed to be present in a corresponding germline genome may be identified by identifying mutations that are present in a reference genome or set of reference genomes. A reference genome or set of reference genomes may be obtained from one or more reference samples that are not (or not all) matched normal samples. For example, the reference sample(s) may be process matched, or may comprise a plurality of normal (i.e. nontumour / non-modified) samples not all of which are matched to the sample for which a somatic indel mutational profile is determined (e.g. pooled normal samples may be used as references for a plurality of tumour samples). A reference genome or set of reference genomes may be obtained from one or more databases. ). A reference genome or set of reference genomes may be obtained from one or more databases. For example, a reference genome may be used and all mutations compared to this reference genome may be assumed to be somatic mutations. Alternatively, a set of reference genomes may be obtained from a database as a catalogue of known germline mutations in one or more populations (e.g. a genetic variation database such as dbSNP https: / / www.ncbi.nlm.nih.gov / snp / , 1000 genomes www.internationalgenome.org / , etc.). The use of a matched normal sample advantageously provides greatest certainty that the mutations identified in the DNA from the tumour sample are somatic mutations. The use of pooled normal samples comprising a matched normal sample may provide similar (though less precise information) and may be useful e.g. when sequencing resources are limited. Compared to the use of a matched normal sample, this may risk excluding more somatic mutations are seemingly germline mutations. The use of a reference genome or set of reference genome advantageously does not require the acquisition and analysis of a separate normal sample. However, the reference genome or set of reference genome is unlikely to capture all germline mutations present in the subject, and to include mutations that are in fact somatic in the subject. This is particularly true if a single reference genome is used rather than a collection capturing common sequence variation. Thus, this may result in a less accurate identification of somatic mutations. The computer-implemented method or tool may take as its inputs an indel and / or an SBS profile. As explained above, an indel profile comprises a count or proportion of indel mutations in each of a plurality of categories, obtained by categorising each somatic indel mutation identified in a sample in a class of a classification as described herein (i.e. a classification comprising the plurality of categories). Similarly, an SBS profile comprises a count or proportion of SBS mutations in each of a plurality of categories, obtained by categorising each somatic SBS mutation identified in a sample in a class of an SBS classification such as the trinucleotide (96 channels) classification described above (i.e. a classification comprising the plurality of categories). The computer- implemented method or tool may take as its inputs at list of somatic indel mutations or sequence data associated with a tumour sample, and may use this data to generate an indel profile using a classification as described herein. Similarly, the computer-implemented method or tool may take as its inputs at list of somatic SBS mutations or sequence data associated with a tumour sample, and may use this data to generate an SBS profile using e.g. the trinucleotide (96 channels) classification described above. The indel / SBS mutations that are included in the indel / SBS profile may be all somatic indel / SBS mutations identified in a sample. The indel / SBS mutations that are included in the indel / SBS profile may be all somatic indel mutations identified in sequence data from a sample. The sequence data may be genome-wide (e.g. from whole genome sequencing or shallow whole genome sequencing), or cover a substantial portion of the genome, such as e.g. exome wide (e.g. from whole exome sequencing) or from a targeted panel assay targeting a large number of regions (e.g. genes) such as e.g. 100 or more genes.

[0080] Embodiments of the disclosure use mutational signature metrics for one or more SBS signatures and / or one or more indel signatures, that have been previously determined, such as e.g. using a method as described herein. For example, the methods of the present disclosure may start from a list comprising mutational signatures metrics (E.g. exposures and / or number of mutations) associated with a plurality of signatures.

[0081] Figure 1 is a flow diagram showing, in schematic form, a method of determining the value of one or more mutational signature metrics for one or more mutational signatures. At optional step 10, a DNA sample is obtained from a tumour of a subject. Optionally, a matched normal sample may also be obtained from the subject. At optional step 12, sequence data is obtained from the tumour (and optionally the matched normal) DNA sample(s). At optional step 14, a plurality of somatic indel and / or SBS mutations are identified using the sequence data. At step 16, a catalogue of somatic indel mutations and / or a catalogue of somatic SBS mutations (indel mutational profile and / or SBS mutational profile) is obtained for the sample, for example by counting the number of indel mutations of each of a plurality of types (also referred to as “indel channels” “classes” or “categories”) to obtain an indel mutational profile, and / or by counting the number of SBS mutations of each of a plurality of types (also referred to as “SBS channels” “classes” or “categories”) to obtain an SBS mutational profile. This may also be described as classifying each of the plurality of indel / SBS mutations into one of a plurality of predetermined categories, by assigning a single one of the plurality of categories to each indel / SBS. This can also be seen as determining the numbers or proportions of the indel / SBS mutations present in the sample that fall in each of the plurality of categories. These numbers or proportions form an indel / SBS profile. Steps 10-14 are optional because determining the value of a mutational signature metric may start from sequence data, or directly from a list of identified indels / SBS or even from an indel / SBS mutational profile associated with the sample. Note that the entirety of the method of Figure 1 is also optional because the methods described herein may start from the values of one or more mutational signature metrics. The indel mutational profile and / or SBS mutational profile may then be used to determine the exposure to one or more indel mutational signatures and / or SBS signatures in a first set at step 18.

[0082] The results of step 18 may then be used at step 20 to identify at least one signature in a second set of signatures (separately for indel signatures and SBS signatures, the first and second set in each case consisting exclusively of indel signatures when an indel profile is analysed and SBS signatures when an SBS profile is analysed) that is likely to be present in the sample. This may comprise identifying an indel / SBS signature that is most similar to the residual between the indel / SBS mutational profile of step 16 (observed mutational profile) and a reconstructed indel / SBS mutational profile obtained using the results of step 18 (i.e. indel / SBS signatures and their exposures identified in step 18). A reconstructed indel / SBS mutational profile may be obtained as a linear combination of one or more indel / SBS signatures, weighted by the exposure to said signatures. Alternatively, step 20 may comprise identifying a signature that most reduces the error between observed and reconstructed mutational profile, when added to the first set of signatures. The process of step 18 may be repeated using the first set of signatures supplemented with a first signature (or set of signatures) of the second set of signatures identified in a first iteration of step 20. At step 22, a further set of exposures may be determined for the sample using the first set of signatures supplemented with any signatures of the second set of signatures identified in one or more iterations of step 20. In embodiments, only a first set of signatures may be used. In such embodiments, steps 20 and 22 may not be performed. The final set of exposures calculated at step 22 (if performed) or step 18 (if steps 20-22 are not performed) may be used as mutational signature metrics in methods as described herein. The method may further comprise receiving (for example from a user, through a user interface, or from one or more databases) one or more of: a first set of indel mutational signatures, a second set of indel mutational signatures, a first set of SBS mutational signatures, a second set of SBS mutational signatures, a set of reference signatures corresponding to the signatures in the first and / or second set, additional information associated with the sample such as e.g. known driver mutations, clinical information, etc. The methods described herein are applicable to indel / SBS profiles that have been obtained from any sequencing approach that allows identification of mutations over a substantial part of the genome. This may be achieved through whole genome sequencing (WGS), whole exome sequencing (WES) or any capture sequencing approach (i.e. targeted / bait based sequencing) that captures a portion of the genomes such as e.g. at least 10% of the genome, at least 20% of the genome, at least 30% of the genome, at least 40% of the genome, at least 50% of the genome, at least 60% of the genome at least 10% of the exome, at least 10% of the exome, at least 20% of the exome, at least 30% of the exome, at least 40% of the exome, at least 50% of the exome, at least 60% of the exome, at least 100 genes, at least 200 genes, at least 300 genes, at least 400 genes, at least 500 genes, and / or at least 1000 genes. Further, when incomplete genome sequencing is used, some of the genome may be imputed based e.g. on comparison with corresponding sequences in more complete profiles. In embodiments, the mutational catalogue and / or the mutational signatures catalogue has been determined from whole genome sequencing data or whole exome sequencing data.

[0083] Any set of mutational signatures (indel mutational signatures catalogue or SBS mutational signatures catalogue) may have been extracted from a plurality of indel / SBS profiles comprising counts of mutations for each of the plurality of categories as described herein, using non-negative matrix factorization (NMF). NMF may be used with Kullback-Leibler divergence (KLD) optimization, repeated bootstrapping (such as e.g. at least 300 bootstraps), and removal of local minima. For example, given a matrix of catalogs C, nonnegative matrix factorization (NMF) may be applied to 20 matrices C, bootstrapped from C. The NMF may be solved using an algorithm that optimizes the Kullback-Leibler divergence (KLD). Solving the NMF may produce a matrix of signatures S and a matrix of exposures E for each NMF run, such that C’ = SE. The NMF may be repeated a number of times (such as e.g. at least 200 or 300 times) for each bootstrap matrix, using random initializations. A set of solutions may be selected solutions that have a final KLD within a predetermined percentage (e.g. 0.1%) of the best solution found (the solution with the lowest KLD). Point estimates of exposures may be obtained as the median of the exposures obtained from bootstrapping.

[0084] Single Base Substitution Signatures

[0085] Embodiments of the present disclosure make use of single base substitution (SBS) signatures. Single base substitution signatures characterise the imprint of a mutational process on the mutational profile of cancer genomes in terms of the types and proportions of single base substitutions that are generated by the mutational process. Single base mutational signatures comprise weights for each of 96 categories of single base substitutions defined by mutation content and trinucleotide context (i.e. the base that is mutated and the surrounding 5’ and 3’ nucleotides). Catalogues of mutational signatures identified in pan-cancer genomes have been described in Alexandrov et al. 2020 and in Degasperi et al., 2022. The signatures described in Alexandrov et al. 2020 are available at cancer.sanger.ac.uk / cosmic / signatures / (hereafter “COSMIC” signatures I database). The signatures described in Degasperi et al., 2022 are available at signal.mutationalsignatures.com / explore / cancer (hereafter “Signal” signatures I database). These include signatures that correspond to the vast majority of the COSMIC signatures, and additional signatures. Corresponding signatures in the Signal and COSMIC databases are assigned the same name.

[0086] Embodiments of the present disclosure make use of particular SBS signatures, selected from: one or more MMRd-associated SBS signatures (such as e.g. signatures selected from: SBS6, 15, 26, 44, 97), one or more Pol-dys-associated SBS signatures (such as e.g. signatures SBS10a and / or SBSIOd), and / or one or more combined MMRd / Pol-dys SBS signatures (such as e.g. SBS14 and / or SBS20). Embodiments of these signatures are provided in Table 12. The examples of the present disclosure use Signal signatures. Therefore, in embodiments the reference to specific SBS signatures refers to signatures as described in Degasperi et al., 2022 and available at signal.mutationalsignatures.com / explore / cancer, such as e.g. the signatures in Table 12b below. SBS6, SBS15, SBS26 and SBS44 are signatures associated purely with post-replicative mismatch repair (MMR) deficiency. They are found in few samples (in large pan cancer cohorts), but are responsible for large mutation burdens in the samples in which they occurs. SBS6 is a signature associated with frequent C>T mutations, particularly in a ACG, CCG and GCG context. In other words, SBS6 is a signature that has higher weights for A[C>T]G, C[C>T]G and G[C>T]G classes of mutations than for any other classes of mutations. SBS15 is a signature associated with frequent C>T mutations, particularly in a GCG context. In other words, SBS15 is a signature that has higher weights for the G[C>T]G class of mutations than for any other classes of mutations. SBS26 is a signature associated with frequent T>C mutations in all contexts, but particularly in ATA, ATG, CTG, GTA, GTG, GTT, TTA and TTT contexts. In other words, SBS26 is a signature that has higher weights for A[T>C]A, A[T>C]G, C[T>C]G, G[T>C]A, G[T>C]G, G[T>C]T, T[T>C]A and T[T>C]T classes of mutations than for any other classes of mutations. SBS44 is a signature associated with frequent C[C>A]T mutations, and frequent T>C and C>T mutations in various contexts, particularly C>T mutations in ACA, ACG, GCA, GCC, GCG or GCT contexts and T>C mutations in GTA and TTT contexts. In other words, SBS44 is a signature that has higher weights for C[C>A]T, A[C>T]A, A[C>T]G, G[C>T]A, G[C>T]C, G[C>T]G, G[C>T]T G[T>C]A and T[T>C]T classes of mutations than for any other classes of mutations. SBS97 is a signature associated with MMR deficiency and replication strand bias. It is associated with frequent G[C>T]G and G[T>C]C mutations. In otherwords, SBS97 is a signature that has higherweights for G[C>T]G and G[T>C]C classes of mutations than for any other classes of mutations. SBS10a is a signature associated with mutations in the error-prone polymerase epsilon (POLE). It was first discovered in uterine cancers. It can cause a hypermutator phenotype. It is associated with frequent C>A mutations at TCT, as well as T>G mutations at TTT and C>T mutations at TCG. In other words, SBS10a is a signature that has higher weights for T[C>A]T, T[T>G]T and T[C>T]G classes of mutations than for any other classes of mutations. SBSIOd is a signature associated with mutations in polymerase delta (POLD). It is associated with frequent C>A mutations in a TCA and TCT contexts. In other words, SBSI Od is a signature that has higher weights for T[C>A]A and T[C>A]T classes of mutations than for any other classes of mutations. SBS14 has been attributed to a mixed phenotype of mismatch repair (MMR) deficiency and POLE dysfunction. Note that SBS10a is associated purely with POLE dysfunction. SBS14 is associated with frequent C>A mutations in a CCA, GCT or TCT context, and frequent C>T mutations in most contexts but particularly in a ACG, GCA, GCC, GCG, and GCT context. In other words, SBS14 is a signature that has higher weights for C[C>A]A, G[C>A]T, T[C>A]T, A[C>T]G, G[C>T]A, G[C>T]C, G[C>T]G, and G[C>T]T classes of mutations than for any other classes of mutations. SBS20 has been attributed to a mixed phenotype of MMR deficiency and POLD dysfunction. Note that SBSIOd is associated purely with POLD dysfunction. SBS20 is associated with frequent C>A mutations in a CCT context. SBS20 also includes relatively frequent C>T and T>C mutations in all contexts. In other words, SBS20 is a signature that has higher weights for the C[C>A]T class of mutations than for any other class of mutations.

[0087] The one or more SBS mutational signatures may be selected from the signatures listed in Table 12 (consensus signatures), corresponding tissue-specific signatures obtained using a conversion matrix, and / or corresponding signatures obtained by extracting one or more signatures from a cohort of samples using the same plurality of categories as those used to obtain the mutational profile for the sample. The one or more SBS mutational signatures may include one or more of indel signatures SBS6, 15, 26, 44, 97, 10a, 10d, 14 and 20 provided in Table 12, tissue-specific signatures corresponding to one or more of SBS6, 15, 26, 44, 97, 10a, 10d, 14 and 20 obtained using a conversion matrix, and / or one or more signatures corresponding to one or more of SBS6, 15, 26, 44, 97, 10a, 10d, 14 and 20 obtained by extracting one or more signatures from a cohort of samples using the same plurality of categories as those used to obtain the mutational profile for the sample. Corresponding signatures are signatures that have similar characterising features, i.e. weights that have the same properties as described above in terms of which SBS categories are associated with the highest weights in the signature.

[0088] Indel Signatures

[0089] Embodiments of the present disclosure make use of indel (InD) signatures. Indel signatures characterise the imprint of a mutational process on the mutational profile of cancer genomes in terms of the types and proportions of indels that are generated by the mutational process. Indel signatures comprise weights for each of a plurality of indel categories. Embodiments of the present disclosure make use of particular indel signatures, selected from: one or more MMRd-associated indel signatures (such as e.g. signatures lnD7 and / or lnD19), one or more Pol-dys-associated indel signatures (such as e.g. signatures Ind14 and / or lnD15), and / or one or more combined MMRd / Pol-dys indel signatures (such as e.g. signatures selected from: lnD16a, lnD16b, lnD20 and lnD21 ). Embodiments of indel signatures lnD7, 19, 14, 15, 16a, 16b, 20 and 21 are provided in Table 9. lnD7 is a signature associated with frequent 1 bp and 2bp deletions, particularly at long mono / dinucleotide repeat tracts (e.g. at least 5, 6, 7 or 8 repeats), and particularly involving 1 bp T deletions. It is similar to experimental signatures extracted from MLH1 , MSH2 and MSH6 gene knockouts. lnD7 is associated with MMRd. Thus, lnD7 may be defined as an indel signature that is associated with higher weights for categories comprising 2 or more bp deletions at repeat tracts of 5 or more repeats (e.g. Del(2,):U(1 ,2):R(5,9)), 1 bp T deletions at repeat tracts of 8 or more repeats in any sequence context (e.g. A[Del(T):R(8,9)]A, A[Del(T):R(8,9)]C, A[Del(T):R(8,9)]G, C[Del(T):R(8,9)]A, C[Del(T):R(8,9)]C, C[Del(T):R(8,9)]G, G[Del(T):R(8,9)]A, G[Del(T):R(8,9)]C, G[Del(T):R(8,9)]G), 1 bp T deletions at repeat tracts of 5 or more repeats in an AA sequence context (e.g. A[Del(T):R(5,7)]A), and 1 bp C deletions at repeat tracts of 6 or more bp (e.g. Del(C):R(6,9)), than for any other categories of indels. lnD19 is associated with PMS2 deficiency. It is characterised by frequent 2 or more bp insertions at longer repeat tracts (e.g. 5 or more repeats). lnD19 may be defined as a signature that has a higher weight for categories comprising 2 or more bp insertions at repeat tracts of 5 or more repeats (e.g. I ns(2 ,) :R(5,9)) , than for any other categories of indels. lnD14 is associated with POLD1 exonuclease mutations. It is associated with frequent 1 bp T insertions in repeat tracts of any length, particularly in repeat tracts of short to medium length (e.g. up to 7 repeats) in a 5’A sequence context. lnD14 may be defined as a signature that has higher weights for categories comprising 1 bp T insertions in repeat tracts of 5 to 7 repeats in a 5’A sequence context (e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, A[lns(T):R(5,7)]G), than for any other categories of indels. lnD15 is associated with POLE exonuclease mutations. It is associated with frequent 1 bp T insertions, particularly at long repeat tracts (e.g. 5 or more repeats). lnD15 may be defined as an indel signature that is associated with higher weights for categories comprising 1 bp T insertions in a plurality of sequence contexts, than for any other categories of indels. The plurality of sequence contexts may include AC (for all repeat lengths), AA, CA, CC, GA, GC (all particularly for repeats of 5 to 7 repeats), such as e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, C[lns(T):R(5,7)]A, C[lns(T):R(5,7)]C, G[lns(T):R(5,7)]A, G[lns(T):R(5,7)]C, and / or A[lns(T):R(8,9)]C. lnD16a is associated with concurrent loss of POLE proofreading and MMR. It is associated with frequent 1 bp T deletions at long repeat tracts (e.g. 5 or more repeats), particularly in AA sequence contexts. Ind16a may be defined as an indel signature that has higher weights for categories comprising 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context (e.g. A[Del(T):R(8,9)]A, A[Del(T):R(5,7)]A), than for any other categories of indels. lnD16b is associated with concurrent loss of POLE proofreading and MMR. It is associated with frequent 1 bp T deletions at long repeat tracts (e.g. 5 or more repeats), particularly in AA sequence contexts, as well as 1 bp T insertions at long repeat tracts (e.g. 5 or more repeats) particularly in AA and AC sequence contexts. Ind16b may be defined as an indel signature that has higher weights for categories comprising 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context (e.g. A[Del(T):R(8,9)]A, A[Del(T):R(5,7)]A) and 1 bp T insertions are repeat tracts of 5 to 7 repeats in AA or AC sequence contexts (e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C), than for any other categories of indels. lnD20 is associated with combined MMRd and POLE dysregulation. It is associated with frequent 1 bp T deletions in AA and to a lower extent AC contexts. lnD20 may be defined as an indel signature that is associated with higher weights for categories comprising 1 bp T deletions in AA contexts in repeat tracts of up to 7 repeats and 1 bp T deletions in AC contexts in repeat tracts of 5 to 7 repeats (e.g. A[Del(T):R(1 ,4)]A, A[Del(T):R(5,7)]A, A[Del(T):R(5,7)]C), than for any other categories of indels. lnD21 is associated with concurrent POLE proofreading defect and MMR. It is associated with frequent 1 bp T insertions, particularly at long repeat tracts (e.g. 5 or more repeats) and 1 bp T deletions, particularly at long repeat tracts (e.g. 5 or more repeats). Ind21 may be defined as an indel signature that is associated with higher weights for categories comprising 1 bp T insertions in repeat tracts of 5 to 7 repeats in 5’A and CG and GG sequence contexts, 1 bp T insertions in repeat tracts of 8 or more repeats AC sequence contexts, 1 bp T deletions in repeat tracts of 8 or more repeats in AA and CC sequence contexts (e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, A[lns(T):R(5,7)]G, C[lns(T):R(5,7)]G, G[lns(T):R(5,7)]G, A[lns(T):R(8,9)]C, A[Del(T):R(8,9)]A, C[Del(T):R(8,9)]C), than for any other categories of indels.

[0090] Thus, the one or more MMRd-associated indel signatures may comprise signatures that are primarily associated with 1 and / or 2 or more bp indels at long repeat tracts. These may include a signature that is characterised by frequent 2 or more bp insertions at long repeat tracts (e.g. 5 or more repeats) such as I n D7 , and a signature that is characterised by 1 and 2 or more bp deletions at repeat tracts of 5 or more repeats such as lnD19. The one or more Pol-dys-associated indel signatures may comprise signatures that are primarily associated with 1 bp T insertions. These may include a signature that is associated 1 bp T insertions in repeat tracts of 5 to 7 repeats in a 5’A sequence context such as In D14, and a signature that is associated with 1 bp T insertions in a plurality of sequence contexts including AC (for all repeat lengths), AA, CA, CC, GA, GC (all particularly for repeats of 5 to 7 repeats), such as lnD15.

[0091] The one or more combined MMRd / Pol-dys indel signatures may comprise signatures that are primarily associated with 1 bp T deletions at long repeat tracts (e.g. 5 or more repeats), 1 bp deletions, 1 bp insertions and deletions at long repeat tracts (e.g. 5 or more repeats), and 1 bp T deletions in AA contexts. These may include (i) a signature that is associated with 1 bp T deletions in AA contexts in repeat tracts of up to 7 repeats and 1 bp T deletions in AC contexts in repeat tracts of 5 to 7 repeats such as lnD20, (ii) a signature that is associated with 1 bp T insertions in repeat tracts of 5 to 7 repeats in 5’A and CG and GG sequence contexts, 1 bp T insertions in repeat tracts of 8 or more repeats AC sequence contexts, 1 bp T deletions in repeat tracts of 8 or more repeats in AA and CC sequence contexts such as lnD21 , (iii) a signature associated with 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context, such as lnD16a, and (iv) a signature associated with 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context and 1 bp T insertions are repeat tracts of 5 to 7 repeats in AA or AC sequence contexts, such as lnD16b.

[0092] The present disclosure describes methods of obtaining mutational signature metrics for indel mutational signatures for a sample, using an indel profile for the sample that classifies indel mutations according to a classification as described herein. A mutational catalogue summarising a list of mutations comprising both somatic insertions and deletions associated with a sample or group of samples may be referred to as an “indel profile”. A prior art classification to obtain indel profiles was described in Alexandrov et al., 2020, and is illustrated in Table 1 . The classification is sometimes referred to as the “COSMIC” indel signatures classification as it is described in the COSMIC database at cancer.sanger.ac.uk / cosmic / signatures / ID. According to the COSMIC classification, indels are classified as insertions or deletions, then if they are of a single base, as C or T, then according to the length of the mononucleotide repeat tract in which they occur (e.g. deletion of a T in a repeat tract of 5 Ts - resulting in a sequence of 4 Ts). Longer indels are classified as occurring at repeat regions (also referred to as “repeat mediated” indels) or with overlapping microhomology at deletion boundaries (whether 5’ or 3’), and according to the size of the indel, repeat and microhomology (e.g. deletion of a TC where the immediately following sequence is a TC is a deletion of 2 bp with repeat size 2, deletion of a TC where the immediately following sequence is TA is a deletion of 2bp with a homology size of 1). This classification scheme results in a total of 83 classes (also referred to as “channels”), listed in Table 1 below.

[0093] | Class description ]

[0094] Table 1. COSMIC classification of indels. The number of repeats is the total number of repeats including the repeats in the indel and those immediately 3’ of the indel and consecutive with the repeats of the indel. The length of the microhomology is the length of the portion of sequence of the indel that is repeated immediately 5’ or 3’ to the indel, but where the repeat is not immediately adjacent to the corresponding sequence in the indel.

[0095] The present disclosure describes a new classification scheme that more completely takes the context of the mutation into account. The classification scheme comprises a total of 476 channels and is described in Table 6 below. The complete classification can be collapsed into any classification scheme that combines one or more of the categories (‘channels’) in Table 6 below. In particular, any one or more of the categories in Table 6 can be combined by analysing the indel count per channel across a cohort of tumour samples, and combining channels that comprise a number of indels below a predetermined threshold (e.g. channels comprising 0 indels across the cohort, or that include fewer than 0.01 % of the indels across the cohort). An embodiment of such a reduced classification scheme comprises a total of 89 classes (also referred to as “channels”) and is shown in Table 2. Table 3 shows classes from Table 1 in which the indels in Table 2 would be classified, together with examples of the types of mutations in each of the classes of Table 2. Any classification scheme that includes a set of channels that distinguish between indels that have the same features (e.g. channels that include indels of the same length, inserting / deleting the same base, occurring in a repeat tract of the same length) other than their sequence context represents an embodiment of a classification scheme of the disclosure.

[0096] Table 2. Consolidated classification of indels. Del=deletion, lns=insertion, NonRep=non- repetitive insertion / deletion (no repetition of indel motif in 3’ context), nMer=repetitive insertions / deletion (insertion / deletion of a repeat motif without microhomology), Mh- del=deletion with microhomology. Si7 9i7

[0097] Table 3. Comparison of the classification schemes of Tables 1 and 2. The classification scheme of Table 1 does not consider complex events. Note that the R:0 notation for deletions in COSMIC corresponds to R1 in the present classification (and similarly R:1 corresponds to R2, etc.), as the present classification marks R as the number of repeats prior to deletion I insertion (i.e. in the present classification RO means 0 copies of an inserted I deleted sequence in addition to the sequence inserted / deleted, R1 means 1 copy of an inserted / deleted sequence prior to insertion / deletion, leading to 2 copies after insertion and 1 copy after deletion, etc.).

[0098] The classification in Tables 6, 2 and 3 (and generally any classification derived from the classification in Table 6 by combining one or more of the channels in Table 6) is based on the concept of a “unit” which is a sequence that is part of the indel and / or 3’sequence context and that can be repeated in the indel and / or 3’ sequence context. In order to classify indels with repeats or microhomology, the unit was defined as the minimal indel prefix that is maximally repetitive in the 3’ sequence context and in the indel itself. In particular, classifying an indel using a classification as described herein can comprise identifying a unit for the indel, wherein the prefix of an indel is the smallest possible unit that has the highest 3’ repetition, highest internal repetition, and smallest spacer (number of bases between the end of an internal repeat tract of the unit and the 3’ region), of a plurality of units of different lengths. A unit is a subsequence of the indel starting from the 5’ end of the indel. A unit may have a length selected from 1 bp to the length of the indel. The length of the prefix (selected unit), is referred to below as U. The words “prefix” and “unit” are used interchangeably below to refer to a selected unit. The word “unit” is also used to refer to candidate units that are being evaluated, one of which is selected as the unit that will be used for the purpose of classifying the indel (selected unit or prefix), as will be explained further below. Depending on the context it will be immediately obvious whether a candidate unit is intended (i.e. in steps leading to identification of a unit) or a selected unit I prefix (in steps associated with classification of an indel based on the selected unit). Thus, classifying an indel may comprise, for each of a plurality of units, determining: the number of 3’ repeats of the unit (3’ repetition, TR), the number of internal repetitions of the unit (IR, repeats of the unit within the indel), and the length of the spacer (S). Classifying an indel may further comprise identifying a unit of the plurality of units as the prefix of the indel, by applying the following criteria in the following order: maximise the number of 3’ repeat units, maximise the number of internal repetitions of the unit, minimise the spacer length, minimise the unit length. Applying criteria in a specific order comprises selecting units of the plurality of units that satisfy the first criterion, then amongst the selected units and if the number of selected units is >1 , selecting units of the plurality of units that satisfy the second criterion, etc. until a single unit remains. The plurality of units may include all subsequence of the indel starting from the 5’ end of the indel and with lengths between 1 bp and the total length of the indel (i.e. the first based of the indel, the first two bases of the indel, the first 3 bases of the indel, etc. up to the total length of the indel). For example, for the indel | 3’ sequence pair: ATCATCGA | ATCATCATCGGGGG, the smallest sequence that is maximally repeated in 3’ and within the indel is ATC (1 internal repeat, i.e. 2 internal copies, and three 3’ repeats, unit length of 3bp). In other words, for any insertion or deletion, the unit (for the purpose of determining the presence of a repeat or microhomology, identity and length of the repeated I microhomology sequence, and number of repeats of a sequence) may be identified by identifying a sequence (unit) within the inserted / deleted sequence that satisfies one or more of the following criteria: (i) has the maximum number of repeats in the sequence immediately 3’ of the inserted / deleted sequence; (ii) has the maximum number of repeats in the inserted / deleted sequence; (iii) has the minimum spacer length (number of bases between the repetitive region of the indel and the 3’ region); and (iv) has the minimum unit length (smallest repetitive unit that yields a segmentation that meets the above criteria). Each of these criteria may be applied hierarchically. Thus, the unit (for the purpose of determining the presence of a repeat or microhomology, identity and length of the repeated I microhomology sequence, and number of repeats of a sequence) may be identified by selecting the one or more sequence(s) that satisfy one of the above criteria, then selecting one or more of the selected sequences that satisfy another one of the above criteria, etc. until a single sequence remains. For example, for any insertion or deletion, the unit (for the purpose of determining the presence of a repeat or microhomology, identity and length of the repeated I microhomology sequence, and number of repeats of a sequence) may be identified by: (i) identifying the sequence (unit) within the inserted / deleted sequence that has the maximum number of repeats in the sequence immediately 3’ of the inserted / deleted sequence; (ii) optionally, when multiple sequences satisfy (i), identify the sequence (unit) within the inserted / deleted sequence amongst the sequences that satisfy (i), that has the maximum number of repeats in the inserted / deleted sequence; (iii) optionally, when multiple sequences satisfy (i) and (ii), identify the sequence (unit) within the inserted / deleted sequence amongst the sequences that satisfy (i) and (ii), that has the minimum spacer length (number of bases between the repetitive region of the indel and the 3’ region); and (iv) optionally, when multiple sequences satisfy (i), (ii) and (iii), identify the sequence (unit) within the inserted / deleted sequence amongst the sequences that satisfy (i), (ii) and (iii), that has the minimum unit length (smallest repetitive unit that yields a segmentation that meets the above criteria).

[0099] When a unit has been identified for an indel as explained above, it may be associated with a number or repetitions in the 3’ sequence prior to insertion / deletion (also referred to as “original repetitions”), R. For deletions, the number of repeats is equal to the number of 3’ repetitions (TR) plus the number of times the unit is repeated within the indel (IR, internal repetitions), plus the first occurrence of the unit (i.e. R=TR+IR+1). For insertions without spacer, the number of repeats is equal to the number of 3’ repetitions (R=TR). For insertions that are not perfectly repetitive (S>0), the number of 3’ repeats is still considered the most relevant and R=TR. A unit that is associated with a non-zero length spacer (S>0) may further be associated with a microhomology length, Mh or M. The microhomology length is the smallest of (i) the length of the prefix (selected unit) + the number of internal repeats multiplied by the length of the prefix (i..e U+IR*U), and (ii) the length of the prefix multiplied by the number of 3’ repeats (i.e. TR*U).

[0100] In Tables 2, 3 and 6, the following notations are used:

[0101] - R=number of repeats of a unit in the 3’ sequence prior to insertion / deletion; RO: R=0, sequence occurs once; R(0,3): sequence occurs from 0 to 3 times prior to insertion I deletion, for an insertion, R=0 means the inserted sequence is not repeated in the 3’ sequence, R=1 means the inserted sequence is repeated once in the 3’ sequence (and was therefore not a repeat sequence prior to insertion), etc.

[0102] - Ins(C) means insertion of a single C, Ins(T) means insertion of a single T, Ins(A) means insertion of a single A, Ins(G) means insertion of a single G.

[0103] - Del(C) means deletion of a single C, Del (T) means deletion of a single T, Del (A) means deletion of a single A, Del (G) means deletion of a single G.

[0104] - Ins(2,4) means insertions of lengths 2 to 4 nucleotides, I ns(5,) means insertions of 5 or more nucleotides.

[0105] - Del(2,4) means deletion of 2 to 4 nucleotides.

[0106] - Del(5,) means deletion of 5 or more nucleotides.

[0107] - A[lns(-)-]A: means a single base insertion in the sequence context ‘AA’ (i.e. sequence prior to insertion was ‘AA’ when the insertion is not repeated, R=0, or ‘A[n]RA’ where [n]R is the repeated sequence when R>0); similar principles apply to all other sequence contexts. When no context is provided, the category includes all contexts (i.e. equivalent to n[lns(-)-]]n), where n can be any nucleotide.

[0108] - [Del(-)-]A means a single base deletion with a 3’A (i.e. deletion or repeat sequence comprising deletion is immediately 5’ of an ‘A’), and any 5’ nucleotide (i.e. deletion is immediately 5’ of any nucleotide); this is equivalent to n[Del(-)-]A.

[0109] - A[Del(-)-]A means a single base deletion with a 5’A and a 3’A (i.e. the nucleotides immediately 5’ and 3’ of the deletion or repeat sequence comprising the deletion are A, i.e. sequence prior to deletion was ‘A[n]RA’ when R>0 and ‘AnA’ when R=0); similar principles apply to all other sequence contexts. When no context is provided, the category includes all contexts (i.e. equivalent to n[Del(-)-]]n).

[0110] - U (for deletions of 2 or more bp in repeat regions) is the length of the smallest unit that was repeated prior to deletion.

[0111] - M (for deletions of 2 or more bp with microhomology) is the length of the microhomology, which is the sequence that is homologous within the deleted region and 3’ of the deletion region, the repeats (homologous sequences) within and outside the deletion not being immediately adjacent to each other.

[0112] Note that 1 bp indels were categorised up to lengths of 9 bases because mutation calls at polytracts that are 10 or more were not made because they typically have low reliability since current sequence data does map well in these regions. However, longer indels could in theory be included in the R(8,9) classes.

[0113] The classification in Table 2 (and generally any classification derived from the classification in Table 6) differs from that in Table 1 (prior art) in many important ways. Firstly, none of the classes in Table 1 consider the sequence context of insertions and / or deletions (where sequence context refers to the nucleotide immediately 5’ and / or immediately 3’ of an insertion / deletion or repeat sequence (also referred to as “repeat tract” comprising the insertion / deletion). The present inventors have found that taking this into account increased the discriminatory capacity of indel signature analyses, enabling to distinguish between patterns that are known to be associated with different mutational mechanisms (e.g. different deficiencies in post-replicative repair, PRR). Further, none of the classes in Table 1 consider the 5’ sequence context of insertions and / or deletions in any way. Some of the classes in Table 1 consider the 3’ sequence context of the indel but only in so far as this is part of a repeat sequence, i.e. the 3’ sequence context of a repeat sequence is not considered. Secondly, the classification in Table 1 does not distinguish between indels at homopolymers >5bp (e.g. there is a single channel 1 :lns:T :5 for all T insertions in repeat tracts of 5 or more Ts, and a single channell :Del:T:5 for all T deletions in repeat tracts of 5 or more Ts, whereas the present inventors have found that different behaviours are associated with indels in repeats tracts of 5 to 7 Ts and repeat tracts of 8 or more Ts). The present inventors have found that separating indels at homopolymers >5bp between different classes (e.g. indels at homopolymers of 5-7bp vs 8 or more bp) increased the discriminatory capacity of indel signature analysis and in particular enable to accurately discriminate between samples that have mismatch repair deficiency (MMRd) and samples that have other replication error such as polymerase dysregulation. Conversely, the present inventors have found that some categories of the prior art classification did not capture many indels, thereby increasing the complexity of the scheme, reducing the power to detect signatures without increasing the discriminatory capacity of signature analyses as they do not actually contribute useful information. For example, the present inventors have found that it was more informative to distinguish between deletions of a single T in different sequence contexts than in different lengths of repeat tract between 1 and 4 (e.g. it was more important to distinguish between G[Del(T):R(1 ,4)]A and G[Del(T):R(1 ,4)]C - deletions of a single T within repeats of 1 to 4 Ts, within a sequence context of GA or GC - than between Ts 1 :Del:T:0, 1 :Del:T:1 , 1 :Del:T:2, 1 :Del:T:3 - deletions of a single T within repeats of 1 , 2, 3 or 4). In other words, the present inventors have found that increased granularity in terms of sequence context could be compensated for in terms of classification complexity by reduced granularity in terms of repeat lengths and / or repeated motif (unit) length.

[0114] The present inventors have identified that while some classes of indels benefited from additional granularity (distinguishing between different 573’ sequence context and additional length categories), the benefits of this vary by category and it is therefore possible to identify a classification that includes such additional granularity only for categories where this is beneficial, and combining categories that include very few data points for a particular dataset. For example, in the examples described below, the present inventors have identified that 1 bp insertions or deletions of a T would be particularly beneficially captured by more granular categories taking detailed sequence context into account, while 1 bp insertions or deletions of a single C could be accurately captured without the sequence context (for insertions) or based only on the 3’ sequence context (for deletions). Other datasets may benefit from more granular categorisation of C deletions based on sequence context.

[0115] Thus, an indel profile for a sample obtained according to the present disclosure may be defined as an indel profile that is obtained by classifying each of the plurality of indel mutations identified in a sample into one of a plurality of predetermined categories, wherein the plurality of predetermined categories comprise a plurality of categories comprising insertions of 1 bp, a plurality of categories comprising insertions of more than 1 bp, a plurality of categories comprising deletions of 1 bp, a plurality of categories comprising deletions of more than 1 bp, wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises categories that differ from each other by the 5’ and / or 3’ sequence context of the insertion / deletion or repeat tract comprising the insertion / deletion and / or wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises a plurality of categories comprising insertions / deletions in repeat tracts of 6 or more bp, said plurality of categories differing by the length of the repeat tract. Similarly, an indel signature according to the present disclosure may be defined as a plurality of weights for each of a plurality of predetermined categories of indel mutations, wherein the plurality of predetermined categories comprise a plurality of categories comprising insertions of 1 bp, a plurality of categories comprising insertions of more than 1 bp, a plurality of categories comprising deletions of 1 bp, a plurality of categories comprising deletions of more than 1 bp, wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises categories that differ from each other by the 5’ and / or 3’ sequence context of the insertion / deletion or repeat tract comprising the insertion / deletion and / or wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises a plurality of categories comprising insertions / deletions in repeat tracts of 6 or more bp, said plurality of categories differing by the length of the repeat tract. All categories, throughout this specification, are non-overlapping, i.e. a mutation can only belong to a single category of a mutational profile (whether indel or SBS profile). The plurality of predetermined categories may comprise at least one category comprising complex indels. Complex indels may be indels that result from a combination of one or more deletions and one or more insertions.

[0116] An insertion or deletion may be referred to as “repeat mediated” if it occurs in a repetitive region. According to the present disclosure, a repetitive region may be defined as a region that includes a plurality (e.g. 2 or more) of perfect, immediately adjacent repeats of a sequence motif. A sequence motif for the purpose of repeats may be defined as a sequence of between 1 and n bases, where n may be selected as 3, 4, 5, 6, 7, 8, 9, 10, 11 , 12 or more. For example n=8 may be convenient. The use of higher values of n requires more extensive cataloguing of such regions, which may be associated with diminishing returns as repeats of longer motifs are less likely. According to the present disclosure, an insertion may be referred to as “repeat mediated” if the insertion includes a motif of 1 or more base pairs that is immediately adjacent to one or more repeats of the same motif either inside or outside of the inserted region. Regions outside of the inserted region are typically 3’ regions. For example, insertion of a single T immediately 5’ of a T (whether a single T or a sequence of multiple T s) creates or adds to a homopolymer (multiple T) sequence and is therefore repeat mediated. By contrast, insertion of a single T immediately 5’ of any nucleotide that is not a T does not create a homopolymer and is therefore not repeat mediated. Thus, an insertion may be considered to be repeat mediated if it creates or increases the length of a homopolymer sequence comprising at least the 5’ end (i.e. any of the last 1 , 2, 3,... ,l nucleotides at the 5’ end of the insertion, where I is the length of the insertion) of the insertion. According to the present disclosure, a deletion may be referred to as “repeat mediated” if the deletion includes a motif of 1 or more base pairs that is immediately adjacent to one or more repeats of the same motif either inside or outside of the deleted region. For example, deletion of a single T immediately 5’ of a T (whether a single T or a sequence of multiple T s) removes or reduces the length of a homopolymer (multiple T) sequence and is therefore repeat mediated. By contrast, deletion of a single T immediately 5’ of any nucleotide that is not a T does not remove or reduce the length of a homopolymer and is therefore not repeat mediated. Thus, a deletion may be considered to be repeat mediated if it removes or reduces the length of a homopolymer sequence comprising at least the 5’ end of the deletion (i.e. any of the last 1 , 2, 3,... , I nucleotides at the 5’ end of the deletion, where I is the length of the deletion).

[0117] A deletion may be referred to as including “microhomology” at the deletion junction if the deletion occurs in a region that included a microhomology. According to the present disclosure, a microhomology may be defined as a plurality of perfect repeats of a sequence motif that are not immediately adjacent to each other. A sequence motif for the purpose of microhomology may be defined as a sequence of at least 1 base. A deletion at a sequence including a microhomology may be referred to as “microhomology-mediated”. A microhomology-mediated deletion may be described by one or more of: the length of the repeated sequence (M, microhomology length), the length of the spacer between the repeat sequence within the deletion and the repeated sequence in the 3’ context of the deletion (S), and the length of the deletion comprising the repeated sequence (microhomology) in the deletion and the spacer (motif size, i.e. M+S, noted as Del(), where M can also be expressed as Mh and is the total length of the microhomology, which can be the length of the minimal repeated unit within the microhomology, U, or a multiple of U if the repeated unit is repeated multiple times within the microhomology).

[0118] Referring back to Figure 1, in the context of indel signatures as described herein, step 16 may comprise performing, for each somatic indel mutation, one or more of the following steps. Step 16 may comprise step 16A of determining whether the indel mutation is an insertion, a deletion, or a combination of insertion(s) and deletion(s) (referred to herein as “complex indel”). In other words, step 16A may comprise classifying each mutation between one of three classes: a class comprising deletions, a class comprising insertions, and a class comprising complex indels. Step 14B may comprise determining, for each mutation classified as an insertion or a deletion, whether the insertion or deletion comprises 1 bp or more than 1 bp. In other words, step 14B may comprise classifying each mutation previously classified as insertion or deletion, between one of two classes for mutations classified as insertions (a class comprising insertions of 1 bp and a class comprising insertions of 2 or more bp) and one of two classes for mutations classified as deletions (a class comprising deletions of 1 bp and a class comprising deletions of 2 or more bp). In other words, step 16B may comprise: (i) for a mutation classified as an insertion at step 16A, classifying the mutation between a first class for insertions of 1 bp and a second class for insertions of more than 1 bp (step 16B), and (ii) for a mutation classified as a deletion at step 16A, classifying the mutation between a first class for deletions of 1 bp and a second class for deletions of more than 1 bp (step 16B’).

[0119] Step 16 may comprise step 16C of classifying each mutation previously classified as an insertion of 1 bp at step 16B: between a first plurality of classes comprising the insertion of a C / G and a second plurality of classes comprising the insertion of an A / T. Step 16 may comprise step 16C’ of classifying each mutation previously classified as a deletion of 1 bp at step 16B’: between a first plurality of classes comprising the deletion of a C / G and a second plurality of classes comprising the deletion of an A / T. The first plurality of classes (both for the 1 bp insertions and the 1 bp deletions) differ from each other by one or more of: the nucleotide immediately 3’ to the indel, the nucleotide immediately 5’ to the indel, or the number of repeats of the inserted I deleted basis at the locus of the mutation prior to the insertion / deletion. The first plurality of classes may comprise at least two classes that include insertions or deletions that have the same characteristic other than their 5’ sequence context. For example, the two classes C[Del(T):R(8,9)]G and G[Del(T):R(8,9)]G have the same characteristics other than their 5’ sequence context. The first plurality of classes may comprise at least two classes that include insertions or deletions that have the same characteristic other than their 3’ sequence context, where the 3’ sequence context refers to the nucleotide immediately 3’ of a homopolymer repeat comprising at last part of the insertion / deletion. For example, the two classes G[Del(T):R(8,9)]G and G[Del(T):R(8,9)]C have the same characteristics other than their 3’ sequence context. The characteristics may be selected from the 5’ or 3’ sequence context of the indel or indel repeat tract, and the number of repeats of the indel base in 3’ of the indel.

[0120] In embodiments, the first plurality of classes for the 1 bp insertions (i.e. plurality of classes for insertions of a single C) comprises or consists of the following classes: a. a category for a single C insertion in an “AA” sequence (A[lns(C):R0]A), b. a category for a single C insertion in an “AT” sequence (A[lns(C):R0]T), c. a category fora single C insertion in any sequence context other than “AA” and “AT”, with 0 to 3 Cs prior to insertion (lns(C):R(0,3)), d. a category for a single C insertion with 4 to 6 Cs priorto insertion (lns(C):R(4,6)), and e. a category fora single C insertion with 7 to 9 Cs prior to insertion (lns(C):R(7,9)). In embodiments, the second plurality of classes for the 1 bp insertions (i.e. plurality of classes for insertions of a single T) comprise or consist of: a plurality of classes for insertions with no T or 1 to 4 Ts prior to insertion, a plurality of classes for insertions with 5 to 7 Ts prior to insertion, and a plurality of classes for 8 or 9 Ts prior to insertion. Each of these plurality of classes may differ by the sequence context around the repeat tract, i.e. the 5’ and / or 3’ nucleotide surrounding the T or T tract (e.g. A-A, A-C, A-G, C-A, C-C, C-G, G-A, G-C, or G-G). In embodiments, the second plurality of classes for the 1 bp insertions (i.e. plurality of classes for insertions of a single T) comprises or consists of one or more (or all) of the following classes: A[lns(T):R(0,4)]A, A[lns(T):R(0,4)]C, A[lns(T):R(0,4)]G, C[lns(T):R(0,4)]A,

[0121] C[lns(T):R(0,4)]C, C[lns(T):R(0,4)]G, G[lns(T):R(0,4)]A, G[lns(T):R(0,4)]C, G[lns(T):R(0,4)]G,

[0122] A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, A[lns(T):R(5,7)]G, C[lns(T):R(5,7)]A, C[lns(T):R(5,7)]C,

[0123] C[lns(T):R(5,7)]G, G[lns(T):R(5,7)]A, G[lns(T):R(5,7)]C, G[lns(T):R(5,7)]G, A[lns(T):R(8,9)]A,

[0124] A[lns(T):R(8,9)]C, A[lns(T):R(8,9)]G, C[lns(T):R(8,9)]A, C[lns(T):R(8,9)]C, C[lns(T):R(8,9)]G,

[0125] G[lns(T):R(8,9)]A, G[lns(T):R(8,9)]C, and G[lns(T):R(8,9)]G.

[0126] In embodiments, the first plurality of classes for the 1 bp deletions (i.e. plurality of classes for deletions of a single C) comprises or consists of the following classes: a class for deletions of a single C in a repeat tract comprising 6 or more Cs prior to deletion (Del(C):R(6,9)), and a plurality of classes for deletions of a single C in a repeat tract comprising fewer than 6 Cs prior to deletion, wherein at least some of the classes differ from each other by the identity of the nucleotide 3’ of the repeat tract. The plurality of classes may comprise: (i) a class for deletions of a C with a G in 3’ of the repeat tracts ([Del(C):R(1 ,5)]G), and (ii) a plurality of classes for deletions of a C with an A or T in 3’ of the repeat tract, wherein the plurality of classes differ by the length of the repeat tract prior to deletion and / or the nucleotide in 3’ of the repeat tract (e.g. of one or more (or all) of the following classes may be used: [Del(C):R1]A, [Del(C):R2]A, [Del(C):R3]A, [Del(C):R(4,5)]A, [Del(C):R1]T, [Del(C):R2]T, [Del(C):R3]T, and [Del(C):R(4,5)]T).

[0127] In embodiments, the second plurality of classes for the 1 bp deletions (i.e. plurality of classes for deletions of a single T) comprises or consists of the following classes: a plurality of classes for deletions with 1 to 4 Ts prior to deletion, a plurality of classes for deletions with 5 to 7 Ts prior to deletion, and a plurality of classes for 8 or 9 Ts prior to deletion. Each of these plurality of classes may differ by the sequence context around the repeat tract, i.e. the 5’ and / or 3’ nucleotide surrounding the T or T tract (e.g. A-A, A-C, A-G, C-A, C-C, C-G, G-A, G-C, or G-G). In embodiments, the second plurality of classes for the 1 bp deletions (i.e. plurality of classes for deletions of a single T) comprises or consists of one or more (or all) of the following classes:

[0128] A[Del(T):R(1 ,4)]A, A[Del(T):R(1 ,4)]C, A[Del(T):R(1 ,4)]G, C[Del(T):R(1 ,4)]A, C[Del(T):R(1 ,4)]C,

[0129] C[Del(T):R(1 ,4)]G, G[Del(T):R(1 ,4)]A, G[Del(T):R(1 ,4)]C, G[Del(T):R(1 ,4)]G, A[Del(T):R(5,7)]A,

[0130] A[Del(T):R(5,7)]C, A[Del(T):R(5,7)]G, C[Del(T):R(5,7)]A, C[Del(T):R(5,7)]C, C[Del(T):R(5,7)]G,

[0131] G[Del(T):R(5,7)]A, G[Del(T):R(5,7)]C, G[Del(T):R(5,7)]G, A[Del(T):R(8,9)]A, A[Del(T):R(8,9)]C,

[0132] A[Del(T):R(8,9)]G, C[Del(T):R(8,9)]A, C[Del(T):R(8,9)]C, C[Del(T):R(8,9)]G, G[Del(T):R(8,9)]A,

[0133] G[Del(T):R(8,9)]C, and G[Del(T):R(8,9)]G.

[0134] Step 16 may comprise step 16D of classifying (i) each mutation previously classified as an insertion of 2bp or more at step 16B between a plurality of classes (step 16D) and (ii) each mutation classified as a deletion of 2bp or more at step 16B’ between a plurality of classes (step 16D’). The plurality of classes of insertions of 2bp or more (i.e. >1 bp) may differ from each other by the length of a repeated sequence (unit) in the insertion and / or the number of repeats of the unit prior to the insertion. A unit (also referred to herein as ‘motif’) is a particular sequence (1 or more bp) that is present in a sequence multiple times. A motif may be considered to be repeated if the motif is present in multiple immediately adjacent copies. The plurality of classes of insertions of 2bp or more may comprises one or more classes with no repeated motif (RO), and one or more classes with a repeated motif (R>0). The one or more classes with no repeated motif may comprise a plurality of classes that differ from each other by the length of the inserted sequence. The one or more classes with no repeated motif comprise insertions of 2 or more bp where no sub-sequence of the insertion is repeated (multiple copies of a motif immediately adjacent to itself) either within the insertion or within the insertion and the sequences immediately adjacent to it. For example, one or more classes with no repeated motif may comprise a first class for insertions of size 2 to 4 bp and a second class for insertions of 5bp or more. The one or more classes with a repeated motif comprise insertions of 2 or more bp where there is a sub-sequence of the insertion that is repeated multiple times either within the insertion or within the insertion and the sequences immediately adjacent to it. For example, one or more classes with a repeated motif may comprise one or more classes with 1 repeat of the motif - the one or more classes differing by the motif length (e.g. a class with 1 repeat of a motif of a first range of sizes - such as 2 to 4 bp, and a class with 1 repeat of a motif of a second range of sizes - such as 5 or more bp), and one or more classes with 2 or more repeats of the motif - the one or more classes differing by the motif length and / or the number of repeats (such as e.g. a class with 2, 3 to 4 repeats of a motif of any size, i.e. 2 or more bp, and a class with 5 or more repeats of a motif of any size, i.e. 2 or more bp). In embodiments, the plurality of classes of insertions of 2bp or more may comprise one or more (or all) of: lns(2,4):R0, lns(5,):R0, lns(2,4):R1 , lns(5,):R1 , lns(2,):R(2,4), lns(2,):R(5,9).

[0135] The plurality of classes of deletions of 2bp or more (i.e. more than 1 bp) may comprise a first plurality of classes comprising deletions in repeat regions and a second plurality of classes comprising deletions with microhomology at the deletion boundary (or boundaries). In embodiments, microhomologies are only considered at the 3' deletion boundary. The first plurality of classes (deletions of >1 bp in repeat regions) may comprise one or more classes comprising deletions with a single repeat (2 copies) prior to the deletion (differing by e.g. the length of the repeated motif, such as motifs of lengths 2 to 4 bp and motifs of length 5 or more bp, e.g. Del(2,4):R1 and Del(5,):R1), and one or more classes comprising more than 2 repeats prior to the deletion. The one or more classes comprising more than 2 repeats prior to the deletion may differ from each other by the length of the repeated motif (U), the number of repeats (R) and / or the length of the deleted sequence (Del()). For example, the one or more classes comprising more than 2 repeats prior to the deletion may comprise a first plurality of classes comprising repeated motifs of

[0136] 1 or 2 base pairs (U(1 ,2), differing from each other by the number of repeats -e.g. R(2,4) or R(5,9)- and / or the length of the deletion, such as e.g. a class for deletions of 2 to 8 bases with 2 to 4 repeats prior to deletion, and a class for deletions of any size with 5 or more repeats prior to deletion, e.g. Del(2,8):U(1 ,2):R(2,4) and Del(2,):U(1 ,2):R(5,9)) and a second plurality of classes comprising deleted units of 3 or more bases (U(3,), differing from each other by the length of the deletion and / or the number of repeats, such as e.g. Del(3,):U(3,):R2, and Del(3,):U(3,):R(3,9), which differ from each other by the number of repeats of the 3 or more bp units prior to deletion). In embodiments, the plurality of classes of deletions of2bp or more in repeat regions may comprise one or more (or all) of: Del(2,4):R1 , Del(5,):R1 , Del(2,8):U(1 ,2):R(2,4), Del(2,):U(1 ,2):R(5,9), Del(3,):U(3,):R2, and Del(3,):U(3,):R(3,9). In embodiments, Del(2,8) is equivalent to Del(2,) because deletions with microhomologies of lengths 9 and above may not be called (e.g. because they are less reliably called). Compared to the classification in Table 1 , this classification only uses

[0137] 2 classes for deletions with a single repeat of the motif prior to deletion, which differ by the range of length of the deletion (e.g. Del(2,4):R1 , Del(5,):R1 , whereas the classification in Table 1 has a separate class for each length, i.e. 2:Del:R:0, 3:Del:R:0, 4:Del:R:0, 5:Del:R:0). For deletions with more than one repeat unit prior to deletion, this classification uses two classes for deletions of repeat units of 1 or 2 base pairs and two classes for deletions of repeat units of 3 or more base pairs, the two classes differing by the range of number of repeats (e.g. 2 to 4 and 5 to 9, for the 1- 2 bp units, and 2 and 3 to 9 for the >3 bp repeat units), whereas the classification in Table 1 has a class foreach length and number of repeats (2:Del:R:1 , 3:Del:R:1 , 4:Del:R:1 , 5:Del:R:1 , 2:Del:R:2, 3:Del:R:2, 4:Del:R:2, 5:Del:R:2, 2:Del:R:3, 3:Del:R:3, 4:Del:R:3, 5:Del:R:3, 2:Del:R:4, 2:Del:R:5, 3:Del:R:4, 3:Del:R:5, 4:Del:R:4, 4:Del:R:5, 5:Del:R:4, 5:Del:R:5). Thus, the classification above uses 4 classes for these types of deletions rather than 20, and 6 classes for all >1 bp deletions without microhomology, instead of 24 in the prior art classification. The present inventors found that this resulted in more informative channels thereby increasing the power to detect signatures (as demonstrated in the examples below, see e.g. Fig.5A, 5B).

[0138] The second plurality of classes (deletions of >1 bp with microhomology at the deletion boundary / boundaries) may differ from each other by the length of the microhomology (M, which is a multiple of the length of the repeat unit U within the microhomology) and / or the length of the deletion comprising the repeated motif in the deletion and the spacer (motif size, i.e. M+S, noted as Del(), where M is the total length of the microhomology, which can be U or a multiple of U if the repeated unit is repeated multiple times within the microhomology). Note that prior art classifications only considered the length of the microhomology (M), not the length of the microhomology and the length of the deletion (including microhomology and spacer sequence). The length of the microhomology may be determined based on the length of the selected unit U (minimal repeated sequence, as explained above) as the smallest of: (i) the sum of the length of the minimal repeated sequence within the indel and the product of the number of additional repeats of the unit within the indel with the length of U (i.e. U+IR*U, where U is the length of the minimal repeat unit (selected unit) and IR is the number of internal repeats after the first identified minimal repeat unit), and (ii) the product of the 3’ number of repeats after the first identified minimal repeat unit with the length of the minimal repeated unit (TR*U, where TR is the number of repeats of the minimal repeat unit in the 3’ sequence context). The second plurality of classes may comprise one or more classes for deletion (motif) sizes of 6 or more bp, the one or more classes differing from each other by the length of the microhomology (such as e.g. respective classes for microhomologies with lengths 1 , 2, 3, and 4 or more), and one or more classes for deletion sizes of fewer than 6 bp that differ from each other by the deletion length and / or the microhomology length (such as e.g. a class for motifs of lengths 2 to 5 with microhomology length 1 , a class for motif lengths 3 to 5 with microhomology length 2, and a class of motif length 4 or 5 with microhomology length 3 or 4). In embodiments, the plurality of classes of deletions of 2bp or more with microhomology at the deletion boundary may comprise one or more (or all) of: Del(2,5):M1 , Del(3,5):M2, Del(4,5):M(3,4)m Del(6,):M1 , Del(6,):M2, Del(6,):M3, and Del(6,):M(4,). The classification in this embodiment separates deletions at micro-homologies by deletion length as well as microhomology length, but combines multiple classes of the prior art classification with microhomology lengths 1 or 2 and indel lengths below 6, which were found not to contribute significant information.

[0139] Repeats and / or microhomologies may be identified in an indel and in the sequence immediately 3’ of the indel. In other words, in embodiments, the sequence 5’ of the indel may be ignored for the purpose of determining the characteristic of a repeated motif (whether in a repeat region ora region with microhomology at the deletion boundary). In other words, repeat regions may comprise repeats immediately 3’ of an indel sequence, and microhomologies at the deletion boundary may comprise microhomologies in the sequence 3’ of the deletion. However, 5’ sequence context around a repeat unit or microhomology may be taken into account.

[0140] Step 16 may comprise, for each indel classified in a category comprising insertions / deletions of more than 1 bp, identifying a motif and a number of repeats or a microhomology of the motif in the sequence 3’ of the indel, by: identifying all sub-sequences starting from the 5’ end of the indel; for each sub-sequence, determine the number of perfect tandem repeats of the sub-sequence in the indel (IR, internal repetition) not including the first occurrence, and in the sequence immediately 3’ of the indel (TR, three prime repetition); determining the length of the remaining sequence in the indel between internal repeats and the end of the indel (S, spacer); and selecting a motif for the indel as the sub-sequence that: maximises TR when a single solution as the highest observed value of TR for the indel, maximises TR and IR when multiple solutions are possible with the same TR, maximises TR, IR and minimises S when multiple solutions are possible with the same TR and IR, or maximises TR, IR and minimises S and the length of the sub-sequence when multiple solutions are possible with the same TR, IR and S. This represents a unified strategy for determining the relationship between an indel and the 3’ context, applicable to all repeat configurations and microhomologies. This approach explicitly identifies the smallest prefix sequence motif that displays a maximal repetitive relationship with the 3' sequence context and within the indel itself. The sub-sequence may be referred to herein as “unit”. The selected subsequence may be referred to as “minimal unit”, “prefix”, “selected unit” or “motif’. The sequence immediately 3’ of the indel may be taken to be the first n base pairs located immediately 3’ of the indel, where n may be e.g. 100, 200, 300 or more. A value of 200 was found to strike a good balance between accuracy and complexity. The plurality of indel mutations may be indels of a size below a predetermined threshold, such as e.g. 100, 150 or 200 bp. A value of 100 was found to strike a good balance between accuracy and complexity.

[0141] The classification may include any of the channels in Table 6, or combinations thereof such as any of the channels in Tables 2-3 which combine one or more channels in Table 6. The classification may partition (i.e. classify) indels into the following groups (also referred to as “classes” or “categories”: (1) a first plurality of categories for 1 bp insertions and deletions; (2) a second plurality of categories for >1 bp insertions; (3) a third plurality of categories for >1 bp deletions without microhomology; and (4) a fourth plurality of categories for >1 bp deletions with microhomology.

[0142] The first plurality of categories (categories for 1 bp insertions and deletions) may comprise categories that differ by their pyrimidine content (i.e. C or T), their 5’ and 3’ context, and / or the length of their homopolymer tract. The length of the homopolymer tract may be discretised between a plurality of ranges of lengths of homopolymer tracts. For example, the present inventors have found three “bins” of length of homopolymer tract to be particularly useful for capturing the diversity of 1 bp indels. Convenient ranges include lengths between 0-4 bp or 0-3bp, 5-7bp or 4-6bp, and 8- 9bp or greater or 7-9bp or greater. In other words, 1 bp indels that have the same pyrimidine content (i.e. C or T), and the same 5’ and 3’ context may be separated between: a first class associated with a first range of homopolymer length (i.e. repeat R), such as 0-4bp, a second class associated with a second range of homopolymer length (e.g. 5-7 bp) and a third class associated with a third range of homopolymer length (e.g. 8bp or more). Note that other specific choices may be made which the number of classes or the precise boundaries of the classes, each specific choice combining in the same class 1 bp indels that have the same pyrimidine content (i.e. C or T), and the same 5’ and 3’ context, and a homopolymer length that falls within the same range. The first plurality of categories may comprise: a plurality of categories for T insertions and deletions, a plurality of categories for C insertions, and a plurality of categories for C deletions. Note that reference to T and C also encompass A and G as the same indel can be assessed on either strand of the genome. For T insertions and deletions, the present inventors found that it was beneficial to use 3 classes of homopolymer lengths, but consider all 5’ and 3’ single nucleotide combinations around the homopolymer variant. This was shown to optimally capture the diversity of 1 bp T indels. The 3 classes of homopolymer lengths may be for examples 0-4bp, 5-7bp and 8bp or more. Thus, the plurality of categories for 1 bp T indels may comprise 27 subcategories, each category being specified by: a 5’ and 3’ single nucleotide combination around the homopolymer of the indel, and a homopolymer length range (e.g. one of 1- 4 bp, 5 - 7 bp, 8 - 9 bp or more). For C insertions, the present inventors found that it was beneficial to use 3 classes of homopolymer lengths, and combine some of the 5’ and 3’ single nucleotide combinations around the homopolymer variant. This was shown to optimally capture the diversity of 1 bp C insertions, which were found to be less frequent than 1 bp T indels. The 3 classes of homopolymer lengths may be for examples 0-3bp, 4- 6bp and 7bp or more. This may combine all C insertions with R>1. For insertions with R=0 a plurality of classes may be distinguished based on the 5’ and 3’ context. The present inventors have found this type of C insertion to be more common and have found it beneficial to distinguish between C insertions with R=0 in a A / A context, C insertions with R=0 in a A / T context, and C insertions with R=0 in any other contexts, which are grouped together with the R>1 C insertions and only separated by range of homopolymer length. For C deletions, the present inventors found that it was beneficial to use different numbers of classes of homopolymer lengths, depending on the 3’ sequence context of the indel. For C deletions with A or T in 3’, 4 different classes of homopolymer lengths were found beneficial, leading to 8 classes, each characterised by a A 3’ context or T 3’ context and a range of homopolymer lengths. Classes of homopolymer lengths separating 1 bp, 2bp, 3bp and 4-5bp homopolymer lengths were found particularly useful. The present inventors found that for C / G 3’ sequence contexts all 1 bp C deletions with homopolymer lengths below a threshold (e.g. 1-5bp homopolymer lengths) could be combined in a single channel without significant loss of information. They further found that all 1 bp C deletions within homopolymers over a threshold length (e.g. more than 5bp) could be combined in a single channel without significant loss of information. Thus, the plurality of categories for 1 bp C deletions may comprise a category for homopolymer lengths above a threshold (e.g. more than 5 bp) in any sequence context, a category for homopolymer lengths below the threshold with a C or G 3’ sequence context, a plurality of categories for homopolymer lengths below the threshold with a A 3’ sequence context (the categories differing by homopolymer length, e.g. 1 bp, 2bp, 3bp, 4-5bp), and a plurality of categories for homopolymer lengths below the threshold with a T 3’ sequence context (the categories differing by homopolymer length, e.g. 1 bp, 2bp, 3bp, 4-5bp).

[0143] For the second plurality of categories (plurality of categories for >1 bp insertions), the present inventors found that indels could be grouped into categories that differ by their range of homopolymer lengths, without significant loss of information. The use of 4 categories that differ by their range of R was found particularly useful, such as e.g. R=0, R=1 , 2 < R < 4, and R > 5. Any other specific choices of ranges of R that differentiate between (i) non-templated (R=0), (ii) duplications (R=1), (iii) moderately repetitive (e.g. 2 < R < 4) and (iv) highly repetitive (R > 5) insertions of >1 bp may be used. In some cases the present inventors further found that some of these categories could be further separated into categories that differ by indel size, in particular distinguishing between short and long indels based on a threshold length (e.g. insertions of up to 4 bp vs 5bp or more). Thus, the plurality of categories for >1 bp insertions may comprise two categories for R=0 (differing by their range of indel lengths, e.g. insertions of up to 4 bp vs insertions of 5 or more bp), two categories for R=1 (differing by their range of insertion length, e.g. insertions of up to 4 bp vs insertions of 5 or more bp), a category for R between 2 and 4, and a category for R=5 or more.

[0144] For the third plurality of categories (categories for >1 bp deletions without microhomology - i.e. no spacer, S=0, exact repeats), the inventors found that indels could be grouped into categories that differ by their repetitive content. They identified that deletions with R=0 could be separated from deletions with R>0 (multiple copies of the unit within the indel and / or in the 3’ sequence context). The deletions with R=0 were further separated by length, leading to two categories for R=0, one for short deletions (length below a threshold) and one for long deletions (length above a threshold). A threshold of 5 was found particularly useful, i.e. creating (i) a class for all deletions of >1 bp without microhomology, with R=0 and with length of deletion up to 4bp, and (ii) a class for all deletions of >1 bp without microhomology, with R=0 and with length of deletion of 5 or more bp. The deletions with R>0 were separated between a plurality of classes that differ by the length of the unit (e.g. a first class with unit lengths below a threshold and a first class with unit lengths above a threshold, such as 3pb). Some of these classes were further separated by degree of repetition, distinguishing a plurality of classes of >1 bp deletions without microhomology, with units in a specified range of lengths, where the classes differ by the range of R covered (e.g. R of up to 4 vs R of 5 or more). The present inventors found it beneficial to keep all longer unit deletions in a single class (e.g. 1 bp deletions without microhomology, with unit lengths of 3 or more, with R>0), while separating shorter unit deletions based on range of R (e.g. 1 bp deletions without microhomology, with unit lengths of 1 or 2, with R> 0 but falling in different classes depending on the R, e.g. R below 5 vs R of 5 or more). Thus, the third plurality of categories may comprise: a plurality of categories for R=0 deletions (the plurality of categories differing by length of the deletion, e.g. a first class for 2 - 4 bp, a second class for> 5 bp), a category for R>0 deletions with unit length above a threshold (e.g. units of 3bp or more), and a plurality of categories for R>0 deletions with unit length below the threshold (the plurality of categories differing by range of R, e,g, a category for number of repeats R below a threshold and a category for number of repeats R at or above the threshold, where the threshold can be e.g. 5).

[0145] For the fourth plurality of categories (categories for >1 bp deletions with microhomology), the present inventors found that it was beneficial to distinguish indels by length of microhomology and length of the deletion. Indeed, they observed that variants with spacers tended to cluster broadly as short (2- 5 bp) and long (> 6 bp). They also observed that microhomology lengths were frequent for shorter lengths (< 3bp), and displayed a long tail (> 4 bp). They therefore designed a plurality of categories that differ by their microhomology length, e.g. microhomology lengths of 1 bp, 2bp, 3 bp, or 4 or more bp. Some of these were further separated between categories that differ by the length of the deletion. Thus, the fourth plurality of categories may comprise: a plurality of categories for microhomology length 1 (differing by the deletion length, such as e.g. deletion length below or above a threshold, e.g. deletion length 2 to 5 bp vs 6 or more bp), a plurality of categories for microhomology length 2 (differing by the deletion length, such as e.g. deletion length below or above a threshold, e.g. deletion length 3 to 5 bp vs 6 or more bp), a category for microhomology length 3 or 4 and deletion length below a threshold (e.g. deletion length 4 or 5), a category for microhomology length 3 and deletion length at or above the threshold (e.g. deletion length 6 or more), and a category for microhomology length 4 and deletion length at or above the threshold (e.g. deletion length 6 or more).

[0146] Note that the particular combinations of channels defined by R, S, U and M values and 5’ and 3’ context as described above that is beneficial may depend on the context. Indeed, combinations are made to preserve informative signal while grouping together categories that are less informative. For example, categories that have a number of indels below a predetermined number or threshold may be grouped together with similar categories. The above provides a grouping that the present inventors have found to be optimal in the context of analysing pan cancer genomes, and this is therefore believed to be likely usable in the majority of cases. However, when particular types of mutations or genomes that have specific unusual features are being investigated, it may well be that a slightly different group of categories would be even more informative. Nevertheless, the present inventors believe and demonstrated below that the present method for classifying indels is more informative than existing methods that are widely applied for cancer genome analysis. Thus, the present and derived approaches are expected to improve on the current state of the art methods to analyse genome wide patterns of indel mutations.

[0147] The one or more indel mutational signatures may be selected from the signatures listed in Table 9 (consensus signatures), corresponding tissue-specific signatures obtained using a conversion matrix, optionally as listed in Table 8, and / or corresponding signatures obtained by extracting one or more signatures from a cohort of samples using the same plurality of categories as those used to obtain the mutational profile for the sample. The one or more indel mutational signatures may include one or more of indel signatures lnD7, 19, 14, 15, 16a, 16b, 20 and 21 provided in Table 9, tissue-specific signatures corresponding to one or more of indel signatures lnD7, 19, 14, 15, 16a, 16b, 20 and 21 obtained using a conversion matrix, optionally as listed in Table 8, and / or one or more signatures corresponding to one or more of indel signatures lnD7, 19, 14, 15, 16a, 16b, 20 and 21 obtained by extracting one or more signatures from a cohort of samples using the same plurality of categories as those used to obtain the mutational profile for the sample. Corresponding signatures are signatures that have similar characterising features, i.e. weights that have the same properties as described above in terms of which indel categories are associated with the highest weights in the signature.

[0148] Prediction of PRR status

[0149] In embodiments of the present invention, a prediction of whether a DNA sample from a tumour of a patient is PRR proficient or deficient is performed. In embodiments, this prediction is performed by a computer-implemented method or tool that takes as its inputs sequence data from the sample or the values of one or more mutational signature metrics derived therefrom, for example as explained above and by reference to Figure 1 , and produces as output a probabilistic score indicative of whether the sample is PRR proficient or deficient, or information derived therefrom such as a classification of the sample as likely PRR deficient I unlikely PRR deficient, and / or a probabilistic score indicative of whether the sample has a particular type of PRR deficiency selected from MMRd, pol-dys and combined MMRd / pol-dys, or information derived therefrom such as a classification of the sample as between a plurality of categories comprising a category for the particular type of PRR deficiency.

[0150] In some embodiments, the present invention provides methods for classifying samples from tumours between a plurality of classes comprising: a class associated with PRR deficiency or a plurality of classes that are associated with different types of PRR deficiency, and a class associated with PRR proficiency. In particular, mutational signature metrics may be evaluated using one or more pattern recognition algorithms. Such analysis methods may be used to form a predictive model, which can be used to classify test data. For example, one convenient and particularly effective method of classification employs multivariate statistical analysis modelling, first to form a model (a "predictive mathematical model") using data ("modelling data") from samples of known subgroup (e.g., from subjects known to have a PRR deficient or PRR proficient tumour, from subjects known to have a tumour with a particular type of PRR deficiency), and second to classify an unknown sample (e.g., "test sample") according to subgroup. Pattern recognition methods have been used widely to characterize many different types of problems ranging, for example, over linguistics, fingerprinting, chemistry and psychology. In the context of the methods described herein, pattern recognition is the use of multivariate statistics, both parametric and non-parametric, to analyse data, and hence to classify samples and to predict the value of some dependent variable based on a range of observed measurements. In the context of the present invention, "supervised" approaches are suitably used, whereby a training set of samples with known class or outcome is used to produce a mathematical model which is then evaluated with independent validation data sets. Here, a "training set" of mutational signature metrics associated with samples is used to construct a statistical model that predicts correctly the "subgroup" of each sample. This training set is then tested with independent data (referred to as a test or validation set) to determine the robustness of the computer-based model. These models may be based on a range of different mathematical procedures such as: (i) for binary classifications: logistic regression models, support vector machine, decision trees, k-nearest neighbour and naive Bayes classifiers; and (ii) for multiclass classification: multinomial regression models, neural networks such as multiclass perceptrons, k-nearest neighbours, naive Bayes classifiers, decision trees, and multiclass support vector machines. The robustness of the predictive models can for example be checked using cross-validation, by leaving out selected samples from the analysis. In embodiments, the model is a logistic regression model (for binary classification) or a multinomial regression model (for multiclass classification).

[0151] In embodiments, the model is a multinomial regression model. The model may be a multinomial regression model that has been trained using a regularisation method, such as an elastic net penalty. Thus, the model may be a multinomial elastic net regression model. Multinomial elastic net regression models are described in Chen et al. 2014. A multinomial elastic net regression model is a multiclass classification model based on regularised logistic regression. Logistic regression determines the probability of a sample belonging to a positive class (p) given a vector x of features associated with the sample using log (such that p = 1 / (1 + e-( / ?o+xLi / ?«))) where p are trained coefficients (Po is an intercept and P; are weights for each of the k predictive features x; considered). By extension, multinomial regression models determine the probability of a sample belonging to a particular class c of a set of classes C (P(y=c|x)) using P(y = c|x) = bclXl'), where for each class c there are a set of trained coefficients bc(intercept), bCj (weights for each of the k predictive features x; considered). Thus, the probability for each class is obtained as the ratio of a term for the class and the sum of terms for all classes, and the probabilities across classes for a sample sum to 1 . The coefficients can be obtained as those that minimise a loss function (e.g. likelihood loss function). When regularisation is used, the coefficients can be obtained as those that minimise the sum of a loss function and one or more penalty for model complexity, such as e.g. the multiclass elastic net penalty.

[0152] Figure 2A is a flow diagram showing, in schematic form, a method of characterising a DNA sample according to the disclosure. At optional step 210, a DNA sample is obtained from a tumour of a subject. Optionally, a matched normal sample may also be obtained from the subject. At optional step 212, sequence data is obtained from the tumour (and optionally the matched normal) DNA sample(s). At optional step 214, the value of a plurality of mutational signature metrics for the tumour DNA sample are obtained. The mutational signature metrics include the number of indels or proportion of indels in the sample associated with one or more indel signatures of postreplication repair (PRR)-deficiency and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)- deficiency, which are determined in step 214D. These numbers or proportions may be based on determined exposures to these indel / SBS signatures, which are obtained in steps 214A-214B, as explained by reference to Figure 1. Note that in Steps 10-14 in Figure 1 and Steps 214A-214C on Figure 2A are optional because the method may start from sequence data, from a mutational profile associated with the sample, or directly from the (previously determined) value of the one or more mutational signature metrics (e.g. exposures) described above. The plurality of mutational signature metrics may comprise: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)-deficiency, wherein the one or more indel signatures of PRR-deficiency comprise one or more indel signatures associated with combined MMRd and pol-dys (MMRD / pol-dys) and one or more indel signatures associated with MMRd; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)- deficiency, wherein the one or more SBS signatures of PRR-deficiency comprise one or more SBS signatures associated with pol-dys, and one or more SBS signatures associated with MMRd. At step 214E, an additional optional metric may be obtaiened as the ratio of the number of indels to the number of SBS in the sample, ora value derived therefrom. The values obtained at steps 214D and 214E may be log transformed, then normalised (e.g. standardised) at step 214F. The values obtained at step 214D are advantageously sums of numbers associated with a plurality of signatures.

[0153] At step 216, it is determined whether said sample has a high or low likelihood of being PRR- deficient, wherein a sample that has a high likelihood of being PRR-deficient has a high likelihood of having mismatch repair deficiency (MMRd), polymerase dysfunction (pol-dys), or combined mismatch repair deficiency and polymerase dysfunction (MMRd / pol-dys). This may comprise classifying said sample between a class associated with a high likelihood of being PRRd and a class associated with a low likelihood of being PRRd, or (as illustrated), step 216A of classifying the sample between a plurality of classes comprising: a class associated with a high likelihood of MMRd, a class associated with a high likelihood of pol-dys, a class associated with a high likelihood of combined MMRd / pol-dys, and a class associated with a high likelihood of PRR-proficiency. Step 216A may be performed using a trained machine learning model that takes as input the values of said mutational signature metrics and produces as output one or more probabilities and / or classification labels indicative of whether the sample has a high or low likelihood of being PRR- deficient or a high or low-likelihood of being in each of the above classes. A class associated with a high likelihood of PRR-proficiency may be a class associated with a low likelihood of PRR- deficiency. A class associated with a high likelihood of MMRd may be referred to as an “MMRd class”, “class associated with MMRd”, or “class comprising samples with MMRd”. A class associated with a high likelihood of pol-dys may be referred to as a “pol-dys class”, “class associated with pol-dys” or “class comprising samples with pol-dys”. A class associated with a high likelihood of combined MMRd / pol-dys may be referred to as a “MMRd / pol-dys class”, “class associated with MMRd / pol-dys” or “class comprising samples with MMRd / pol-dys”. A class associated with a high likelihood of PRR-proficiency may be referred to as a “PRR-proficient”, “normal” or “negative control” class, or “class comprising samples without PRRd”.

[0154] The machine learning model may have been trained using training data comprising the values of said mutational signature metrics for a plurality of training samples with known PRR status. The plurality of training samples may each be associated with a label selected from: PRR proficient, MMRd, pol-dys, combined MMRd / pol-dys. Methods of training such a model are also described herein. The machine learning model may generate one or more probabilistic scores at step 216B. The probablistic scores may comprise a first probabilistic score that is such that the sample has a high likelihood of being PRR-deficient when the probabilistic score is at or below a threshold and a low likelihood of being PRR-deficient when the probabilistic score is above the threshold. Thus, this score can be compared to a threshold at step 216C to determine whether the tumour is PRRdeficient. The machine learning model may further produce: a second probabilistic score associated with a second class associated with a high likelihood of MMRd, a third probabilistic score associated with a third a class associated with a high likelihood or pol-dys, and a fourth probabilistic score associated with a fourth class associated with a high likelihood of combined MMRd / pol-dys. The method may comprise determining that the sample has a low likelihood of being PRR-deficient when said first probabilistic score for said sample is above a threshold, or determining that the sample has a high likelihood of being PRR-deficient when said first probabilistic score for said sample is at or below the threshold. Determining that the sample has a high likelihood of being PRR-deficient when said first probabilistic score for said sample is at or below the threshold comprises may comprise determining that the sample has the type of PRRd associated with the class of the second, third or fourth class that is associated with the highest probabilistic score. The machine learning model according to any embodiment may be selected from: one or more logistic regression models, a support vector machine, one or more decision trees, a k-nearest neighbour classifier, a naive Bayes classifiers, a multinomial regression model, a neural network, and a multiclass support vector machine.

[0155] At step 218, the results of any of the steps above are provided to a user. These may include one or more of: the value of the mutational signature metrics, a value derived therefrom (such as e.g. a probabilistic score or classification label), a determination of whether the sample has a high likelihood or a low likelihood of being PRR-deficient, and a determination of whether the sample has a high likelihood of having a particular type of PRR-deficiency selected from MMRd, pol-dys and combined MMRd / pol-dys.

[0156] Determining whether said sample has a high or low likelihood of being PRR-deficient at step 216 may comprise using said values of said one or more mutational signature metrics to classify said sample between a class associated with a high likelihood of being PRRd and a class associated with a low likelihood of being PRRd. The classification may be binary or may be a multi-class classification. Thus, the class associated with a high likelihood of being PRRd may comprise a plurality of classes. The plurality of classes may be associated with non-overlapping subsets of: mismatch repair deficiency (MMRd), polymerase dysfunction (pol-dys), or combined mismatch repair deficiency and polymerase dysfunction (MMRd / pol-dys). For example, a first class may be associated with MMRd, a second class may be associated with pol-dys alone or combined with MMRd, and a third class may be associated with PRR proficiency. As another example, a first class may be associated with MMRd alone or combined with pol-dys, a second class may be associated with pol-dys alone, and a third class may be associated with PRR proficiency. Alternatively, a first class may be associated with MMRd, a second class may be associated with pol-dys alone, a third class may be associated with combined MMRd / pol-dys, and a third class may be associated with PRR proficiency. Determining whether said sample has a high or low likelihood of being PRR-deficient may be performed based on the values of one or more further metrics in addition to the values of the one or more mutational signature metrics. For example, in the embodiment illustrated on Figure 2A, the ratio of numbers of indels to number if SBS or a value derived therefrom is optionally additionally used. For example, a value derived from said ratio may be (total number of indel+1 ) / (total number of SNV+1), log((total number of indel+1 ) / (total number of SNV+1)) (where the log can be in any base), or a normalised version thereof.

[0157] Optional normalisation may be performed as illustrated on Figure 2A (step 214F). Normalisation of any value (including but not limited to mutational signature metrics) may be performed by standardisation across a reference set of samples, such as e.g. samples used fortraining a model to predict the likelihood of PRR deficiency. Normalisation may also be referred to as “scaling”. Thus, determining the value of a mutational signature metric for the sample may comprise scaling the value of said mutational signature metric. Scaling the mutational signature metrics may advantageously increase the comparability of the values of the respective variables and reduce the risk that metrics that are on different scales disproportionately affect the probabilistic score obtained. Scaling may be performed using any method known in the art, such as e.g. by normalisation (also known as min-max scaling, i.e. transforming a variable such that the range of possible values for the variable ranges between 0 and 1), or by standardisation (where values are centred around the mean with a unit standard deviation by, for each observation, subtracting the mean and dividing by the standard deviation for the variable). The present inventors have found standardisation to result in better performing models. Standardisation was found to strike a good balance between simplicity and improving the comparability of the variables thus improving the performance of the PRR deficiency identification process. The scaling may be performed using one or more parameters for each mutational signature metric, such as e.g. a value by which every value for a particular metric should be divided in order to obtain the corresponding derived (i.e. normalised) value. Thus, the method may further comprise receiving or determining the value of said one or more parameters. The values of said one or more parameters (e.g. mean and standard deviation of a mutational signature metric for standardisation) may be values obtained using a training set of samples. For examples, the values in T able 11 (last 2 columns) may be used. These may be used in combination with the model parameters in Table 11 , which were obtained using the same training data. Alternative model parameters and normalisation parameters corresponding to those in Table 11 may be used instead, which parameters have been obtained using a different training dataset. A log transformation may be any log transformation, e.g. Iog2.

[0158] The method may further comprise receiving (e.g. from a user through a user interface, or from a database) or determining any of the predetermined thresholds used. For example, the threshold used to compare the first probabilistic score may be determined (or may have been determined) using test data comprising the values of said probabilistic score for a plurality of samples that have a known PRR deficiency status. For example, the predetermined threshold(s) may be chosen so as to optimise (maximise or minimise, as the case may be) one or more performance metrics such as accuracy, specificity or sensitivity of detection of samples from PRR-deficient tumours.

[0159] The probabilistic scores may be determined using a multinomial regression model. A probabilistic score for each particular class c (P(y=c|x)) of a set of classes C comprising the first, second, third and fourth classes may be determined as P(y = each class c, coefficients bc(intercept), and bCj (weights for each of the k predictive features x considered) are trained parameters of the model and xi are the values provided as input to the model (including the plurality of mutational signature metrics and optionally the ratio of number of indels to number of SBS or values derived therefrom).

[0160] The machine learning model used or trained according to any embodiment may comprise trained parameters comprising weights that apply to the inputs of the model (including the mutational signature metrics) and characterise the relative importance of the inputs to the prediction made by the machine learning model. The machine learning model may comprise trained parameters comprising a plurality of weights each associated with a respective class of a plurality of classes represented in the model, wherein the trained weights associated with a class characterise the relative importance of the inputs to the prediction made by the machine learning model of whether the sample belongs to the class. The plurality of weights may assign importance to the following variables in the following order (where importance refers to the absolute values of the weights and one or more of the variables may be omitted depending on the embodiment): (i) for the prediction of a first probabilistic score associated with a first class associated with a high likelihood of PRR- proficiency: a mutational signature metric associated with one or more SBS signatures associated with mismatch repair deficiency (MMRd), a mutational signature metric associated with one or more indel signatures associated with mismatch repair deficiency (MMRd), a mutational signature metric associated with one or more SBS signatures associated with pol-dys, a mutational signature metric associated with one or more indel signatures associated with pol-dys, a mutational signature metric associated with one or more indel signatures associated with combined MMRd / pol-dys, a ratio of the number of indels to the number of SBS in the sample or value derived therefrom, and a mutational signature metric associated with one or more SBS signatures associated with combined MMRd / pol-dys; (ii) for the prediction of a second probabilistic score associated with a second class associated with a high likelihood of MMRd: a mutational signature metric associated with one or more SBS signatures associated with mismatch repair deficiency (MMRd), a mutational signature metric associated with one or more indel signatures associated with mismatch repair deficiency (MMRd), a ratio of the number of indels to the number of SBS in the sample or value derived therefrom, a mutational signature metric associated with one or more SBS signatures associated with combined MMRd / pol-dys, a mutational signature metric associated with one or more SBS signatures associated with pol-dys, a mutational signature metric associated with one or more indel signatures associated with pol-dys, and a mutational signature metric associated with one or more indel signatures associated with combined MMRd / pol-dys; (iii) for the prediction of a third probabilistic score associated with a third a class associated with a high likelihood or pol- dys: mutational signature metric associated with one or more SBS signatures associated with pol- dys, a ratio of the number of indels to the number of SBS in the sample or value derived therefrom, a mutational signature metric associated with one or more indel signatures associated with combined MMRd / pol-dys, a mutational signature metric associated with one or more indel signatures associated with pol-dys, and a mutational signature metric associated with one or more indel signatures associated with MMRd; and / or (iv) for the prediction of a fourth probabilistic score associated with a fourth class associated with a high likelihood of combined MMRd / pol-dys: a mutational signature metric associated with one or more indel signatures associated with combined MMRd / pol-dys, mutational signature metric associated with one or more SBS signatures associated with pol-dys, a mutational signature metric associated with one or more SBS signatures associated with combined MMRd / pol-dys, a ratio of the number of indels to the number of SBS in the sample or value derived therefrom, a mutational signature metric associated with one or more indel signatures associated with pol-dys, a mutational signature metric associated with one or more indel signatures associated with MMRd, a mutational signature metric associated with one or more SBS signatures associated with pol-dys, and a mutational signature metric associated with one or more SBS signatures associated with MMRd.

[0161] The trained parameters may further comprise an intercept, for example when a linear model is used (such as e.g. a logistic regression model or multinomial regression). When a multiple linear model is used (e.g. multinomial regression), the trained parameters may comprise an intercept for each of the plurality of classes represented in the model. The trained parameters may have values selected from Table 11 , where the second to fifth columns provide the values of trained parameters for each of a first to fourth classes as indicated. As the skilled person understands, the exact values of the weights used are likely to depend on the training data used. For example, the examples herein demonstrate howto obtain suitable values using training data comprising colorectal cancer, uterine cancer and stomach cancer samples. Using a different training data set (comprising additional samples and / or different samples such as e.g. samples from other types of tumours, or different definitions for the mutational signatures, such as e.g. the SBS signatures in Table 12a instead of Table 12b) may result in different weights. However, the relative importance of the variables may remain similar. In embodiments, the machine learning model is able to provide a prediction of whether a sample has a high or low likelihood of being mismatch repair (PRR)-deficient with above 99% accuracy (as evaluated using the AUC metric), on at least one test set of samples. The test set of samples preferably comprises at least 50 samples, or at least 100 samples. The test set of samples may comprise samples from one or more types of tumours. The one or more types of tumours in the test set of samples may be represented in the training set used to train the machine learning algorithm. The test set of samples may comprise colorectal cancer samples, stomach cancer samples and / or uterine cancer samples. The test set of samples and the training set of samples preferably comprise samples that are known to be PRR deficient and samples that are known to be PRR proficient. The test set of samples and / or the trainings et of samples preferably comprise a plurality of samples that are known to be PRR deficient and a plurality of samples that are known to be PRR proficient. The test set of samples and / or the training set of samples preferably comprise a plurality of samples that are known to be PRR deficient and to have a type of PRR deficiency selected from pol-dys, MMRd and combined MMRd / pol-dys, and a plurality of samples that are known to be PRR proficient. The test set of samples and / or the training set of samples preferably comprise: a plurality of samples that to have pol-dys, a plurality of samples that to have MMRd and a plurality of samples that to have combined MMRd / pol-dys, PRRd status and type may be determined based on driver mutation status (e..g mutations in specific genes involved in PRR) and / or immunohistochemistry.

[0162] According to any embodiment, determining the value of one or more indel mutational signature metrics for the sample may comprises cataloguing the somatic indel mutations in said sample to produce an indel profile for that sample, wherein the value of said indel mutational signature metrics is derived from said indel profile. Determining the value of one or more SBS mutational signature metrics for the sample may comprises cataloguing the somatic SBS mutations in said sample to produce an SBS profile for that sample, wherein the value of said SBS mutational signature metrics is derived from said SBS profile. Cataloguing the somatic mutations in said sample comprises determining the number of mutations in the mutational catalogue which are attributable to each of a plurality of base substitution classes or indel classes (depending on whether an indel profile or an SBS profile is being generated) which are determined to be present. The base substitution classes may include all possible trinucleotide substitution classes. The base substitution classes may be described according to the “96 channels convention” known in the art, i.e. the product of 6 types of substitution multiplied by 4 types of 5’ base (A,C,G,T) and 4 types of 3’ base (A,C,G,T). Trinucleotide substitution classes are listed in Table 12 (first column). The indel classes may include all classes of an indel classification as described herein.

[0163] Obtaining an indel profile for the sample may comprise classifying each of the plurality of indel mutations into one of a plurality of predetermined categories, wherein the plurality of predetermined categories comprise a plurality of categories comprising insertions of 1 bp, a plurality of categories comprising insertions of more than 1 bp, a plurality of categories comprising deletions of 1 bp, a plurality of categories comprising deletions of more than 1 bp, and optionally at least one category comprising complex indels, wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises categories that differ from each other by the 5’ and / or 3’ sequence context of the insertion / deletion or repeat tract comprising the insertion / deletion and / or wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises a plurality of categories comprising insertions / deletions in repeat tracts of 6 or more bp, said plurality of categories differing by the length of the repeat tract.

[0164] According to any embodiment, the sequence data that is used or from which the mutational signature metrics have been derived may be obtained or may have been obtained using whole genome sequencing. Whole genome sequencing enables accurate identification of both single base substitutions and indels, and therefore enables determination of mutational signature metrics for both SBS signatures and indel signatures. The method may further comprise providing to a user one or more of: the value of the mutational signature metrics, a value derived therefrom (such as e.g. a probabilistic score or classification label), a determination of whether the sample has a high likelihood or a low likelihood of being PRR-deficient, and a determination of whether the sample has a high likelihood of having a particular type of PRR-deficiency selected from MMRd, pol-dys and combined MMRd / pol-dys. The method may further comprise obtaining a germline sample from the subject and / or obtaining sequence data from a germline sample from the subject. The tumour sample may be a sample comprising tumour cells or genetic material derived therefrom. The tumour sample may be a sample of cells or tissue that has been obtained directly from a tumour (e.g. a tumour biopsy). The tumour sample may be a sample comprising cells or genetic material derived from a tumour, such as e.g. a liquid biopsy sample comprising circulating tumour cells or circulating tumour DNA.

[0165] Determining a mutational signature metric for a mutational signature may comprise comparing the mutation profile (indel or SBS profile) of the sample to said mutational signature (indel or SBS signature, as the case may be), wherein a mutational signature comprises a plurality of weights characterising the effect of a mutational process associated with the signature on a genome in terms of the proportion of indels / SBS in each of the plurality of predetermined categories.

[0166] The method may further comprise receiving (for example from a user, through a user interface, or from one or more databases) one or more of: one or more mutational signature(s) of PRR, clinical features associated with the sample, parameters of a predictive model used at step 16A, etc. Uses of Predictor Outcome

[0167] A prediction of whether a tumour is likely to be PRR deficient can be used in the treatment of cancer. Thus, the invention also provides a method of treating cancer in a subject, wherein the method comprises administering or recommending a subject for administration of a particular therapy, depending on whether a tumour of the subject is identified as likely to be PRR deficient. For example, tumours that are deficient in mismatch repair are known to be susceptible to immune checkpoint inhibition, and tumours with PRR deficiency even with intact MMR are also likely to be susceptible to such therapies. Therefore, a subject with a tumour identified as likely to be PRR deficient may be treated or recommended for treatment with a immune checkpoint inhibitor. The immune checkpoint inhibitor may be administered alone or in combination with a DNA damaging agent.

[0168] Figure 2B illustrates a method of providing a prognosis and / or treating a subject that has been diagnosed with cancer, according to embodiments described herein. The method may comprise optional step 30 of obtaining a DNA sample from a tumour of a subject. Optionally, a matched normal sample may also be obtained from the subject. The step of obtaining a sample from a subject may comprise physically obtaining the sample from the subject. Alternatively, the sample may have been previously obtained and no interaction with the subject may be required. In other words, obtaining a DNA sample may comprise receiving a previously acquired DNA sample. At optional step 32, sequence data is obtained from the tumour (and optionally the matched normal) DNA sample(s). The step of obtaining sequence data from a DNA sample may comprise sequencing the DNA sample. Alternatively, sequence data may have been previously obtained. Thus, obtaining sequence data may comprise receiving the data from one or more databases, or from a user through a user interface. At step 34, it is determined whether the tumour sample has a low or high likelihood of being PRR deficient and / or whether the tumour sample has a low or high likelihood of being having a particular type of PRR deficiency, using methods described herein such as e.g. by reference to Figure 2A. Based on this determination, the subject may be classified as having a good or poor prognosis at step 36A (as will be explained further below). Instead or in addition to this, the subject may be classified at step 36B as being likely to respond or unlikely to respond to a particular course of treatment, where responder / non-responder status is known to be associated with PRR-deficiency (i.e. tumours that are PRR-deficient are known to be more or less likely to respond to the particular course of treatment, compared to tumours that are not PRR deficient). At optional step 38, a particular course of treatment (which may comprise one or more different individual therapies) may be identified based on the results of step 36B. For example, a subject that has been identified at step 36B as unlikely to respond to the particular course of therapy may be identified as likely to benefit from a therapy that is different from the particular course of therapy. Alternatively, a subject that has been identified at step 36B as likely to respond to the particular course of therapy may be identified as likely to benefit from a therapy that includes the particular course of therapy. At optional step 40, the subject may be treated with the therapy identified at step 40.

[0169] In particular, MMR deficient cancers and pol-dys cancers have been identified as having an increased likelihood of response to immunotherapy, and particularly checkpoint inhibitors (CPI). CPI therapy includes for example treatment with an anti-CTL4 or anti-PD(L)1 drug. Thus, also described herein are methods of determining whether a subject that has been diagnosed as having a cancer is likely to benefit from treatment with an immunotherapy, preferably a CPI therapy, the method comprising determining the PRR status of a tumour from the subject (i.e. determining whether the subject has a high or low likelihood of having a tumour that is PRR deficient) using the methods described herein. The method may further comprise classifying the subject between a group that is likely to respond to CPI therapy, and a group that is not likely to respond to CPI therapy. For example, the method may comprise determining whether a sample from a tumour of the subject has a high or low likelihood of being PRR deficient (as explained above), or to have PRR deficiency of a particular type. A subject may then be classified in the group that is not likely to respond to CPI therapy if the sample is determined to have a low likelihood of being PRR deficient, and in a group that is likely to respond to CPI therapy otherwise. Alternatively, a subject may be classified in the group that is not likely to respond to CPI therapy if the likelihood of PRR deficiency (e.g. as captured in a probabilistic score as described above) is below a threshold, and in the group that is likely to respond to CPI therapy otherwise.

[0170] In some cases CPI therapy may comprise CTLA-4 blockade (cytotoxic T-lymphocyte associated protein 4, Gene ID:1493), PD-1 inhibition (PDCD1 , programmed cell death 1 , Gene ID:5133), PD- L1 inhibition (CD274, CD274 molecule, Gene ID: 29126), Lag-3 (Lymphocyte activating 3; Gene ID: 3902) inhibition, Tim-3 (T cell immunoglobulin and mucin domain 3; Gene ID: 84868) inhibition, TIGIT (T cell immunoreceptor with Ig and ITIM domains; Gene ID: 201633) inhibition and / or BTLA (B and T lymphocyte associated; Gene ID: 151888) inhibition. The CPI therapy may be an anti- PD1 or anti-PDL1 therapy (also referred to as anti-PD(L)1 inhibitor). The inhibitor may be a therapeutic antibody. For example, the CPI therapy may be a PD-1 inhibitor such as pembrolizumab, nivolumab, or tislelizumab. Pembrolizumab is a therapeutic antibody that has been approved by the FDA (U.s> Food and Drug Administration) for patients with unresectable or metastatic microsatellite instability-high (MSI-H) or mismatch repair deficient (dMMR) solid tumours that have progressed following prior treatment. This indication is independent of PD-L1 expression assessment, tissue type and tumour location. Nivolumab is a therapeutic antibody used to treat various cancers including melanoma, lung cancer, renal cell carcinoma, Hodgkin lymphoma, head and neck cancer, colon cancer, and liver cancer. Tislelizumab is a therapeutic antibody under investigation for the treatment of advanced solid tumours. The CPI therapy may be a PDL-1 (also referred to as “PD-L1”) inhibitor such as atezolizumab, avelumab, or durvalumab. Atezolizumab is a therapeutic antibody used to treat urothelial carcinoma, non-small cell lung cancer (NSCLC), triple-negative breast cancer (TNBC), small cell lung cancer (SCLC), and hepatocellular carcinoma (HCC). It was the first PD-L1 inhibitor approved by the FDA. Avelumab is a therapeutic antibody used for the treatment of Merkel cell carcinoma, urothelial carcinoma, and renal cell carcinoma. Durvalumab is a therapeutic antibody that has been approved by the FDA for the treatment of certain types of bladder and lung cancer. As another example, the CPI therapy may be a CTLA-4 inhibitor, such as ipilimumab or tremelimumab. Ipilimumab is a therapeutic antibody approved by the FDA for the treatment of melanoma, and under investigation for the treatment of non-small cell lung cancer, small cell lung cancer, bladder cancer and metastatic hormone-refractory prostate cancer. Tremelimumab is a therapeutic antibody under investigation for the treatment of melanoma, mesothelioma and non-small cell lung cancer.

[0171] As explained above, PRR deficiency encompasses MMR deficiency, polymerase dysfunction, and combinations of these. However, prior to the present invention, no method existed to identify tumours that have any of these deficiencies. At most, MMR deficient vs proficient tumours could be identified. This missed out on potentially providing life-saving treatments for tumours that have polymerase dysfunction orcombined MMRd / pol-dys, that could not be reliably identified using prior art methods.

[0172] Further, MMR deficient cancers have been identified as having a decreased likelihood of response to fluorouracil based treatment (e.g. adjuvant 5-fluorouracil chemotherapy) and / or an increased likelihood of response to non-fluorouracil based treatments (Devaud & Gallinger, 2013; Jover et al., 2009). Thus, also described herein are methods of determining whether a subject that has been diagnosed as having a cancer is likely to benefit from treatment with chemotherapy, preferably a fluorouracil based therapy or a non-fluorouracil based therapy, the method comprising determining the PRR status of a tumour from the subject using the methods described herein. Such a method may further comprise classifying the subject between a group that is likely to respond to fluorouracil based therapy, and a group that is not likely to respond to fluorouracil-based therapy. For example, the method may comprise determining whether a sample from a tumour of the subject has a high or low likelihood of being PRR deficient (as explained above), or more specifically MRRd or combined MMRd / pol-dys. A subject may then be classified in the group that is likely to respond to fluorouracil-based therapy if the tumour is determined to have a low likelihood of being PRR deficient (or to have a low likelihood of having MRRd or combined MMRd / pol-dys), and in a group that is not likely to respond to fluorouracil-based therapy otherwise. Alternatively, a subject may be classified in the group that is not likely to respond to fluorouracil-based therapy if the likelihood of PRR deficiency or more specifically MRRd or combined MMRd / pol-dys (e.g. as captured in a probabilistic score as described above) is above a threshold, and in the group that is likely to respond to fluorouracil-based therapy otherwise. Alternatively, such a method may comprise classifying the subject between a group that is likely to respond to non-fluorouracil based therapy, and a group that is not likely to respond to no- fluorouracil-based therapy. For example, the method may comprise determining whether a sample from a tumour of the subject has a high or low likelihood of being PRR deficient (as explained above). A subject may then be classified in the group that is likely to respond to non-fluorouracil- based therapy if the tumour is determined to have a high likelihood of being PRR deficient, and in a group that is not likely to respond to non-fluorouracil-based therapy otherwise. Alternatively, a subject may be classified in the group that is not likely to respond to non-fluorouracil-based therapy if the likelihood of PRR deficiency (e.g. as captured in a probabilistic score as described above) is below a threshold, and in the group that is likely to respond to fluorouracil-based therapy otherwise.

[0173] Any treatment described herein may be used alone or in combination with another treatment. For example, any treatment with a drug may be used in combination with one or more chemotherapies, one or more course of radiation therapy, and / or one or more surgical interventions. In particular, any treatment described herein may be used in combination with a treatment for which the subject has been identified as likely to be responsive. For example, a subject may be identified as likely to be deficient for homologous recombination (HRdeficient) using one or more methods known in the art. Such a subject may be treated or identified as likely to benefit from treatment with a PARP inhibitor or platinum-based drug. For example, a subject may be identified as likely to be HRdeficient using the methods described in WO 2018 / 115452 or WO 2017 / 191074, or likely to respond to a PARP inhibitor or a platinum-based drug using the methods described in WO 2017 / 191073. As a particular example, a method of treating a subject that has been diagnosed as having cancer may comprise: determining whether the subject is likely to benefit from treatment with an immunotherapy, preferably a CPI therapy, the method comprising determining the PRR status of a tumour from the subject using the methods described herein; and determining whether the subject is likely to benefit from treatment with a PARP inhibitor or platinum based therapy, the method comprising determining the HR status of a tumour from the subject, for example using the methods described in WO 2018 / 115452 or WO 2017 / 191074. Such a method may further comprise treating the subject with an immunotherapy (e.g. a CPI therapy, such as a PD1 / PDL1 inhibitor) if the subject has been identified as likely to be PRR deficient, and / or treating the subject with a PARP inhibitor or platinum-based therapy if the subject has been identified as likely to be HR deficient.

[0174] Additionally, the PRR status of a tumour has been shown to be associated with different prognosis in cancer (see e.g. Sinicrope, 2009). For example, MMR deficient tumours have been associated with improved prognosis compared to non-MMR deficient tumours, for example in terms of disease-free survival and overall survival. Thus, also described herein are methods of providing a prognosis for a subject that has been diagnosed as having a cancer, the method comprising determining the PRR status of a tumour from the subject. The method may further comprise classifying the subject between a group that has good prognosis, and a group that has poor prognosis. For example, the method may comprise determining whether a sample from a tumour of the subject has a high or low likelihood of being PRR deficient (as explained above), or having PRR deficiency of a particular type (as explained above). A subject may then be classified in the group that has poor prognosis if the sample is determined to have a low likelihood of being PRR deficient, and in a group that has good prognosis otherwise. Alternatively, a subject may be classified in the group that has poor prognosis if the likelihood of PRR deficiency (e.g. as captured in a probabilistic score as described above) is below a threshold, and in the group that has good prognosis otherwise.

[0175] Whether a prognosis is considered good or poor may vary between cancers and stage of disease. In general terms a good prognosis is one where the overall survival (OS), disease free survival (DFS) and / or progression-free survival (PFS) is longer than that of a comparative group or value, such as e.g. the average for that stage and cancer type. A prognosis may be considered poor if OS, DFS and / or PFS is lower than that of a comparative group or value, such as e.g. the average for that stage and type of cancer. Thus, in general terms, a “good prognosis” is one where survival (OS, DFS and / or PFS) and / or disease stage of an individual patient can be favourably compared to what is expected in a population of patients within a comparable disease setting. Similarly, a “poor prognosis” is one where survival (OS, DFS and / or PFS) of an individual patient is lower (or disease stage worse) than what is expected in a population of patients within a comparable disease setting.

[0176] The subject is typically a mammalian, preferably a human patient. The present inventors have shown that the methods described herein are applicable to any cancer type. The methods described herein may be particularly useful in the context of subjects having or being likely to have a cancer where PRR deficiencies are relatively common. In particular, the methods described herein may be used to characterise any type of cancer that is known to have PRR deficient subpopulations or in which PRR deficiencies have been reported in at least some patients. The cancer may be ovarian cancer, breast cancer, endometrial cancer (uterus / womb cancer), kidney cancer (renal cell), lung cancer (small cell, non-small cell and mesothelioma), brain cancer (gliomas, astrocytomas, glioblastomas), melanoma, merkel cell carcinoma, clear cell renal cell carcinoma (ccRCC), lymphoma, gastrointestinal cancer (e.g. colorectal cancer), small bowel cancers (duodenal and jejunal), leukemia, pancreatic cancer, hepatobiliary tumours, germ cell cancers, prostate cancer, head and neck cancers, bladder cancer, thyroid cancer and sarcomas. The cancer may be colorectal cancer, endometrial / uterine cancer, stomach cancer, bladder cancer, central nervous system cancer, or lung cancer. PRRd has been identified in significant proportions of cases in each of these types of cancers. In embodiments, the cancer may be colorectal cancer, breast cancer, endometrial cancer, prostate cancer, bladder cancer or thyroid cancer, all of which are known to have MMR deficient subpopulations. As another example, the cancer may be colorectal cancer, endometrial / uterus cancer, biliary caner, bone / soft tissue cancer, breast cancer, central nervous system cancer, choroid melanoma, carcinoma of unknown primary (CUP), oesophagus cancer, head and neck cancer, kidney cancer, liver cancer, lung cancer, lymphoid cancer, neuroendocrine tumour (NET), ovarian cancer, pancreatic cancer, prostate cancer, skin cancer, stomach cancer, urinary tract cancer. All of these have been tested with the methods described herein. In embodiments, the cancer is colorectal cancer or uterine cancer, both of which are shown herein to contain high proportions of PRRd cancers. The links between MMR deficiency and prognosis as well as therapy response in colorectal cancer has been extensively studied and as such there is strong evidence that treatment and prognosis in such cancers can be adjusted using information regarding the PRR status of such cancers. Such information is more accurately obtained using the methods described herein, compared to the prior art. As such, the treatment strategy designed for a subject and / or the prognosis provided for a subject having colorectal cancer can be improved using the methods of the present invention.

[0177] Systems

[0178] Figure 3 shows an embodiment of a system for characterising a DNA sample and / or for providing a prognosis or treatment recommendation, according to the present disclosure. The system comprises a computing device 1 , which comprises a processor 101 and computer readable memory 102. In the embodiment shown, the computing device 1 also comprises a user interface 103, which is illustrated as a screen but may include any other means of conveying information to a user such as e.g. through audible or visual signals. The computing device 1 is communicably connected, such as e.g. through a network, to sequence data acquisition means 3, such as a sequencing machine, and / or to one or more databases 2 storing sequence data. The one or more databases 2 may further store one or more of: mutational signatures information, training data, parameters (such as e.g. parameters of a machine learning model used to predict whether a tumour is PRR-deficient, e.g. weights of a logistic regression model, architecture and parameters of a decision tree model, etc.), clinical and / or sample related information, etc. The computing device may be a smartphone, tablet, personal computer or other computing device. The computing device is configured to implement a method for characterising a DNA sample, as described herein. In alternative embodiments, the computing device 1 is configured to communicate with a remote computing device (not shown), which is itself configured to implement a method of characterising a sample, as described herein. In such cases, the remote computing device may also be configured to send the result of the method of characterising a DNA sample to the computing device. Communication between the computing device 1 and the remote computing device may be through a wired or wireless connection, and may occur over a local or public network 6 such as e.g. over the public internet. The sequence data acquisition means may be in wired connection with the computing device 1 , or may be able to communicate through a wireless connection, such as e.g. through WiFi and / or over the public internet, as illustrated. The connection between the computing device 1 and the sequence data acquisition means 3 may be direct or indirect (such as e.g. through a remote computer). The sequence data acquisition means 3 are configured to acquire sequence data from nucleic acid samples, for example genomic DNA samples extracted from cells and / or tissue samples. In some embodiments, the sample may have been subject to one or more preprocessing steps such as DNA purification, fragmentation, library preparation, target sequence capture (such as e.g. exon capture and / or panel sequence capture). Preferably, the sample has not been subject to amplification, or when it has been subject to amplification this was done in the presence of amplification bias controlling means such as e.g. using unique molecular identifiers. Any sample preparation process that is suitable for use in the determination of a genomic alteration profile (whether whole genome or sequence specific) may be used within the context of the present invention. The sequence data acquisition means is preferably a next generation sequencer.

[0179] The following is presented by way of example and is not to be construed as a limitation to the scope of the claims.

[0180] EXAMPLES

[0181] Small insertions and deletions (indels, <100 bp) represent the second most prevalent form of genetic variation after substitutions. Indel mutagenesis is substantial and non-random, reflecting underlying mutational processes, be they benign by-products of normal physiology or malign consequences of exogenous exposures and / or endogenous dysfunction.

[0182] Despite its deleterious effects, small insertions and deletions (indels) have received far less attention than substitutions. Recent computational advances have surmounted previous technical challenges, enabling the study of mutational processes driving indel formation. In particular, recent advancements in indel detection and annotation have led to the identification of 18 indel signatures (IDs) in human cancers (Alexandrov et al., 2020). These signatures were defined using an 83- channel classification system (i.e., 83 indel subtypes, herein referred to as the COSMIC-83 format), founded on characteristics such as the length of flanking mono / polynucleotide repeat or sequence homology at the indel junction. Subsequent re-analysis of the same dataset using a revised algorithm reported nine additional de novo IDs (Jin et al. 2022).

[0183] Accurate characterization of indel patterns is crucial for biological and clinical purposes. For instance, high proportion of microhomology-mediated deletions is a key predictor of clinically actionable homologous recombination deficiency (HRd), carrying the greatest weight in HRd detection and classification algorithms (Davies et al. 2017, Nguyen et al. 2020). Additionally, the diagnosis of mismatch repair deficiency (MMRd) relies on measuring genome-wide indel mutagenesis and / or analyzing panels of mono / dinucleotide repeats (Thibodeau et al. 1993, Aaltonen et al. 1993, Boland er al. 1998). Recent studies have also highlighted how 2-5bp short deletions and 2-4bp insertions are distinct readouts of TOP1 activity in RNaseH2-null cells and TOP2A dysfunction, respectively.

[0184] Motivated by the mounting importance of indel mutagenesis in accurate tumour classification and vulnerability prediction (Chung et al. 2021), the inventors established a ‘ground truth’ set of mutational signatures. They focused on ‘Post-Replicative Repair deficiency’ (PRRd), encompassing defects in MMR and replicative polymerase proofreading - biological abnormalities that often display exquisite sensitivity to immune checkpoint inhibition (ICI). In particular, the inventors generated isogenic human CRISPR-edited cellular models of post- replicative repair dysfunction (PRRd), including individual and combined gene-edits of mismatch repair (MMR) and replicative polymerases (POLE and POLD1). Diverse, unique mutational footprints of MMR deficiency and polymerase mutants were revealed.

[0185] However, they identified inherent constraints in the prevailing indel classification schema that hamper its discriminatory power. In particular, the prevailing indel signature classification falls short in discriminating these signatures from background mutagenesis and between each other, as it condenses important biological signals into two indel subclasses (channels), which limits signature analyses. To address this, the inventors devised a novel classification system that includes considering the 5’ and 3’ flanking nucleobases (i.e. considering the 5’ and 3’ sequences flanking indels and including informative motifs (e.g. longer homopolymeric tracts) while removing uninformative channels), resulting in 89 indel channels. They showed that incorporating flanking nucleobases information for indel classification provides paths to uncovering new aetiologies of indel mutagenesis, offering mechanistic insights and clinical added value.

[0186] In particular, through analysis of 4,775 whole cancer genomes from the 100,000 Genomes Project, the inventors uncovered 37 indel signatures (InDs), 27 of which are new. Further, the inventors developed a classifier, PRRDetect, which outperforms other approved biomarkers for immunotherapy, such as Tumour Mutational Burden (TMB).

[0187] Example 1 - Diversity of indel patterns in PRRd

[0188] In this example, the inventors used isogenic (knock-out and knock-in) cell lines to study the diversity of indel patterns in PRR deficiency. They further used this analysis to explore the strengths and limitation of the currently used indel signature classification (COSMIC). They show that the current approach fails to identify known phenotypes of these samples, and identify particular channels of the current classification that are not specific enough to capture these differences. Results

[0189] The inventors generated a ‘ground truth’ set of isogenic cellular models, with CRISPR edits in genes that produce indel mutagenesis when dysfunctional. They used a hTERT RPE-1 (TP53- / “) cell line, creating knockouts (KO) of DNA mismatch-repair-associated genes (AMSH2, AMLH1, AMSH3, ASETD2, AARID1A, AEX01), two knock-in (KI) pathogenic missense mutants of replicative DNA polymerase POLE (p.P286R, p.L424V), two KI pathogenic missense mutants of POLD1 (exonuclease mutant p.S478N, and polymerase mutant p.R689W), and three mutants with combined polymerase and MMRd (POLD1s478N / M / WLH7, POLD1s478N / +A / WSH2 and POLEP286RAMSH2). AS illustrated on Figure 6A, successfully edited clones were grown for ~45- 50 days to allow for mutation accumulation, 2-5 daughter subclones were picked per genotype and expanded for whole-genome sequencing (WGS). Somatically acquired indels that had accumulated during culture were identified. Indel catalogues were constructed using COSMIC’s 83-channel method as described in Alexandrov et al., 2020. The indel pattern associated with each genotype is shown on Figure 6F and was obtained by subtracting background indel mutagenesis as described in Zou et al., 2021 .

[0190] Elevated indel burdens were observed in ten out of 1 1 gene-edits when compared to an unedited control clone (Fig. 6b). The mutation burden was approximately two-fold higher in AMSH3 and POLD1R689W, ten-fold in POLD1S478N, POLEL424Vand POLEP286R, 50-fold in AMSH2, AMLH1 , and particularly significant in combined gene-edits: around 200-fold in POLD1S478N / +AMLH1 and POLEP286RAMSH2, and 300-fold in POLD1S478N / +AMSH2. All lines except ASETD2 showed variations in their COSMIC-83 indel profiles compared to control (Fig. 6K). The inventors noted discriminative characteristics between and within gene-edited clones (Fig. 6F, C-D, L-O). Exclusive 1 bp T deletions at homopolymers of 6bp or more (poly-T6+) were observed for AMSH3, whilst POLD1S478Nand POLEP286Rshowed exclusive 1 bp T insertions at poly-T6+. POLD1R689W, POLEL424Vand all three combined polymerase / MMRd edits predominantly exhibited 1 bp T insertions (but not exclusively), with variations of 1 bp T deletions among different genotypes. Together, these experiments revealed diverse indel signatures among different PRRd mutants. Remarkably, mutations within the same gene but affecting different protein domains manifested signature variations (e.g., POLD1 p.S478N vs p.R689W).

[0191] The inventors also examined the substitution profiles all edited lines (Fig. 6E, 9A-G). Interestingly, MMRd lines showed lower substitution-to-indel ratios compared to control, while polymerase- dysregulated lines exhibited markedly increased ratios. This suggests that genome instability is primarily driven by an excess of indel mutagenesis in MMRd, whereas substitution mutagenesis plays a more significant role in polymerase dysregulation. Furthermore, mutational asymmetry analyses revealed enrichment of both substitutions and indels on the leading strand for POLE mutants while P0LD1 mutants exhibited lagging strand bias, specifically T insertions at homopolymeric runs of 5-7 nts. This is in keeping with the hypothesized preferential activity of Pol E and Pol 5 in leading and lagging strand synthesis respectively, suggesting that POLE / POLD1 mutants tend to accumulate 1 bp A insertions on the nascent strand while replicating through 5-7 nts poly-T-tracts. This lends support to the proposition that polymerase E and 5 are more proficient at detecting incorrectly paired bases at template adenines. Interestingly, indel signatures of combined MMRd / polymerase mutants were not simply the sum of the individual mutational processes. There was complexity in the mutational profiles and a marked strand bias effect on many more channels, likely reflecting the biological interactions of Pol E and Pol 5 with MMR in suppressing indel formation during replication of repetitive DNA.

[0192] Nevertheless, while the diversity of experimental indel profiles was appreciable among PPRd genotypes, it was difficult to disambiguate gene-edit signatures from background mutagenesis. Clustering analyses and direct comparisons of gene-edit profiles to control showed extremely high similarity (cosine similarity > 0.9, Fig. 6C, 6L, 6N). Discrimination between polymerase- dysregulation and MMRd gene-edit signatures was also limited (Fig. 6D). Clustering by cosine similarity revealed mainly two groups of signatures: deletion-driven MMRd signatures and insertion-driven polymerase mutant signatures (Fig. 60). The inventors thus questioned the sufficiency of the COSMIC-83 classification schema, given that all the variation between the ten gene-edits was effectively concentrated in two channels: T insertions and T deletions at poly-T6+.

[0193] Experimental indel catalogues were compared to known COSMIC indel signatures (v3.3) (Alexandrov, 2020) (shown on Fig. 6G). The results of this are shown on Fig. 6J. Surprisingly, the inventors found that the indel catalogues of DMSH2, DMSH6 and DMLH1 cells did not show any similarity to COSMIC ID7 previously reported to be due to MMR deficiency (MMRd). Instead, DMSH2, DMSH6 and DMLH1 indel patterns showed greatest similarity to COSMIC indel signatures ID1 and ID2, ascribed to normal replication-associated nascent and template strand slippage respectively (Fig. 6F, G, H). The COSMIC-83 taxonomy collapses all indels at homopolymers > 5bp into a single channel (i.e., T6+, for deletions and insertions separately) (Fig. 6G). It is known that the probability of indel formation increases with increasing length of polynucleotide repeat tract (Lujan et al., 2015) in MMRd, a characteristic hallmark known as micro sate I lite instability (MSI) (Strand, 1993). The inventors postulated that MSI cannot be distinguished from the signature of background replication errors (ID1 and ID2) using the current COSMIC taxonomy because COSMIC’s indel classification aggregates indels occurring at repetitive tract lengths of greater than five into a single channel called “+6”, i.e. all discriminatory signals were condensed into ‘deletion at T6+’ or ‘insertion at T6+’ channels, which reduced the separative capacity during signature extractions (Fig. 6F-G, where Fig. 6G shows COSMIC signatures ID1 , ID2 and ID7, respectively believed to be thought to be associated with N-slip, T- slip and MMRd). Hence, MMRd signatures could not be distinguished from signatures of normal replication errors. This contrasts with corresponding PRRd SBS signatures, which capture the diversity of substitution patterns amongst MMRd and / or polymerase-dysregulated cancers (Fig. 6E, 9B-G). Moreover, COSMIC ID7 lacks signal within reputedly the most informative poly-T6+ channel. Instead, signal in this channel is only present in ID2 (Fig. 6G), resulting in systematic misattribution of all MMRd gene-edit signatures to ID2 (Fig. 6J, 6H). Additionally, indel signatures of POLE and POLD1 mutants and all combined polymerase / MMRd edits were indistinguishable from ID1 using the COSMIC-83 taxonomy (Fig. 6J, 6H) and sometimes indistinguishable from each other (Fig. 6D). poLD1R689Wsignature did not resemble any reported signatures. Since a vast proportion of mutagenesis in gene-edited lines was erroneously assigned to ID1 and / or ID2 (Fig. 6H), the inventors explored whether modifying the information presented in individual indel channels could improve the resolution to disentangle ostensibly alike but distinct biological signatures without compromising power for signature extraction.

[0194] Methods

[0195] Cell Lines. A summary of all human cell line models used in this study is provided in Table 4. All cells were grown in DMEM / F12 medium (Gibco / Thermo Fisher Scientific) supplemented with 10% FBS, at 37°C and 5% CO2 in a humidified incubator. hTERT RPE-1 ATP53 cell (a gift from M. Tarsounas, Department of Oncology, University of Oxford, UK) was generated from a previous study (Zimmermann, 2018). To generate the remaining isogenic CRISPR-edited cell lines, 200k RPE-1 ATP53 cells per each edit were electroporated with pre-formed ribonucleoprotein complex (RNP with final concentrations of 120 pmol gRNAs and 100 pmol Alt-R Cas9) in supplemented SE buffer, using nucleocuvette and AMAXA 4D-Nucleofector (Lonza) on program EH-158 according to manufacturer’s instructions. Following electroporation, cells were re-plated into fully supplemented DMEM / F12 medium to recover for 48 h. For knock-ins, homology-directed repair (HDR) donor oligos were supplied with the RNP for electroporation, and cells were re-plated for recovery in medium spiked in with final concentrations of 2pM M3814 (Selleckchem) and 0.5pM Alt-R HDR Enhancer (iDT) for 24 h. All cells were then cultured for additional 2-4 days to allow for gene editing and eventually subjected to limiting dilution on 96-w plates to isolate single-cell clones. Targeted clones were screened for successful frameshift edits for knockouts or incorporation of intended missense mutations for knock-ins via Sanger sequencing. All gRNAs and sequencing primers used are summarised in Table 5.

[0196] Table 4. Cell line models used in these examples.

[0197] Table 4b. Mutation counts in cell line model samples used in these examples.

[0198] Table 5. Guide RNAs, HDR donors, and sequencing primers used for generating and genotyping the isogenic CRISPR edits in RPE-1 cells. Bold underlined letters in the HDR donors specify the intended missense nucleotide changes to the reference alleles; Italic underlined letters indicate the silent CRISPR / Cas9-blocking mutations incorporated to improve HDR. HDR, homology- directed repair; g, guide RNA; F, forward primer; R, reverse primer. Mutation accumulation experiment and sequencing. Although double mutants were more challenging and slower to establish from single cells than single mutants, once established, their doubling times were comparable to those of the single mutants. All edited clones were expanded and maintained in culture for over 45 -55 days (approximately 40-50 cell doublings) to allow for mutation accumulation in the edited genetic backgrounds. Then, a second round of single cell limiting dilution was performed to isolate 2-4 daughter subclones per edit, providing bottleneck to capture mutations that had occurred since the isolation of the initial edited parental clones. Genomic DNA was isolated from daughter subclones using Qu / c / c-DNA Miniprep Plus Kit (Zymo Research) following manufacturer’s protocol. WGS libraries were prepared and sequenced with a paired-end 150 bp configuration on Illumina NovaSeq 6000 platform by Novogene (Beijing, China), aiming for an average genome-wide sequencing depth of 25x per sample.

[0199] Somatic variant calling. Short reads were aligned to human reference genome GRCh38 / hg38 using BWA-MEM. Substitutions, indels and rearrangements in genetic reference clones and daughter subclones were called as previously described (Zou et al., 2021). Post-processing filters were applied to improve the specificity of mutation-calling. Specifically, forSNV calls by CaVEMan, we used CLPM == 0 and ASMD >= 140. To reduce false positive calls by Pindel, we used QUAL >= 250 and REP < 10. For rearrangements, the default assembly score was used. Variant allele fraction (VAF) distribution for each subclone was examined, and those with an average VAF of < 0.4 were designated as polyclonal and subsequently excluded from all quantitative analyses (i.e. , estimation of mutation density and mutation burden). A filter for variant allele frequency (> 0.2) was applied to substitutions and indels. De novo substitutions and indels in subclones were obtained by subtracting from respective parental clone whenever available, or by removing mutations shared among subclones. De novo mutation counts are provided in Table 4b. Rearrangements were too few to be informative.

[0200] Mutational signature analysis of experimental samples. The principle of determining gene edit- associated mutational signatures with bootstrapping and background subtraction method (Zou et al, 2021) follows the steps: (i) determine the background mutational profile in wildtype control; (ii) determine the difference between the mutational profile of the edited clone and background mutation profiles; and (iii) remove background mutation profile from the mutation profile of the edited subclone. The inventors also used Uniform Manifold Approximation and Projection (UMAP) (Mclnnes & Healy, 2018), a manifold learning technique for dimensionality reduction to cluster the mutation profiles. In UMAP two-dimensional representation, mutational profiles similar to each other are plotted near each other, whereas profiles that are dissimilar are plotted apart. Example 2 - A new indel classification framework

[0201] In this example, the inventors propose a new indel classification framework, and demonstrate that it remedies at least some of the problems of the existing indel classification framework.

[0202] Results

[0203] Indel patterns have a propensity to occur at specific sequence motifs (Tanay et al., 2008). The inventors thus hypothesized that incorporation of surrounding sequence context could improve biological discriminatory value of indel signature analyses.

[0204] A new classification framework was devised, illustrated on Figures 4A and 4B and detailed in Table 6. The inventors classified all indels according to whether they were deletions, insertions or complex indels (a combination of insertion and deletion). Within insertion and deletion categories, the indels were sub-classified by motif size (i.e. the size of the indel, 1 bp vs >=2bp). For 1 bp indels, the inventors considered the nucleotide content (CIG motifs from A / T motifs, i.e. deletion of a C / G or A / T), and the 5’ and 3’ nucleotides (flanking bases, i.e. the 5’ nucleotide immediately adjacent to the indel motif, and the 3’ nucleotide immediately adjacent to the indel or repeated sequence, as well as the minimal indel prefix that is maximally repetitive in the 3’ sequence context and within the indel itself, see Methods). For longer indels (>=2bp), the classification framework further takes the polynucleotide repeat motif length and repeatedness of the indel into account. For deletions occurring with microhomology at the indel junction, the classification framework takes the length of the indel motif (L) and the microhomology length (M, number of bases involved in the microhomologous sequence) into consideration.

[0205] This comprehensive taxonomy resulted in 476 non-overlapping indel sub-categories (channels) (see Table 6). The inventors examined whether all 476 channels were informative. By investigating counts of indels (indel frequency distributions across all channels) across 18,522 tumours covering most cancer types from the ICGC / TCGA, Hartwig, and the GEL 100,000 Genomes project, the inventors were able to discard non-informative channels (i.e. removing channels were the sum of indels show no indels across all samples and tumour types (channels with no signal), and consolidating those that show few indels or are redundant (channels with low signal)), reducing the total number of channels to 89 (see Tables 2 and 6). This is illustrated on Figures 5A-C. Although the total number of channels is not vastly different from the COSMIC 83 channels, the encoded information is quite different because of the incorporation of sequence context, and the division of polynucleotide repeat lengths into additional categories of longer repetitive sequences. The incorporation of sequence context and stratification of mono / polynucleotide repeat tracts into additional categories present alternative information to the mutational signature extraction and assignment process, increasing the likelihood of detecting new biologically meaningful signatures. Overall, compared to the COSMIC-83, the 89-channel taxonomy expands upon 1 bp T indels (where most of the signals were) into a larger array of channels, and condenses longer indels and / or genome motifs that are infrequent in he genome (where signals were scant or non-existent) into fewer indel subcategories. Although the final numbers are not vastly different between the two classification systems, the inventor’s data-driven approach to incorporate sequence context information and to stratify mono / polynucleotide repeat tracts into additional categories to enhance signal per represented channel provides alternative information to the mutational signatures and assignment process, increasing the likelihood of detecting new biologically meaningful signatures.

[0206] To investigate the discriminatory power of the new classification method, the inventors applied it to the ground truth experimental dataset used in Example 1. From the 89-channel profiles alone, the inventors could already see distinguishing features between genotypes that were not appreciable using the 83-channel classification (see e.g. Fig. 7A-C). Cosine similarities between experimental catalogues and the control were much lower with the 89-channel format than with the COSMIC-83 format (Fig. 7B, 7H), indicating that the new classification schema improved separation of gene-edits from the background (mean cosine similarity, 0.68 ± 0.08 for 89-channel vs 0.89 ± 0.11 for COSMIC-83, P=1 ,917e-07). Next, the channels were used to determine signatures associated with each gene-edit with two methods (sigProfiler, Islam et al., 2022, and mvNMF, Yu et al., 2018 as implemented in MuSiCal, Jin et al., 2022). The resultant signatures were better distributed across the entire profile, exhibiting signal in multiple channels (Fig. 6F, 7D, 7J). Gene-edit signatures were also more readily discernible among themselves (mean signature pairwise cosine similarity, 0.57±0.25 for 89-channel vs 0.64 ± 0.3 for COSMIC-83, P = 1 ,483e-05, Fig. 7G, 7L). This suggests that the 89-channel format was more effective in capturing the biological variation in indel mutagenesis amongst PRRd mutants.

[0207] Interestingly, the inventors noted that while MMRd deletions are particularly amplified at longer homopolymer tracts (8-9bp > 5-7bp > 0-4bp), polymerase mutants displayed strikingly different distribution of excess insertion mutagenesis at shorter homopolymers (5-7bp > 8-9bp > 0-4bp). The higher indel rates in shorter homopolymers conferred by defective proofreading of Pol E and Pol 5 likely reflects the distance over which they interact with duplex DNA upstream of the polymerase active site. Indeed, crystal structures of Pol E and Pol 5 have shown numerous contacts made within 5-7bp of the polymerase active sites with duplex DNA, with experimental model reinforcing this optimal distance, explaining how proofreading may offer reduced protection against indels outside of this “footprint” (i.e., unpaired bases further upstream of the active site; longer runs where MMR plays a more crucial role). These unique insights were only appreciable owing to the effectiveness and enhanced resolution of our new 89-channel format at capturing the biological variation.

[0208] To compare the discriminatory capacity of both classification systems, the inventors also performed de novo signature extraction on the ground truth experimental samples of Example 1 (n=37, Fig.lOA). Using the COSMIC-83 system, only two signatures were extracted - one dominated by T insertions at poly-T6+ (ID83A) and the other dominated by T deletions at poly-T6+ (ID83B) (Fig. 10B). In contrast, the 89-channel format yielded four signatures, matching our expectation of a predominantly deletion-driven MMRd-signature (CH89B), a predominantly insertion-driven polymerase signature (CH89D), and two distinct signatures with varying levels of insertion / deletion combinations (CH89A and CH89C) (Fig. 10C).

[0209] Finally, to determine whether this observed relationship between channel information content and signature extraction extended to other datasets and workflows, the inventors applied three different algorithms (Degasperi et al. 2022, Jin et al. 2022, Islam et al. 2022) to an unrelated cohort of 52 colorectal WGS from ICGC (International Cancer Genome Consortium, Zhang et al. 2019, dcc.icgc.org / ) (Fig. 10D). All three algorithms failed to discern all available signatures using the COSMIC-83 format, reaching a discrimination limit of five, yielding sparse signatures with signal density highly concentrated in two channels (Fig. 10d, e, g). Contrarily, the 89-channel format consistently enabled the detection of additional signatures across all algorithms used (Fig.lOf, h). The extracted signatures also displayed signal across more channels, highlighting the superior performance of the 89-channel classification over COSMIC-83 in uncovering additional, true mutational processes.

[0210] 

[0211]

[0212]

[0213] classification (see Table 2).

[0214] Below are examples of descriptions of some of the classes in the full indel classification: A[lns(C):R0]A = Insertion of a single C, no repeats (R=0, U=1), 5’=A, 3’=A; A[lns(C):R0]G = Insertion of a single C, no repeats (R=0, U=1), 5’=A, 3’=G; A[lns(C):R0]T = Insertion of a single C, no repeats (R=0, U=1), 5’=A, 3’=G; A[lns(C):R1]A = Insertion of a single C, 1 repeat (R=1 , U=1 ), 5’=A, 3’=A, i.e. A|C|CA; A[lns(C):R1]G = A|C|CG; A[lns(C):R1]T = A|C|CT; A[lns(C):R2]A = Insertion of a single C, 2 repeat (R=2, U=1), 5’=A, 3’=A, i.e. A|C|CCA; lns(2,4):M = Insertion of 2 or more base pairs, any number of repeats, with microhomology; lns(5,):M = Insertion of 5 or more base pairs, any number of repeats, with microhomology; lns2:U1 :R0 = Insertion of 2 bp, U=1 , R=0; lns3:U1 :R0 = Insertion of 3 bp, U=1 , R=0; lns4:U1 :R0 = Insertion of 4 bp, U=1 , R=0; lns2:U2:R0 = Insertion of 2 bp, U=2, R=0; lns(5,):R0 = Insertion of 5 or more bp, R=0; lns2:U1 :R1 = Insertion of 2 bp, U=1 , R=1 ; I ns2 : U 1 :R(5,9) = Insertion of 2 bp, U=1 , repeats=5, 6, 7, 8, or 9 (5+); lns(5,):U2:R1 = Insertion 5bp, unit size=2, repeat=1 ; A[Del(C):R1]A = Deletion of a single C, U=1 , R=1 ;

[0215] Del2:U1 :R1 = Deletion of 2bp, U=1 , R=1 ; Del3:U1 :R1 = Deletion of 3bp, U=1 , R=1 (1 repeat prior to deletion); Del2:U2:R2 = Deletion 2bp, unit size 2 bp, 2 repeats prior to deletion; Del3:U3:R(5,9) = Deletion 3bp, motif size 3bp, 5 or more repeats prior to deletion; Del2:M1 = Deletion 2bp, microhomology length 1 (1 base pair is repeated at the start of the indel and the start of the 3’ context); Del3:M2 = Deletion 3bp, microhomology length 2 bp; Del(7,):M1 = Deletion 7 or more bp, microhomology length 1 ; Complex(0,1) = Complex indel, length 0 or 1 ; Complex(2,5) = Complex indel, length 2 to 5. Note that the reasoning applied to interpret the notation for categories that are associated with a description applies also to categories not explicitly described for brevity.

[0216] In Table 6, the following notations apply:

[0217] A[lns(C):R0]A refers to: o A[: 5’ nucleotide o lns(C):R0: Insertion of a single C, 0 repeats prior to insertion o ]A: 3’ nucleotide; lns2:U1 :R0 refers to: o Ins2: Insertion of 2bp o U1 : Insertion of a motif of 1 bp, o RO: 0 repeats prior to insertion lns2:U1 :R(5,9) refers to: o Ins2: Insertion of 2bp o U1 : Insertion of a motif of 1 bp, o R(5,9): 5 to 9 repeats prior to insertion

[0218] Del3:M1 refers to: o Deletion of 3bp o Microhomology length 1

[0219] Del(2,4):R1 refers to: o Del(2,4): Deletion of 2 to 4 bp, o R1 : 1 repeat prior to deletion

[0220] - Del(2 ,8): U(1 ,2):R(2,4) refers to: o Del(2, 8): Deletion of 2 to 8 bp o R(2,4): 2 to 4 repeats prior to deletion o U(1 ,2): repeated unit of 1 or 2 bp lns(C):R(0,3) refers to: o Ins(C): insertion of a single C o R(0,3): 0 to 3 repeats prior to insertion o Any sequence context

[0221] [Del(C):R1]A refers to: o [: any 5’ sequence o Del(C): deletion of a single C o R1 : 1 repeat prior to deletion o ]A: 3’ A

[0222] The above are provided as examples but the same reasoning apply to interpret all of the notations in Table 6. Methods

[0223] See Example 1 .

[0224] De novo extraction of indel signatures from hTERT-RPE-1 samples. De novo extraction and decomposition of mutational signatures was performed in Python v.3.8.5 using SigProfilerExtractor (v.1.1.18) (Islam, 2022), along with SigprofilerMatrixGenerator (v.1.2.4) (Bergstrom et al, 2019). The recommended default settings (including 500 NMF replicates) were applied (github.com / AlexandrovLab / SigProfilerExtractor). De Novo extraction was also performed using MuSiCal (Jin et al., 2022) - which is based on mvNMF, with default settings (20 bootstraps, 500 repeates per bootstrap, matched clustering). The number of indel signatures selected per channel set was determined by maximal drop in average silhouette width of clustered signatures. For extraction using MuSiCal, the default hyperparameters included random initialization (init hyperparameter), the multi-view non-negative matrix factorization (MVNMF) algorithm (method hyperparameter), 20 MVNMF replicate runs (n_replicates hyperparameter), and between 10,000 and 1 ,000,000 MVNMF iterations per replicate run (minjter and maxjter hyperparameters, respectively).

[0225] Genomics England (GEL) cohort. WGS data for colorectal (n=2,146), endometrial (n=695), stomach (n=181), bladder (n=347), brain (CNS, n=392), lung (n=958) and skin cancers (n=56) were considered in this study, subset from the curated 12,222 sample cohort previously described (Degasperi, 2022). Indels were called with Strelka v.2.4.7 using somatic calling mode. The GEL version 8 dataset can be accessed via www.genomicsengland.co.uk.

[0226] Indel segmentation. The inventors exploited the relationship between indels and their associated 3’ sequence context to determine, for each indel, the minimal indel prefix that is maximally repetitive in the 3’ sequence context and within the indel itself. In brief, they list each prefix in the indel and calculate the maximally repetitive substring in the associated suffix (rest of the indel sequence) and 3’ context (3’ adjacent sequence). They term these partitions of the indel and context “segmentations”, and select the segmentation that contains the smallest prefix, with maximum repetition in the 3’ sequence context and the indel. Segmentation produces multiple values quantifying the repetitiveness of the prefix sequence (termed, “unit”), and these values, along with the sequence context, may be used to group indels into biologically relevant channels. Note that the COSMIC classification (83 channels) also considered motif repeatedness but does not consider the sequence context outside of the repeat, and quantifies repeats including what is termed here as “prefix” (i.e. R0 here is R1 in the 83-channel classification). The segmentation process will now be described in more detail.

[0227] Mechanisms of indel formation often involve insertion or deletion of nucleotide sequence located 3’ to the event. Previous approaches to indel classification have used this property to stratify indels into mechanistically-associated groups by determining the repetitiveness of the inserted I deleted sequence with respect to the 3’ sequence context (see e.g. Alexandrov et al., 2020). However, the inventors have observed that significant ambiguity exists in binary threshold-based classification schemes (i.e “Repeat-mediated” vs. “Microhomology-mediated”). Specifically, the inventors found that restricting indel classification to predefined and mechanistically associated categories can obscure and collapse mutational signatures produced through engineered genetic perturbations.

[0228] Previous approaches quantifying indel repetitiveness have used different measures of repetition dependent on the mechanistically-associated class. The COSMIC classification classifies insertions and deletions separately (see Fig. 3A, 3B), using C / T classes for 1 bp insertions (which are then separated by homopolymer length classes), insertion length classes for >1 bp insertions (which are then separated by the number of repeated units), C / T classes for 1 bp deletions (which are then separated by homopolymer length classes), deletion length classes for >1 bp deletions without microhomology (which are then separated by the number of repeated units), and deletion length classes for >1 bp deletions with microhomology (which are then separated by the microhomology length). Further, the COSMIC approach applies a scheme that simply separates subclasses within each of these classes based on a pre-established scheme where every homopolymer length I number of repeat units I insertion length I deletion length I microhomology length is its own class up to a threshold number (all instances above the threshold being grouped together). This fails to appreciate that there are in fact distributions of these values that are not monotonous in real cancer genomes, such that classes that cover subranges of these parameters naturally emerge. The inventors instead sought a unified strategy for determining the relationship between an indel and the 3’ context. Here, they demonstrate this strategy, which they term “indel segmentation”, which explicitly identifies the smallest prefix sequence motif that displays a maximal repetitive relationship with the 3' sequence context and within the indel itself. This unified strategy enables to characterise in detail all indel types within a common scheme, leading to granular classes of indels that have unique combinations of the segmentation parameters. These granular classes were then combined in a data driven way, maintaining highly informative classes and combining less informative ones. Thus, the proposed approach provides a data driven scheme that aims to best capture signal in a dataset, where for some categories of indels a more granular classification is informative whereas other classes are adequately captured with fewer classes. Thus, compared to the COSMIC approach, although the present approach does not add many channels in terms of absolute number of channels, it better captures information by allocating more channels to classes of indels that benefit from this additional granularity, and fewer channels to classes of indels that are adequately captured with fewer indels. The direct result of this is an improvement in the identification of indel signatures in cancer genomes, and hence in the analysis of cancer genomes for purposes such as prognosis, treatment identification and treatment response prediction, since patterns that were obscured by inadequate classification can now be uncovered.

[0229] Pre-processing indels. In this work, the inventors followed the standard convention of normalizing indels through left-alignment (VCF standard). Indel calls being “left-aligned” means that for example when there is a deletion in what is a polynucleotide tract, (e.g. a T deletion in a TTTTT), it is difficult to know which T is actually deleted from that string of Ts (any of the Ts in the tract could have been deleted leading to the same resulting sequence). All indel callers by default follow the convention to left align meaning that they will assume that the deleted T is the first one of the tract (the same principle applies to any indels, the use of a poly-T tract is provided for illustration purposes). Because the proposed method depends on comparing indel sequences, it is preferred that indels be called with high confidence (VAF > 0.2). The approach described here focuses specifically on insertions, deletions and complex events with lengths < 10Obp, and a 3’ sequence context of up to length 200bp. The segmentation approach generalizes to indels of all lengths, including 1 bp indels where the prefix is unambiguous. The use of indels <100bp is a commonly followed approach in the field because it was historically difficult to validate longer indels with Sanger PCR sequencing. This limitation no longer applies and there is no reason why the present approach could not be applied with any other threshold. The commonly used threshold was used simply for ease of comparability with previous work. Similarly, where context of an indel is taken into account, this is typically limited to the 3’ sequence (sequence downstream of the indel), and 200bp is a reasonable and convenient length to consider. In the present work, the 5’ context is also taken into account for at least some classes of indels.

[0230] Indel segmentation process. The following procedure is applied to each indel variant:

[0231] 1. Starting from the 1st position of the indel, generate multiple ‘units’, each representing different prefixes of length k (k from 1 to the length of the indel) bases from the start of the indel.

[0232] E.g for the indel | 3’ sequence pair: ATCATCGA | ATCATCATCGGGGG The possible units are:

[0233] - A

[0234] - AT

[0235] - ATC

[0236] - ATCA

[0237] - ATCAT

[0238] - ATCATC

[0239] - ATCATCG

[0240] - ATCATCGA

[0241] 2. For each “unit”, determine how many perfect tandem repeats of this unit exist in the 3' sequence context.

[0242] 3. For each “unit”, determine the number of exact tandem repeats (i.e. immediately adjacent exact repetitions of the unit) of the unit within the indel, not including the first occurrence, starting from the end of the prefix sequence. 4. Term the remaining length of the indel the "spacer", which separates the repetitive region within the indel and the repetitive region in the 3' sequence context.

[0243] The above results in the following set of characterizing values for each segmentation:

[0244] Unit length (U): The length of the "unit" in base pairs (bp), and associated nucleotide sequence

[0245] Internal Repetition (IR): The number of times the unit sequence is tandem-repeated within the indel, and the nucleotide sequence of those repetitions

[0246] Spacer (S): The length in bp of the remaining sequence between the last tandem repetition of the unit within the indel, and the end of the indel, as well as the associated nucleotide sequence.

[0247] Three Prime Repetition (TR): The number of times the unit sequence is tandem-repeated in the 3' sequence context, and the associated nucleotide sequence.

[0248] For example, in the above example, we have for each of the possible units:

[0249] A: U=1 with sequence “A”, IR=0, S=7 with sequence “TCATCGA”, TR=1 with sequence “A”;

[0250] AT: U=2 with sequence “AT”, IR=0, S=6 with sequence “CATCGA”, TR=1 with sequence “AT”;

[0251] ATC: U=3 with sequence “ATC”, IR=1 with sequence “ATC”, S=2 with sequence “GA”, TR=3 with sequence “ATCATCATC”;

[0252] ATCA: U=4 with sequence “ATCA”, IR=0, S=4 with sequence “TCGA”, TR=1 with sequence “ATCA”;

[0253] ATCAT: U=5 with sequence “ATCAT”, IR=0, S=3 with sequence “CGA”, TR=1 with sequence “ATCAT”;

[0254] ATCATC: U=6 with sequence “ATCATC”, IR=0, S=2 with sequence “GA”, TR=1 with sequence “ATCATC”;

[0255] ATCATCG: U=7 with sequence “ATCATCG”, IR=0, S=1 with sequence “A”, TR=0;

[0256] - ATCATCGA: U=8 with sequence “ATCATCGA”, IR=0, S=0, TR=0.

[0257] For each indel, there are k possible segmentations corresponding to the k different prefixes beginning at the first position of the indel, each represented by the above four values (U, IR, S, TR). For example, in the above indel I 3’ context pair, there are 8 possible segmentations and therefore 8 possible units. Different segmentations capture different relationships between an indel and its associated 3’ context. To capture the relationship most relevant for indel classification, two criteria are used. These make two assumptions logically grounded in stratification of indels into mechanistically-associated categories (Alexandrov et al. 2020, Kucab et al. 2019, Zou et al. 2021). First, the scheme assumes that repetitive motifs drive and enhance indel mutagenesis. Second, in the absence of additional information, the scheme assumes the minimally sufficient representation of the unit is the most informative. The scheme selects the segmentation that is most concordant with the above criteria. This is equivalent to trying to optimize for the following properties, in the following order:

[0258] 1 . Maximize 3’ repeat coverage (TR*U). This favours a segmentation where the unit is highly repeated in the 3’ region, because this repetition is the most likely mechanistically relevant, following assumption 1 .

[0259] 2. Maximize internal repeat coverage (IR*U). As for 3’ repetition, maximizing internal repetition is consistent with assumption 1. 3. Minimize spacer length (S). Minimizing the number of bases between the repetitive region of the indel and the 3’ region in turn maximizes the length of the repetitive region, consistent with assumption 1 .

[0260] 4. Minimize unit length (U). The smallest repetitive unit that yields a segmentation that meets the above criteria is also the minimally sufficient unit, in accordance with assumption 2.

[0261] The top segmentation for each indel according to the criteria above is selected as the most optimal segmentation (most representative of the process of indel formation). To concisely describe the optimal segmentation - it is the smallest possible unit that has the highest 3’ repetition, internal repetition, and smallest spacer, out of all possible segmentations. Considering the example above: ATCATCGA | ATCATCATCGGGGG The segmentation is:

[0262] Unit I Repetitive motif (U): ATC, 3pb Internal rep (IR): ATC, 1 Rep Spacer (S): GA, 2bp 3’ Repetition (TR) ATCATCATC, 3 reps

[0263] When one of the values of IR, S orTR is zero, the sorting procedure then prioritises maximizing or minimizing the subsequent value, following the above list. This allows indels that may have no 3’ repetition to still be assigned a unique segmentation with a minimally sufficient unit and maximal repetition. In cases where there is no repetition at all, the priority of minimizing the length of the spacer results in maximizing the unit length by counting the entire indel as the unit.

[0264] Accounting for original context (Reps). For insertions or deletions of perfectly repetitive sequence with no spacer (S=0), the original number of times the unit is repeated in the DNA sequence before mutagenesis is biologically relevant. For deletions, this number of repetitions is equal to the number of 3’ prime repetitions (TR), plus the number of times the unit is repeated within the indel (IR) plus the first occurrence of the unit (1). Thus for deletions, we calculate an adjusted value which we denote as “Original Repetitions” or “R”: (Deletions) R = TR + IR + 1. For insertions, we do not need to adjust the number of repetitions, as TR accurately represents the number of times the unit was present in the DNA prior to mutagenesis occurring. Therefore for insertions: (Insertions) R = TR. Finally, for all indels where repeats are not perfect (i.e S > 0), the number of 3’ repetitions remains biologically relevant: (All others) R = TR. To summarize, a derived value R is obtained where: (i) for deletions with S=0: R = TR + IR + 1 ; (ii) for all other deletions: R = TR; and (iii) for insertions: R = TR.

[0265] Accounting for microhomology length (Mh-length). For indels where the associated segmentation indicates a spacer, we additionally calculate the length of the unit and internal repetition as the length of microhomology between the indel and the 3’ sequence context as: Mh = Mh = min( (1+IR) * U , TR * U) where we take the minimum length of the two regions as the minimally sufficient, and exact matching, region of microhomology (i.e. the microhomology length is the minimum of the internal coverage - internal repeats including the initial unit - and the 3’ coverage - repeats in the 3’ sequence).

[0266] Constructing indel channels using indel segmentations. With the segmentation procedure defined above and the associated values (R, Mh, U etc), “channels”, or, groups of indel variants sharing common properties, can be constructed. Here, we construct channels by defining boundaries on the segmentation values. Although multiple choices are possible for the particular boundaries applied for segmentation, the present inventors identified a particularly advantageous set of boundaries that results in sufficient numbers of mutations per class (at least when analysing human cancers as described herein) such that no single group is underpowered. For example, grouping repeats of 5 to 7 Ts in A[lns(T):R(5,7)]A and 8 or 9 Ts in A[lns(T):R(8,9)]A rather than e.g. group 1 to 3 Ts, 4 to 6 Ts, and 7 to 9 Ts, resulted in not single category with numbers of mutations so low as to be underpowered (while not creating groups that have so many mutations that they likely unduly aggregate signals that could be analysed separately with sufficient power. They show (see, Figure 7), that using channels constructed using segmentation can more effectively distinguish between mutational signatures associated with post-replicative repair dysfunction. A detailed description of each channel, including the specific boundaries on segmentation values, is presented in Table 2 above. This partitions indels into the following groups:

[0267] 1) 1 bp insertions and deletions

[0268] For all indels of 1 bp, variants were translated to their associated pyrimidine context (i.e., C or T) and their 5’ and 3’ sequence context, as well as the length of the associated homopolymer tract considered. Examining the frequency of insertions and deletions stratified by tract length, we found roughly three densities at 0 - 4 bp, 5 - 7 bp, and 8 - 9 bp or greater. We used these two facets for channel construction. a) T insertions / deletions

[0269] We considered all 5’ and 3’ single nucleotide combinations around a central homopolymer variant, stratified into 3 homopolymer length classes (0 / 1 - 4 bp, 5 - 7 bp, 8 - 9 bp). This produced 27 channels in total for insertions and for deletions. b) C insertions

[0270] We observed C insertions were relatively infrequent across datasets and thus sought to concentrate signal in meaningful channels. We therefore only considered sequence context for non-templated insertions (R = 0) at specific sequence contexts (A I A, and A I T) producing two channels. We then stratified the remaining C insertions by length (0 - 3 bp, 4 - 6 bp, and 7 - 9 bp) producing 3 channels. c) C deletions

[0271] In the datasets examined in this study, C deletions were more frequent than C insertions, but still necessitated channel collapsing to aggregate signal. We noted that mechanistically, the 3’ sequence context is particularly relevant for deletions. We therefore constructed 6 channels describing C deletions at 1 bp, 2 bp and 3 bp, or 4 - 5 bp long homopolymers adjacent to A or T nucleotides in the 3’ context, producing 8 channels. We additionally considered shorter homopolymers (1 - 5 bp) at C / G sequence contexts in a single channel. Finally, we aggregated C insertions at longer homopolymers into a singular channel. In total, this produced 10 channels.

[0272] 2) > 1 bp insertions

[0273] For > 1 bp insertions, which are relatively infrequent across the datasets considered in this study, we stratified variants by repetitive content - R=0 (non-tem plated), R=1 (duplication), 2 < R < 4 (moderately repetitive) and R > 5 (highly repetitive). Where additional stratification was possible without compromising signal to noise, we partitioned indels by size, 2 - 4 bp (short) and > 5 bp (long). In total, this produced 6 channels.

[0274] 3) > 1 bp deletions without microhomology

[0275] For > 1 bp deletions without a spacer, and therefore consisting of exact repeats, we similarly stratified variants by repetitive content. For deletions where only one copy of the relevant variant existed, we stratified by length (2 - 4 bp, > 5 bp), producing two channels. For deletions with multiple copies of the relevant unit within the indel or in the 3’ sequence context, we considered the length of the unit, short (1 - 2 bp), or long (> 3 bp). For short unit deletions, we found we could partition variants further by degree of repetition (2 < R < 4 , 5 < R < 9), but could not similarly partition long unit deletions. In total, this also produced 6 channels.

[0276] 4) > 1 bp deletions with microhomology

[0277] For > 1 bp deletions where a spacer was identified during segmentation, the length of putative microhomology was calculated as above. We observed that variants with spacers tended to cluster broadly as short (2- 5 bp) and long (> 6 bp). We also observed that microhomology lengths were frequent for shorter lengths (< 3bp), and displayed a long tail (> 4 bp). We therefore considered two properties for constructing channels: 1) length of microhomology (1 bp, 2 bp, 3 bp, > 4 bp) and 2) the length of the variant. After collapsing channels with redundant information, this produced 7 channels in total.

[0278] In total, these 89 channels were used to represent the spectrum of indel variation observed in this study. We once again note that alternative partitioning procedures are possible and remain a topic for further study.

[0279] Indel channel construction. Using the segmentation values for each indel, the inventors constructed a set of 476 non-overlapping indel channels. By surveying the indel frequency distribution across all 476 channels in 17,253 tumour covering most cancer types from ICGC / TCGA, Hartwig, and the GEL 100,000 Genomes Project, the inventors discarded channels with no signal and consolidated those with low signal to reduce the total number of indel channels to a set of 89 channels that capture relevant biological variation. The final 89-channel set was used for the experimental gene-edit dataset and the cancer WGS from the GEL 100,000 Genome Project considered in this study. Generally, channels may be grouped into 6 broad divisions, 1 bp insertions, 1 bp deletions, >1 bp insertions, >1 bp deletions, > 1 bp deletions with evidence of microhomology. Channels were constructed such that each indel could 840 only be unambiguously assigned to one single channel. A complete description of each channel and exemplar indels are included in Table 2.

[0280] Example 3 - Indel signatures in seven cancer types

[0281] In this example, the inventors used the new indel classification framework described in Example 2 on a cohort of human cancers, and identified a new set of biologically meaningful consensus indel signatures.

[0282] Results

[0283] To explore the impact of the new indel taxonomy (89-channel indel classification system) on signature discovery in human cancers, beyond PRRd phenotypes, the inventors applied it to seven tumour types (n=4,775) known to have high tumour mutational burden (TMB) including due to clinically-relevant abnormalities such as MMRd, environmental exposures such as ultraviolet (UV) radiation and other causes such as APOBEC-related mutagenesis. This included genomes from colorectal cancers (n=2,146), endometrial cancers (n=695), stomach cancers (n=181), bladder cancers (n=347), brain cancers (CNS, n=392), lung cancers (n=958) and skin cancers (n=56) from the Genomics England 100,000 Genomes Project. A total of 4,775 samples were analysed (see Fig. 8A).

[0284] The inventors performed mutational signature analysis separately for each tumour-type as described in Degasperi et al. 2022, then derived from these a set of consensus indel signatures (herewith referred to as InD) using the 89-channel system. They identified 33 consensus indel signatures (InDs), illustrated on Figures 7B-D, 11 and shown in Table 9.

[0285] Ten of the identified signatures share characteristics with previously reported COSMIC ID signatures (Alexandrov et al, 2020): lnD1 , 2a, 3a / b, 4a, 6, 8, 9a, 13 and 18. The remaining 23 signatures were new. The inventors deduced putative aetiologies for nine new signatures, through experimental validation orthrough associations with genomic features such as substitution, double substitution patterns and extended sequence motifs.

[0286] Exogenous exposures underlie five InDs. lnD3a and 3b often co-occurred in lung cancers with tobacco exposure. lnD3a / b clustered with experimental signatures induced by benzo(a)pyrene (BaP) and its metabolite benzo(a)pyrene diol epoxide (BPDE) (Fig. 12, 13), supporting the notion that they represent modulated versions of tobacco-related DNA damage. lnD13, characterized by T deletions at TT dinucleotides and mostly seen in skin cancers, is linked to UV damage; and lnD18, found exclusively in colorectal samples, is secondary to colibactin exposure (lnD18 was found to correspond to COSMIC’s ID18, associated with colibactin exposure (Pleguezuelos- Manzano, 2020)). lnD32 was identified in samples with prior exposure to platinum and was associated with a new platinum- associated signature, SBS112. Twenty InDs have probable endogenous origins (Fig. 13). Several have been previously described, including lnD1 and lnD2a - errors associated with nascent (N-Slip) and template strand slippage (T-Slip) in replication, respectively (corresponding to COSMIC’s ID1 and ID2, respectively). lnD1 and lnD2a were seen universally across all tumour types except CNS and skin cancers, which showed a tissue-specific variant, lnD2b (Fig. 8a). lnD4a is attributed to TOP1 transcription-associated mutagenesis (Reijns, 2022). lnD6, marked by microhomology-mediated deletions, is associated with deficiency in homologous recombination (HR) repair (corresponding to COSMIC’s ID6). lnD8, which had deletions with little to no microhomology at deletion junctions, likely reflects the footprint of non- homologous end-joining activity and / or radiotherapy. lnD3a was found to correspond to ID3, previously attributed to tobacco exposure.)

[0287] Interestingly, we extracted eight gene-specific MMRd and polymerase-dysregulated InDs. lnD7 is associated with MMRd. MMRd-lnD7 contrasts with COSMIC ID7. lnD7 is characterized by the expected excess of 1 bp and 2bp deletions, particularly at longer mono / dinucleotide repeat tracts. lnD7 clustered with experimental signatures of z1 MLH1 , z1 MSH2 and z1 MSH6. We also identified lnD19 (MMRd due to PMS2 deficiency), lnD14 (associated with POLD1 exonuclease mutations), lnD15 (associated with POLE exonuclease mutations), lnD16a and 16b (resulting from concurrent loss of POLE proofreading and MMR), lnD21 (associated with combined POLD1 proofreading defect and MMRd), and lnD20, which we found through experimental investigations, was due to MMRd occurring on a POLE dysregulation background (i.e. POLE+MMRd, likely late onset MMRd on a background of POLE dysregulation). Thus, the new consensus signatures provide disambiguation of MMRd and polymerase-dysregulated signatures highlighting nine disparate signatures (instead of the three COSMIC indel signatures characterising dysregulation of post- replicative repair machinery, i.e. the signatures of normal replication slippage ID1 , ID2 and the MMRd signature ID7). lnD9a, correlated with SBS2 and SBS13 hypermutation, featured 1 bp C deletions at TCT and TCA motifs, identical to motifs seen in SBS2 / 13, particularly at short poly-T tracts. It is presumptively induced by APOBEC (Fig. 12c), which is corroborated by experimental evidence from an APOBEC over-expression DT40 model (DeWeerd, 2022). It is most prevalent in ovarian cancers. We propose a mutagenesis mechanism wherein following C to U deamination at TCT by APOBEC, UNG removal of uracil leaves an uninformative abasic site. Template strand slippage can then occur at this short repetitive T tract, leading to a C deletion (Fig. 12D). For reasons currently unclear, we also found similar C-deletion-dominated lnD9b / c, which though resembling lnD9a, lacked the predilection for a preceding T, and were possibly caused by alternative mechanisms.

[0288] The remaining 12 signatures are of uncertain aetiology. Five InDs are probably artefacts: lnD27 and lnD28 often co-occurred, incurring thousands of indels, and were related to SBS57, potentially an amplification or a sequencing artefact (correlated to SBS57 which has been described as an artefact of uncertain aetiology). lnD28m was possibly a mixed signature of lnD28 and lnD4, remaining to be resolved with larger cohorts. lnD5 and lnD10 were ubiquitous and possibly artefacts.

[0289] While C insertions dominated both lnD26 and lnD30 at poly-C tracts followed by a 3’A, lnD30 C insertions induced thousands of insertions at poly-C3-4, whereas lnD26 C insertions mainly occurred at longer poly-C5 (CCCCC) and did not induce hypermutation.

[0290] Three InDs (lnD31 , 24, 12) showed striking correlations with signatures of other classes. lnD31 displayed distinct C deletions at short homopolymers (<5bp) followed by 3’G, T deletions at short homopolymers (<5 bp) followed by 3’A, and longer deletions (>= 2bp) with microhomologous sequences. It was only reported in samples with novel rare SBS1052, and often co-occurred with lnD8. lnD24 deletions peaked prominently at GTA and GTG and were strongly correlated with DBS8, which is defined by double substitutions at the same motifs (TGTG>TAGG / TTGG). In D12 exhibited C deletions between AA and AT and was associated with DBS25 (associated with TBS1 , see Degasperi et al. Science 2022) featuring a tall peak at TT dinucleotide (excess of TTT>AAA followed by TTT>GAA and TTT>CAA). Despite clear co-occurrence, the causes for these signatures remain cryptic. lnD4b and lnD29 shared features with lnD4a and lnD8, respectively. Whether they represented tissue-specific variants, were mixed, or caused by different mechanisms requires further investigations. lnD11 appears related to lnD1 and may be an oversplit signature frequently enriched in high indel burden samples, such as those with MMRd and polymerase dysfunction. Seen in bladder and colorectal cancers, lnD23 showed a striking pattern of longer (>=5bp) at non-repeats. These insertions were almost exclusively tandemly duplicated from immediate neighboring sequences. lnD33 was most prominent in one CNS tumour treated with temozolomide; however, its aetiology remains unknown. lnD2b is another version of lnD2, seen in Skin and CNS and could represent a tissue-specific version of lnD2 (likely associated with T-Slip). Further samples would be required to confirm or refute this. lnD3b is a distinct signature to lnD3a yet seen in lung cancers with tobacco exposure. It may represent a modulated version of the tobacco-associated mutational process. lnD4b is due to the activity of TOP1 / RNASeH2 (Reijns et al, Nature 2022). lnD8 is RT / NHEJ DB2 / 11 is associated with the activity of the Non-homologous End Joining (NHEJ) pathway, and described in the context of samples that have been exposed to radiation. It is therefore likely to be associated with NHEJ and / or radiotherapy. It is characterised by deletions with little to no microhomology at deletion junctions. lnD12 is associated with DBS25 (Degasperi et al., 2022), a signature characterized by an excess of TT>AA that, on inspection, revealed a triple base substitution signature (TBS). lnD24 is characterised by T deletions at GTA and GTG, and is strongly correlated with DBS8 which are defined by TGTG>TAGG and TGTG>TTGG double substitutions.

[0291] Five new indel signatures of unknown aetiology were also identified, including lnD5, lnD23 (characterised by long templated insertions (5+bp) at non-repeat (R1), seen in bladder and colorectal cancers; these insertions were almost exclusively tandemly duplicated from immediate neighbouring sequences.), lnD26 (characterised by C insertions at poly-C tracts followed by a 3’A; insertions mainly occurred at longer poly-C5 and did not induce hypermutation), lnD29, and lnD33 (most prominent in one CNS tumour treated with temozolomide). Ind23 is a signature characterised by long templated insertions (5+bp) at non-repeat (R1), seen primarily in bladder and colorectal cancers.

[0292] In summary, five InDs were of likely exogenous origins (lnD3a, 3b, 13, 18, 32), 20 were endogenous (lnD1 , 2a, 2b, 4a, 4b, 6, 7, 8, 9a, 9b, 9c, 11 , 14, 15, 16a, 16b, 19, 20, 21 , 29), and 12 had uncertain sources (lnD5, 10, 12, 23, 24, 26, 27, 28, 28m, 30, 31 , 33 (Table 10). Thus, by using the new indel classification described in Example 2, the inventors were able to confirm 10 previously described signatures, and identify 15 new signatures with putative causes and associations (of which 8 are associated with dysfunction of post-replicative MMR and polymerase activity) - i.e. a total of 21 signatures with associated aetiology, 8 new signatures with unknown cause, and 4 artefactual signatures. By contrast, the current classification (COSMIC) identified 18 signatures, of which 10 have a putative aetiology (ID1 , ID2, ID3, ID6, ID7, ID8, ID13, ID17 and ID18 - although the present work demonstrates that at least ID7 did not in fact correctly identify the associated aetiology), and 9 signatures with unknown I artefactual aetiology (ID4, ID5, ID9, ID10, ID11 , ID12, ID14, ID15, ID16).

[0293] Methods

[0294] See Examples 1 and 2.

[0295] Indel signature extraction. The approach used for signature extraction was motivated by previous work (Degasperi, 2022). The inventors observed that hypermutator samples (samples containing > 100K indels) strongly influenced signature extraction using the standard beta-divergence NMF model. They sought to filter out hypermutator samples from the initial extraction. For each tissue, they first removed samples with a total indel burden < 100, and then clustered sample profiles according to their cosine similarity (cosine distance, 1 - cosine similarity) using hierarchical clustering with complete linkage. Clusters of similar samples were determined by thresholding the resulting dendrogram such that the average silhouette width was maximised and within-cluster variation was minimised. To determine cluster-specific hypermutators, for each cluster they fit the total burden per sample using a two-component Gaussian mixture model (mixtools, cran.r- project.org / web / packages / mixtools / vignettes / mixtools.pdf) compared to a one-component mixture model, using the Bayesian Information Criteria for model selection. Hypermutators were defined as the union of: (i) samples with total burden > 3rd Quartile + 1 ,5*IQR (interquartile range) where quartiles were calculated over the total dataset, and (ii) samples with a greater than 50% probability of being generated by the higher-burden Gaussian distribution. Normal and hypermutator clusters were then manually reviewed per tissue, and only normal clusters used in primary extraction.

[0296] Signature extraction was performed per tissue as in Degasperi et al. (2022), with an increased number of bootstraps (40) and repetitions per bootstrap (1000), to increase final solution stability.

[0297] Signature refitting and reliably determining excess variation. Following Degasperi et al. 2022, the inventors sought to determine if excess variation indicative of rare or unextracted signatures was present in the sample catalogue. In refitting tissue-specific catalogues, hypermutator samples included, with signatures extracted from non-hypermutator samples using FitMS (Degasperi et al., 2022), the inventors observed that profiles with lower indel counts displayed higher degrees of error (as measured by total residual, normalized by sample indel mutational burden). Generally, error decreased logarithmically as indel burden increased. Using a single threshold on fitting error to define samples with excess variation, as in Degasperi et al. 2022, therefore excluded a large proportion of samples. To more accurately calibrate the expectation of fitting error, the inventors performed a parametric bootstrapping procedure to generate a sample-specific expected error distribution. For each sample, they constructed a multinomial distribution using the per-channel density from the normalized reconstructed profile produced by FitMS. Using this distribution, they simulated 10,000 sample profiles with a total burden equal to the sample burden, and then fit these profiles with FitMS and calculated the resulting error distribution. Comparing the experimentally derived error distribution to the resulting null distribution allowed to estimate an empirical p value. This procedure was repeated for all samples in a GEL cohort, and p values corrected for multiple testing. To control the false discovery rate at 5%, samples with an adjusted p value less than 0.05 were selected for further analysis to determine rare signatures.

[0298] Rare signature extraction and refinement. For each tissue, samples with excess variation were clustered using the residual after subtracting out FitMS reconstructed profiles. Hierarchical clustering with average linkage and Euclidean distance resulted in multiple clusters per tissue, and per-cluster secondary signature extraction was performed using the primary tissue-specific signature set to initialise the signature matrix. An additional 1 - 5 rare signatures were determined per cluster, and the number of rare signatures was determined as the minimal value of n such that rare signatures were not found to perfectly recapitulate cluster members, or match common signatures. All extracted rare signatures across a tissue were subject to manual curation to identify the recurrent patterns, and the examples that displayed minimal mixing with common signatures were selected. These are signatures that were unique and could not faithfully be decomposed into previously identified common signatures. Using this consensus set of common and rare signatures per tissue, all samples in each catalogue were re-fit using FitMS to determine per-sample signature exposures.

[0299] Signature Fit Multi-Step (FitMS) is an algorithm designed to estimate signature exposures taking advantage of the concept of common and rare signatures. In general, given a mutational catalogue c, a signature fit algorithm attempts to find a set of nonnegative exposures e that indicate the number of mutations associated with each signature in a given signature matrix S, such that c ~ Se. FitMS has two steps. In the first step, only common signature exposures are estimated. In the second step, the presence of potential rare signatures is estimated. In particular, the algorithm attempts to improve the fit by adding a small number of rare signatures (one by default). This is achievable through two possible strategies: constrainedFit or errorReduction. The constrainedFit strategy uses constrained non-negative least squares (limSolve R package) to estimate the residual between the observed and reconstructed catalogues, using only common signatures. If this residual resembled a rare signature (cosine similarity of at least 0.8) then we assumed that rare signature was present in the sample. In the errorReduction strategy, the error (KLD) between the original catalogue and the fit obtained using only common signatures was compared with the error obtained using one additional rare signature, for all rare signatures considered. A rare signature is considered present if the reduction in error is at least 15%. Sample exposures are recomputed using both common signatures and any additional rare signatures. In particular, in the constrainedFit strategy: common signatures are fitted using a non-negative least squares algorithm with the additional constraint that the difference between the original catalogue c and the reconstructed catalogue Se should be mostly positive, c - Se > -T ■ Zi ci, with T = 0.003 (limSolve R package). The residual R = c - Se is then compared to the rare signatures, and if there are rare signatures with cosine similarity of at least 0.8 to R, then the rare signature with the highest cosine similarity is chosen. Finally, the common signatures and the selected rare signature are fitted into the catalogue using a non-negative KLD optimization (NNLM R package). In particular, in the errorReduction strategy: common signatures are fitted using a non-negative KLD optimization. All rare signatures are then fitted one at a time along with the common signatures. Rare signatures that caused the mean absolute deviation between c and Se to reduce at least 15% with respect to using the common signatures alone, are considered. Finally, if more than one rare signature is considered, the rare signature that induced the highest cosine similarity between the catalogue c and the model Pe is selected.

[0300] Deriving consensus indel signatures (InD). Indel signatures from all 7 GEL cancer cohorts were clustered (111 tissue specific InDs) using hierarchical clustering with average linkage and 1 - cosine similarity as distance. This was used to derive a set of consensus signatures following the framework previously described (Degasperi, 2022). The associated tissue-specific signature to reference signature conversion matrix is shown in Table 8. This was also used to transform tissuespecific signature exposures to consensus signature (final InD) exposures for downstream analyses.

[0301] Annotating InD reference signatures using gene-edit and exposure-based signatures. Variants in samples generated in previous studies (Zou et al., 2021) were reclassified using our refined indel classification scheme, and 89-channel profiles constructed. Experimental signatures were obtained via background subtraction and determined using a bootstrapping and cosine similaritybased framework previously described (Zou et al. 2021). Consensus indel signatures and all experimentally derived signatures were clustered using (1 - cosine similarity) as distance metric and hierarchical clustering. ‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘ L‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0‘0 ■■gj91sn|Q"9>| J9ppE|g_|96d

[0302] Table 9. Consensus signatures 1 , 2a, 2b, 4a, 3b.

[0303] Table 9 (continued). Consensus signatures 4a, 4b, 5, 6, 7.

[0304] Table 9 (continued). Consensus signatures 8, 9a, 9b, 9c, 10.

[0305] Table 9 (continued). Consensus signatures 11 , 12, 13, 14, 15.

[0306] Table 9 (continued). Consensus signatures 16a, 16b, 18, 19.

[0307] Table 9 (continued). Consensus signatures 20, 21, 23, 24.

[0308] Table 9 (continued). Consensus signatures 26, 27, 28, 28m.

[0309] Table 9 (continued). Consensus signatures 29, 30, 31 , 32, 33.

[0310] Example 4 - A clinical classifier of PRR dysfunction

[0311] PRRd subtypes, typified by MSI, are clinically actionable with potential selective sensitivity to immunotherapies (Cristescu et al. 2018; Wang et al. 2019; Das et al. 2022). Current methods of detecting PRRd mainly rely on immunohistochemistry (IHC) staining of MMR proteins (but not for polymerase mutants) and / or PCR-based assays to determine MSI at selected genomic loci. These assays are not sensitive or robust enough, especially in non-epithelial tissues. The inventors used insights from the work described in the above examples to construct a classifier for PRRd stratification, reporting MMRd, polymerase dysfunctional (Pol-dys), and mixed MMRd / Pol-dys as distinct classes, versus PRR-proficiency.

[0312] The inventors trained multiple multinomial elastic net regression models on 571 GEL cancers confidently assigned as MMRd (n=214), Pol-dys (n=36), mixed MMRd / Polymerase dysregulation (n=41), or PRR-proficient (negative controls, n=280) based on confirmed causal genotypes and allelic status, and / or supporting immunohistochemistry (Fig. 14a). Supporting immunohistochemistry is standard practice in the clinic. Usually we stain for the 4 MMR proteins: MLH1 / PMS2 and MSH2 / MSH6, and determine whether there is concomitant losses of the noted pairs (as expected since they tend to heterodimerize). Samples treated as controls had neither MMRd and / or Pol-dys confirmed through lack of driver mutations in key MMR genes (i.e. , MLH1 , MSH2, MSH6, PMS2), POLE, POLD1 , and displayed no evidence of signatures associated with these abnormalities. The inventors trained multiple multinomial elastic net regression models applying 7:3 partitioning iteratively across the dataset. Multinomial regression (also known as polytomous logistic regression, multiclass logistic regression, softmax regression, multinomial logit (mlogit), maximum entropy (MaxEnt) classifier, and conditional maximum entropy model) is a classification method that generalises logistic regression to multiclass problems. It predicts a score associated with assigning a sample i to a category k, as a linear combination of predictor variables (i.e. a linear function f(k,i)=(3k-Xj where Pk are regression coefficients (3o,k, Pi,k ... pM,k corresponding to the intercept (Po,k) and M predictive variables for class k (Pi,k ...pM,k), xi is a vector of values of the M predictive variables for sample i, and f(k,i) is the score for sample i and class k). Probabilities are obtained from each class as P(Yt= 1, and the vectors

[0313] Pk are jointly estimated for all classes for example using a maximum likelihood estimation. Elastic net regression is a regularised regression method that includes Li (Lasso) and L2 (ridge) penalties.

[0314] Through exploring all possible models, the inventors identified exposures of SBS and InD signatures associated with Pol-dys, MMRd, and mixed Pol-dys / MMRd, as well as the ratio of total indels to total substitutions as the most effective predictive features (see Methods for details, Fig. 14b). The final model, termed PRRDetect (Post-Replicative Repair Detect), was re-trained on the entire dataset (n=571 ) and achieved an AUC (area under the curve) of 1 at distinguishing PRRd samples from PRR-proficient samples (Fig. 14c). The ability to distinguish individual groups of MMRd, Pol-dys, and mixed MMRd / Pol-dys tumours from the rest of the cohort was also excellent (AUC=1 , 0.999 and 0.999, respectively). Then, in an independent validation cohort of 504 ICGC breast cancers and 847 GEL cancers, where the true labels of PRRd were known, PRRDetect achieved an AUC (area under the ROC curve) of 1 at distinguishing PRR-dysfunction from PRR-proficient samples, performing superiorly to other MSI / MMRd detection tools, including MSIseq (Huang et al. 2015), MMRDetect (Zou et al. 2021) and Tumour Mutational Burden (TMB) - an approved biomarker for immunotherapies (Fig. 14c, d).

[0315] Figure 14b shows the distribution of coefficients for each of 8 genomic features tested (including the intercept) from ten replicates of training in cross validation. The genomic features are (from left to right): intercept, sum of exposure of MMRd-associated SBS signatures, sum of exposure of MMRd-associated indel signatures, sum of exposure of combined MMRd / Pol-dys SBS signatures, sum of exposure of combin...

Claims

CLAIMS1. A computer-implemented method of determining whether a subject has a tumour that is deficient for post-replicative repair, the method including the steps of, for a DNA sample obtained from said tumour: determining the values of a plurality of mutational signature metrics for the sample, wherein the plurality of mutational signature metrics comprise: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)-deficiency; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)-deficiency; and based on said values of said mutational signature metrics, determining whether said sample has a high or low likelihood of being PRR-deficient, wherein a sample that has a high likelihood of being PRR-deficient has a high likelihood of having mismatch repair deficiency (MMRd), polymerase dysfunction (pol-dys), or combined mismatch repair deficiency and polymerase dysfunction (MMRd / pol-dys).

2. The method of claim 1 , wherein the one or more indel signatures of PRR-deficiency are selected from: one or more indel signatures associated with mismatch repair deficiency (MMRd), one or more indel signatures associated with polymerase dysfunction (pol-dys), and one or more indel signatures associated with combined MMRd and pol-dys (MMRD / pol-dys).

3. The method of any preceding claim, wherein the one or more SBS signatures of PRR- deficiency are selected from: one or more SBS signatures associated with mismatch repair deficiency (MMRd), one or more SBS signatures associated with polymerase dysfunction (pol- dys), and one or more SBS signatures associated with combined MMRd and pol-dys (MMRD / pol-dys).

4. The method of any preceding claim, wherein the one or more indel signatures of PRR- deficiency comprise one or more indel signatures associated with MMRd and / or wherein the one or more SBS signatures of PRR-deficiency comprise one or more SBS signatures associated with MMRd.

5. The method of any preceding claim, wherein determining whether said sample has a high or low likelihood of being PRR-deficient comprises classifying the sample between a plurality of classes comprising: a class associated with a high likelihood of MMRd, a class associated with ahigh likelihood of pol-dys, a class associated with a high likelihood of combined MMRd / pol-dys, and a class associated with a high likelihood of PRR-proficiency; and / or wherein the subject is a human subject and / or wherein the subject has a cancer selected from: colorectal cancer, stomach cancer, uterine cancer, lung cancer, bladder cancer, and central nervous system cancer.

6. The method of any preceding claim, wherein the plurality of mutational signature metrics comprise: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)-deficiency, wherein the one or more indel signatures of PRR-deficiency comprise one or more indel signatures associated with combined MMRd and pol-dys (MMRD / pol-dys), one or more indel signatures associated with MMRd, and optionally one or more indel signatures associated with pol-dys; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)-deficiency, wherein the one or more SBS signatures of PRR-deficiency comprise one or more SBS signatures associated with pol-dys, one or more SBS signatures associated with MMRd, and optionally one or more SBS signatures associated with combined MMRd and pol-dys.

7. The method of any preceding claim, wherein the method further comprises determining the ratio of the number of indels to the number of SBS in the sample, and wherein determining whether said sample has a high or low likelihood of being PRR-deficient is based on said ratio or a value derived therefrom and said mutational signature metrics, optionally wherein determining whether said sample has a high or low likelihood of being PRR-deficient is based on said mutational signature metrics and a value derived from said ratio by applying one or more transformations selected from: log transformation, normalisation and adding a fixed number to the numerator and denominator.

8. The method of any preceding claim, wherein the plurality of mutational signature metrics comprise: the proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)-deficiency; and the proportion of SBS in the sample associated with one or more SBS signatures of postreplication repair (PRR)-deficiency.

9. The method of any preceding claim, wherein determining whether said sample has a high or low likelihood of being PRR-deficient comprises using a trained machine learning model that takes as input the values of said mutational signature metrics and produces as output one ormore probabilities and / or classification labels indicative of whether the sample has a high or low likelihood of being PRR-deficient, optionally wherein the machine learning model has been trained using training data comprising the values of said mutational signature metrics for a plurality of training samples with known PRR status, optionally wherein the plurality of training samples are each associated with a label selected from: PRR proficient, MMRd, pol-dys, combined MMRd / pol-dys.

10. The method of claim 9, wherein the machine learning model is a multinomial classification or regression model, or a regularised multinomial classification or regression model, optionally wherein the machine learning model is a multinomial regression model trained with elastic net regularisation.11 . The method of claim 9 or claim 10, wherein determining whether said sample has a high or low likelihood of being PRR-deficient comprises: generating, by said machine learning model using said values of said one or more mutational signature metrics, one or more probabilistic scores; and based on said one or more probabilistic scores, determining whether said sample has a high or low likelihood of being PRR-deficient, optionally by comparing said one or more probabilistic scores with respective predetermined thresholds, wherein the probabilistic scores comprise at least a first probabilistic score that is such that the sample has a high likelihood of being PRR-deficient when the probabilistic score is at or below a threshold and a low likelihood of being PRR-deficient when the probabilistic score is above the threshold.

12. The method of claim 11 , wherein the first probabilistic score is a probability of the sample belonging to a class associated with a high likelihood of PRR-proficiency, and the method comprises determining that the sample has a low likelihood of being PRR-deficient when said first probabilistic score for said sample is above the threshold, optionally wherein the threshold is 0.5.

13. The method of any of claims 9 to 12, wherein the machine learning model produces as output: a first probabilistic score associated with a first class associated with a high likelihood of PRR-proficiency, a second probabilistic score associated with a second class associated with a high likelihood of MMRd, a third probabilistic score associated with a third a class associated with a high likelihood or pol-dys, and a fourth probabilistic score associated with a fourth class associated with a high likelihood of combined MMRd / pol-dys.

14. The method of claim 13, wherein the method comprises determining that the sample has a low likelihood of being PRR-deficient when said first probabilistic score for said sample isabove a threshold, or determining that the sample has a high likelihood of being PRR-deficient when said first probabilistic score for said sample is at or below the threshold, optionally wherein determining that the sample has a high likelihood of being PRR-deficient when said first probabilistic score for said sample is at or below the threshold comprises determining that the sample has the type of PRRd associated with the class of the second, third or fourth class that is associated with the highest probabilistic score.

15. The method of any preceding claim, wherein determining whether said sample has a high or low likelihood of being PRR-deficient does not use the tumour mutational burden of the sample.

16. The method of any preceding claim, wherein determining whether said sample has a high or low likelihood of being PRR-deficient uses values derived from said values of said mutational signatures by applying a log transformation and / or a normalisation.

17. The method of any preceding claim, wherein the mutational signature metrics comprise one or more of: the sum of the number of indels or proportion of indels in the sample associated with each of a plurality of indel signatures associated with MMRd; the sum of the number of indels or proportion of indels in the sample associated with each of a plurality of indel signatures associated with pol-dys; the sum of the number of indels or proportion of indels in the sample associated with each of a plurality of indel signatures associated with combined MMRd / pol-dys; the sum of the number of single base substitutions (SBS) or proportion of SBS in the sample associated with each of a plurality of SBS signatures associated with MMRd; the sum of the number of single base substitutions (SBS) or proportion of SBS in the sample associated with each of a plurality of SBS signatures associated with pol-dys; and the sum of the number of single base substitutions (SBS) or proportion of SBS in the sample associated with each of a plurality of SBS signatures associated with combined MMRd / pol- dys.

18. The method of any preceding claim, wherein the one or more SBS mutational signatures are selected from: one or more MMRd-associated SBS signatures, one or more Pol-dys-associated SBS signatures, and one or more combined MMRd / Pol-dys SBS signatures; optionally wherein the one or more MMRd-associated SBS signatures are selected from: an SBS signature (e.g. SBS6) that has higherweights for A[C>T]G, C[C>T]G and G[C>T]G classesof mutations than for any other classes of mutations, an SBS signature (e.g. SBS15) that has higher weights for the G[C>T]G class of mutations than for any other classes of mutations, an SBS signature (e.g. SBS26) that has higher weights for A[T>C]A, A[T>C]G, C[T>C]G, G[T>C]A, G[T>C]G, G[T>C]T, T[T>C]A and T[T>C]T classes of mutations than for any other classes of mutations, an SBS signature (e.g. SBS44) that has higher weights for C[C>A]T, A[C>T]A, A[C>T]G, G[C>T]A, G[C>T]C, G[C>T]G, G[C>T]T G[T>C]A and T[T>C]T classes of mutations than forany otherclasses of mutations, and an SBS signature (e.g. SBS97) that has higherweights for G[C>T]G and G[T>C]C classes of mutations than for any other classes of mutations; and / or wherein the one or more Pol-dys-associated SBS signatures are selected from: an SBS signature (e.g. SBS10a) that has higher weights for T[C>A]T, T[T>G]T and T[C>T]G classes of mutations than for any other classes of mutations, and an SBS signature (e.g. SBSI Od) that has higher weights for T[C>A]A and T[C>A]T classes of mutations than for any other classes of mutations; and / or wherein the one or more combined MMRd / Pol-dys SBS signatures are selected from: an SBS signature (e.g. SBS14) that has higher weights for C[C>A]A, G[C>A]T, T[C>A]T, A[C>T]G, G[C>T]A, G[C>T]C, G[C>T]G, and G[C>T]T classes of mutations than for any other classes of mutations, and an SBS signature (e.g. SBS20) that has higher weights for the C[C>A]T class of mutations than for any other class of mutations.

19. The method of any preceding claim, wherein the one or more indel signatures are selected from: one or more MMRd-associated indel signatures (such as e.g. signatures lnD7 and / or lnD19 as defined in Table 9), optionally wherein the one or more MMRd-associated indel signatures are signatures that are primarily associated with 1 and / or 2 or more bp indels at repeat tracts or 5 or more repeats and / or wherein the or more MMRd-associated indel signatures comprise a signature that is characterised by frequent 2 or more bp insertions at long repeat tracts (e.g. 5 or more repeats - e.g. I n D7), and a signature that is characterised by 1 and 2 or more bp deletions at repeat tracts of 5 or more repeats (e.g. lnD19); one or more Pol-dys-associated indel signatures (such as e.g. signatures Indi 4 and / or lnD15 as defined in Table 9), optionally wherein the one or more Pol-dys-associated indel signatures comprise signatures that are primarily associated with 1 bp T insertions, and / or wherein the one or more Pol-dys associated indel signatures comprise a signature that is associated 1 bp T insertions in repeat tracts of 5 to 7 repeats in a 5’A sequence context (e.g. lnD14), and a signature that is associated with 1 bp T insertions in a plurality of sequence contexts including AC for all repeat lengths, and AA, CA, CC, GA, GC for repeats of 5 to 7 repeats (e.g. I n D 15); and one or more combined MMRd / Pol-dys indel signatures (such as e.g. signatures selected from: lnD16a, lnD16b, lnD20 and lnD21 as defined in Table 9), optionally wherein the one or more combined MMRd / Pol-dys indel signatures comprise signatures that are primarily associated with1 bp T deletions at long repeat tracts (e.g. 5 or more repeats), 1 bp deletions, 1 bp insertions and deletions at long repeat tracts (e.g. 5 or more repeats), and 1 bp T deletions in AA contexts, and / or wherein the one or more combined MMRd / Pol-dys indel signatures comp (i) a signature that is associated with 1 bp T deletions in AA contexts in repeat tracts of up to 7 repeats and 1 bp T deletions in AC contexts in repeat tracts of 5 to 7 repeats (e.g. lnD20), (ii) a signature that is associated with 1 bp T insertions in repeat tracts of 5 to 7 repeats in 5’A and CG and GG sequence contexts, 1 bp T insertions in repeat tracts of 8 or more repeats AC sequence contexts, 1 bp T deletions in repeat tracts of 8 or more repeats in AA and CC sequence contexts (e.g. In D21 ), (iii) a signature associated with 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context (e.g. lnD16a), and (iv) a signature associated with 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context and 1 bp T insertions are repeat tracts of 5 to 7 repeats in AA or AC sequence contexts (e.g. lnD16b).

20. The method of claim 19, wherein the one or more MM Rd-associated indel signatures are selected from: an indel signature (e.g. lnD7) that has higher weights for categories comprising 2 or more bp deletions at repeat tracts of 5 or more repeats (e.g. Del(2,):U(1 ,2):R(5,9)), 1 bp T deletions at repeat tracts of 8 or more repeats in any sequence context (e.g. A[Del(T):R(8,9)]A, A[Del(T):R(8,9)]C, A[Del(T):R(8,9)]G, C[Del(T):R(8,9)]A, C[Del(T):R(8,9)]C, C[Del(T):R(8,9)]G, G[Del(T):R(8,9)]A, G[Del(T):R(8,9)]C, G[Del(T):R(8,9)]G), 1 bp T deletions at repeat tracts of 5 or more repeats in an AA sequence context (e.g. A[Del(T):R(5,7)]A), and 1 bp C deletions at repeat tracts of 6 or more bp (e.g. Del(C):R(6,9)), than for any other categories of indels; and an indel signature (e.g. lnD19) that is associated with PMS2 deficiency and / or that has a higher weight for categories comprising 2 or more bp insertions at repeat tracts of 5 or more repeats (e.g. I ns(2 ,) :R(5,9)) , than for any other categories of indels.

21. The method of claim 19 or claim 20, wherein the one or more pol-dys-associated indel signatures are selected from: an indel signature (e.g. lnD14) that is associated with POLD1 exonuclease mutations and / or that has higher weights for categories comprising 1 bp T insertions in repeat tracts of 5 to 7 repeats in a 5’A sequence context (e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, A[lns(T):R(5,7)]G), than for any other categories of indels; and an indel signature (e.g. lnD15) that is associated with POLE exonuclease mutations and / or that has higher weights for categories comprising 1 bp T insertions in a plurality of sequence contexts, than for any other categories of indels. The plurality of sequence contexts may include AC (for all repeat lengths), AA, CA, CC, GA, GC (all particularly for repeats of 5 to 7 repeats), such as e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, C[lns(T):R(5,7)]A, C[lns(T):R(5,7)]C,G[lns(T):R(5,7)]A, G[lns(T):R(5,7)]C, and / or A[lns(T):R(8,9)]C.

22. The method of any of claims 19 to 21 , wherein the one or more combined MMRd / Pol-dys indel signatures are selected from: an indel signature (e.g. lnD16a) that is associated with concurrent loss of POLE proofreading and MMR and / or that has higher weights for categories comprising 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context (e.g. A[Del(T):R(8,9)]A, A[Del(T):R(5,7)]A), than for any other categories of indels; an indel signature (e.g. lnD16b) that is associated with concurrent loss of POLE proofreading and MMR and / or that has higher weights for categories comprising 1 bp T deletions at repeat tracts of 5 or more repeats in an AA context (e.g. A[Del(T):R(8,9)]A, A[Del(T):R(5,7)]A) and 1 bp T insertions are repeat tracts of 5 to 7 repeats in AA or AC sequence contexts (e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C), than for any other categories of indels; an indel signature (e.g. lnD20) is associated with combined MMRd and POLE dysregulation and / or that has higher weights for categories comprising 1 bp T deletions in AA contexts in repeat tracts of up to 7 repeats and 1 bp T deletions in AC contexts in repeat tracts of 5 to 7 repeats (e.g. A[Del(T):R(1 ,4)]A, A[Del(T):R(5,7)]A, A[Del(T):R(5,7)]C), than for any other categories of indels; and an indel signature (e.g. lnD21) that is associated with concurrent POLE proofreading defect and MMR, and / or that has higher weights for categories comprising 1 bp T insertions in repeat tracts of 5 to 7 repeats in 5’A and CG and GG sequence contexts, 1 bp T insertions in repeat tracts of 8 or more repeats AC sequence contexts, 1 bp T deletions in repeat tracts of 8 or more repeats in AA and CC sequence contexts (e.g. A[lns(T):R(5,7)]A, A[lns(T):R(5,7)]C, A[lns(T):R(5,7)]G, C[lns(T):R(5,7)]G, G[lns(T):R(5,7)]G, A[lns(T):R(8,9)]C, A[Del(T):R(8,9)]A, C[Del(T):R(8,9)]C), than for any other categories of indels.

23. The method of any preceding claim, wherein the one or more indel signatures are signatures that comprise a plurality of weights fora respective plurality of predetermined categories of indels, wherein the plurality of predetermined categories comprise a plurality of categories comprising insertions of 1 bp, a plurality of categories comprising insertions of more than 1 bp, a plurality of categories comprising deletions of 1 bp, a plurality of categories comprising deletions of more than 1 bp, wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises categories that differ from each other by the 5’ and / or 3’ sequence context of the insertion / deletion or repeat tract comprising the insertion / deletion, and wherein the plurality of categories comprising insertions of 1 bp and the plurality of categories comprising deletions of 1 bp comprises a plurality of categories comprising insertions / deletions in repeat tracts of 6 or more bp, said plurality of categories differing by the length of the repeat tract.

24. A method of predicting whether a subject with cancer is likely to respond to an immunotherapy, the method comprising characterising a sample obtained from a tumour in the subject as having a high or low likelihood of being PRR-deficient using a method of any preceding claim, wherein when the sample is characterised as having a high likelihood of being PRR-deficient, the subject is likely to respond to immunotherapy.

25. A method of providing a tool for characterising a DNA sample obtained from a tumour, the method including the steps of: obtaining mutational signature profiles for a plurality of training samples associated with known PRR-deficiency status; obtaining a training data set comprising, for each of a plurality of training samples associated with known PRR-deficiency status, the value of a plurality of mutational signature metrics comprising: the number of indels or proportion of indels in the sample associated with one or more indel signatures of post-replication repair (PRR)-deficiency; and the number of single base substitutions (SBS) or proportion of SBS in the sample associated with one or more SBS signatures of post-replication repair (PRR)-deficiency; and training a machine learning model to predict, based on said values of said one or more mutational signature metrics, whether each training sample has a high or low likelihood of being PRR-deficient, wherein a sample that has a high likelihood of being PRR-deficient has a high likelihood of having mismatch repair deficiency (MMRd), polymerase dysfunction (pol-dys), or combined mismatch repair deficiency and polymerase dysfunction (MMRd / pol-dys).

26. A system comprising: a processor; and a non-transitory computer readable medium comprising instructions that, when executed by the processor, cause the processor to implement the method of any of claims 1 to 25.

27. A non-transitory computer readable medium or media comprising instructions that, when executed by at least one processor, cause the at least one processor to perform the method of any of claims 1 to 25.

Images