Methods and systems for cancer detection

Abstract

A computer-implemented method for predicting whether a subject has a cancer, comprising providing multimodal data comprising at least two feature types, said feature types comprising a metric of methy

Description

Field of the Invention The present invention relates to methods and systems for assigning a probability of cancer and to related products. It is particularly, although not exclusively, concerned with methods of using methylome sequencing data from samples including cell free DNA (cfDNA) to predict the presence of cancer, including predicting a probability of cancer, in a patient. Background Methylomics is used to assess the methylation levels of circulating DNA, such as cell free DNA (cfDNA) including circulating tumour DNA (ctDNA). The enzymatic addition of a methyl group to a cytosine base of DNA leads to the epigenetic silencing of a gene, and is a vital component of gene regulation. However, it has been observed that abnormal DNA methylation is present in diseases such as cancer, lupus, and muscular dystrophy. Characterisation of the human methylome has recently come to the forefront of epigenetic studies in the wake of ever improving sequencing technologies. Loyfer et al. 2023 recently performed the first genome-wide characterisation of the human cell methylome. The authors sequenced the methylome in 77 primary cell types, and identified a total of 7,104,162 non-overlapping, continuous blocks of homogeneously methylated CpG sites, before using this information to investigate variation in methylation patterns across cell and tissue types . Heider et al. 2020 describes detection of circulating tumour DNA (ctDNA) from dried blood spots after DNA size selection. WO2024 / 083860A1 describes non-invasive disease detection and monitoring, including cancer detection using methylation signal(s) from cell-free DNA (cfDNA). Despite these advances, there remains a need for methods for detecting, predicting or providing a probability of cancer in a patient. The present invention has been devised in light of the above considerations . Suxamary of the Invention The inventors devised a classification framework for predicting cancer in cfDNA containing samples. In particular, the inventors identified that patterns of differential methylation between healthy and cancerous samples containing cfDNA can be used to train a classifier model for predicting the presence of a cancer. The inventors found that processing methylation sequencing data to identify "differentially methylated blocks" (DMBs) of DNA specific to target cancer types, and calculate a copy number feature (such as median absolute deviation, MAD from copy number neutrality) for each sample, provided two features which were sufficient to train and test a random forest machine learning classification model for predicting the presence of cancer. The inventors showed the robustness in their approach, in that predictive DMBs for ovarian cancer could be identified from both their own sample data (73 DMBs, Table 3), and public data (716 DMBs, Table 4), with every one of the 73 DMBs of Table 3 also present in the larger set of 716 DMBs in Table 4. The inventors demonstrated the power of this approach with an AUG of 0.955 on the hold-out test set (Example 3). Further, the inventors bolstered the predictive power of this first model with additional features including an accumulated signal in tissue-specific hypomethylated blocks feature, a fragment length ratio feature, and / or a fragment 5'-end sequence motif group feature. The inventors demonstrated the power of this further approach with an AUG of 0.991 on the hold-out test set (Example 4). The predictive power of the classification framework described herein is therefore clear, including its use therefore in predicting cancer in individual patients, in particular the early detection of cancer; detection of minimal residual disease (MRD) or cancer recurrence following treatment, such as following surgical treatment; monitoring cancer progression or remission; monitoring response to an anti-cancer therapy; and / or assessing the stage and / or severity of cancer. Additionally, the inventors provide an optimised cfDNA extraction protocol that builds on existing protocols for blood spot cfDNA extraction (see, e.g., WO2020 / 104670A1) and NEB Enzymatic Methyl-seq® ("EM-seq™"). As described herein, the inventors were able to detect signal (e.g. methylation profile, copy number aberration patterns and fragmentomics features) even with low input DNA, such as samples of 15pg or less of cfDNA. Thus, according to a first aspect, there is provided a method, in particular a computer-implemented method, for predicting whether a subject has a cancer, comprising: providing multimodal data derived from a cell-free DNA (cfDNA)-containing sample obtained from the subject, said multimodal data comprising at least two feature types, said feature types comprising: a first feature type comprising a metric of methylation of each of at least 10 differentially methylated blocks of homogeneous methylation ("DMBs") in a cell-free DNA (cfDNA)-containing sample obtained from the subject and a second feature type selected from: a metric of somatic copy number alterations of at least 10 non-overlapping bins of some or all or the genome of the subject; an accumulated signal in tissuespecific hypomethylated blocks feature; a fragment length ratio feature; and a fragment 5'-end sequence motif group feature; inputting the multimodal data into a multimodal machine learning classifier that has been trained on a labelled training data set comprising corresponding multimodal data for each of: (i) a plurality of subjects known to have had the cancer the time of sample collection ("cases"); and (ii) a plurality of subjects known to not have the cancer at the time of sample collection ("controls"); and causing the multimodal machine learning classifier to classify the subject into the cases class and thereby as having cancer or the controls class and thereby not having cancer based on at least the inputted multimodal data. In some embodiments the classifier may output a numerical probability of cancer. In some embodiments, the present invention provides a computer-implemented method for predicting whether a subject has a cancer, comprising: providing multimodal data comprising at least two feature types, said feature types comprising a metric of methylation of each of at least 10 differentially methylated blocks of homogeneous methylation ("DMBs") in a cell-free DNA (cfDNA)-containing sample obtained from the subject and a metric of somatic copy number alterations of at least 10 non-overlapping bins of the genome of the subject in a cell-free DNA (cfDNA)-containing sample obtained from the subject; inputting the multimodal data into a multimodal machine learning classifier that has been trained on a labelled training data set comprising corresponding multimodal data for each of: (i) a plurality of subjects known to have had the cancer the time of sample collection ("cases"); and (ii) a plurality of subjects known to not have the cancer at the time of sample collection ("controls"); and causing the multimodal machine learning classifier to classify the subject into the cases class and thereby as having cancer or the controls class and thereby not having cancer based on at least the inputted multimodal data, optionally wherein the classifier outputs a numerical probability of cancer. In some embodiments, the metric of methylation comprises the fraction of methylated fragments (i.e. the number of methylated reads mapping to the DMB divided by the total number of reads (methylated and unmethylated) summed over said at least 10 DMBs. In some embodiments, the metric of somatic copy number alterations comprises the median absolute deviation (MAD) from copy number neutrality, such as the trimmed median absolute deviation (t-MAD) or ichorCNA. In some embodiments, the at least 10 non-overlapping bins of some or all of the genome of the subject are each 100 bp, 1 kb, 1 Mb or 10 Mb in size. In some embodiments, the at least 10 non-overlapping bins of some or all of the genome of the subject comprise at least 20, 30, 40, 50, 60, 70, 80, 90, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000, 5000, 10000, 50000 or at least 100000 non-overlapping bins. In some embodiments, the at least 10 DMBs comprise cancer-specific hypermethylated DMBs. In particular, each of said cancer-specific hypermethylated DMBs may be a DMB that satisfies 1, 2, 3 or all of the following conditions : (i) the DMB block comprises at least 1, 2, 3, 4 or at least 5 CpGs, wherein all CpGs, if there is more than 1, share substantially the same level of methylation in a given sample; (ii) the DMB is no more than 1000 bp, 5000 bp or no more than 10000 bp in length; (iii) the median fraction of methylated fragments mapping to the DMB in a plurality of samples from controls is less than 0.005; and (iv) the 90th quantile fraction of methylated fragments mapping to the DMB in a plurality of samples from cancer cases is at least 0.1 greater than the 90th quantile fraction of methylated fragments mapping to the DMB in a plurality of samples from controls. A DMB satisfying the above conditions may be considered to capture cancer-specific methylation signal. By applying the above conditions to a suitable data set of samples derived from patients known to have a particular cancer, DMBs informative for that particular cancer may be identified, which then find use in the methods of the present invention. In some embodiments of the present invention, the cancer is selected from: ovarian cancer, breast cancer, prostate cancer, gastrointestinal cancer (e.g. colorectal cancer, oesophagus cancer stomach cancer), endometrial cancer (uterus / womb cancer), kidney cancer (renal cell), lung cancer (small cell, non-small cell and mesothelioma), central nervous system cancer including brain cancer (gliomas, astrocytomas, glioblastomas), melanoma (including choroid melanoma and skin cancers), merkel cell carcinoma, clear cell renal cell carcinoma (ccRCC), carcinoma of unknown primary (CUP), lymphoid cancer (such as e.g. lymphoma), small bowel cancers (duodenal and jejunal), leukaemia, pancreatic cancer, hepatobiliary tumours, germ cell cancers, bone / soft tissue cancer, head and neck cancers (such as e.g. adenoid cystic carcinoma, ACC), pancreatic cancer, cervical cancer (e.g. Cervical Squamous Cell Carcinoma and endocervical adenocarcinoma, CESC), liver cancer, bladder cancer (such as e.g. bladder carcinoma, BLCA), urinary tract cancer, neuroendocrine tumour (NET), thyroid cancer and sarcomas. For example, the cancer may be any cancer represented in The Cancer Genome Atlas (TCGA) such as LAML (Acute Myeloid Leukemia), ACC (Adrenocortical carcinoma), BLCA (Bladder Urothelial Carcinoma, LGG (Brain Lower Grade Glioma, BRCA (Breast invasive carcinoma), CESC (Cervical squamous cell carcinoma and endocervical adenocarcinoma), CHOL (Cholangiocarcinoma), LCML (Chronic Myelogenous Leukemia), COAD (Colon adenocarcinoma), ESCA (Esophageal carcinoma), GBM (Glioblastoma multiforme), HNSC (Head and Neck squamous cell carcinoma), RICH (Kidney Chromophobe), KIRC (Kidney renal clear cell carcinoma), KIRP (Kidney renal papillary cell carcinoma), LIHC (Liver hepatocellular carcinoma), LUAD (Lung adenocarcinoma), LUSC (Lung squamous cell carcinoma), DLBC (Lymphoid Neoplasm Diffuse Large B-cell Lymphoma), MESO (Mesothelioma), OV (Ovarian serous cystadenocarcinoma), PAAD (Pancreatic adenocarcinoma), PCPG (Pheochromocytoma and Paraganglioma), PRAD (Prostate adenocarcinoma), READ (Rectum adenocarcinoma), SARC (Sarcoma), SKCM (Skin Cutaneous Melanoma), STAD (Stomach adenocarcinoma), TGCT (Testicular Germ Cell Tumors), THYM (Thymoma), THCA (Thyroid carcinoma), UCS (Uterine Carcinosarcoma), UCEC (Uterine Corpus Endometrial Carcinoma), and UVM (Uveal Melanoma). The cancer may be at any stage and specifically includes metastatic cancer. In particular embodiments of the present invention, the cancer is ovarian cancer. In some embodiments, such as where at least 10 DMBs are selected from In some embodiments, such as where at least 10 DMBs are selected from the cancer is ovarian cancer, the the 716 DMBs set forth in Table 4 the cancer is ovarian cancer, the the 73 DMBs set forth in Table 3. In some embodiments, such as where the cancer is ovarian cancer, the at least 10 DMBs comprise at least 20, 30, 40, 50, 60, 70 or all of the 73 DMBs set forth in Table 3. In some embodiments, such as where the cancer is ovarian cancer, the at least 10 DMBs comprise at least 20, 30, 40, 50, 100, 150, 200, 250, 300, 350, 400, 450, 500, 550, 600, 650, 700 or all 716 DMBs set forth in Table 4. In some embodiments, the multimodal data comprises as a second feature type, said metric of somatic copy number alterations of at least 10 non-overlapping bins of some or all or the genome of the subject. In some embodiments, the multimodal data further comprises at least one, at least two or three additional feature types selected from the group : (i) an accumulated signal in tissue-specific hypomethylated blocks feature; (ii) a fragment length ratio feature; and (iii) a fragment 5'-end sequence motif group feature. In some embodiments, the accumulated signal in tissue-specific hypomethylated blocks feature comprises the fraction of unmethylated reads (# of unmethylated / (# of unmethylated + # of methylated) that map to blocks of tissue-specific hypomethylation (i.e. blocks that are homogeneously hypomethylated only in the relevant tissue, but show substantially greater methylation in other tissues). The signal may be median-normalised. In particular, the tissue-specific blocks of hypomethylation may be specific for one of the following tissue types: breast-basal, breast-luminal, lung, ovarian and prostate. Loyfer et al, 2023 reported that the top 25 differentially unmethylated regions for each cell type can serve as sensitive biomarkers for quantification of the presence of DNA from a specific cell type in a mixture. Without wishing to be bound by any particular theory, the present inventors consider the accumulated tissue-specific methylation signal to be informative of the type of cancer (e.g. tissue of origin of the primary cancer). As shown in Figure 9 herein, the fraction of unmethylated reads that overlap the ovary-specific blocks of hypomethylation was higher in cfDNA taken from samples of plasma from patients with ovarian cancer than the fraction seen in healthy control plasma. The present inventors believe this may indicate an increase in ovary-derived cfDNA in the plasma or ovarian tumour-derived cfDNA that retains the ovary-specific hypomethylation. In other words, an increase in an "ovary-like" hypomethylation pattern in the cfDNA of plasma samples from ovarian tumour patients compared to controls demonstrates an "ovary signal" that may complement the "cancer signal" detected by the cancer-specific DMBs. In this way the multimodal classifier may exploit both cancer signal and tissue of origin signal to better detect the present of a cancer and to determine the type of cancer. Without wishing to be bound by any particular theory, the present inventors believe the inclusion of the accumulated signal in tissue-specific hypomethylated blocks feature to be advantageous in multicancer detection, including in classifying not only the presence of cancer per se, but of one or more specific types of cancer. In some embodiments, the method of the present invention is for predicting whether a subject has a particular type of cancer and the multimodal data comprises the accumulated signal in tissue-specific hypomethylated blocks feature. Generally, the tissue-specific hypomethylated blocks match the cancer type(s) for which prediction or detection is carried out. In particular, the tissue-specific hypomethylated blocks may comprise one or more of breast-basal, breast-luminal, lung, ovary and prostate and the cancer type(s) may comprise one or more of the cancer types breast, lung, ovarian and prostate, respectively. In some embodiments, the fragment length ratio feature comprises the ratio of: (i) the count of cfDNA fragments of length in the range 75160 bp to (ii) the count of cfDNA fragments of length in the range 167-225 bp. The cfDNA fragment counts may be as determined from DNA sequencing data derived from the cfDNA-containing sample obtained from the subject. In some embodiments, the fragment 5'-end sequence motif feature comprises the frequency, optionally centred and scaled frequency, of a) cfDNA fragments having a 4-mer 5'-end sequence belonging to a group that is underrepresented in cancer samples and b) cfDNA fragments having a 4-mer 5'-end sequence belonging to a group that is overrepresented in cancer samples. In particular, the threshold to consider the 5'-end sequence as underrepresented vs overrepresented may be an adjusted p-value of t-test of 0.0001. In some embodiments, the 4-mer 5'-end sequence motifs overrepresented in cancer comprise the Group 1 4-mer sequences set forth in Table 5 and the 4-mer 5'-end sequence motifs underrepresented in cancer comprise the Group 2 4-mer sequences set forth in Table 5. In some embodiments, the machine learning classifier exhibits predictive performance measured as area under curve of the receiver operating characteristic curve (AUROC) of at least 0.60, 0.70, 0.80 or 0.90, as assessed on a test data set comprising labelled data for a plurality of subjects known to have said cancer at the time of sample collection and a plurality of subjects known not to have any cancer at the time of sample collection, said test data set differing from said labelled training data set. Typically, the test data set comprises data for at least 10, 20, 30, 40 or at least 50 subjects of each class (e.g. at least 10 subjects known to have said cancer at the time of sample collection and at least 10 subjects known not to have any cancer at the time of sample collection). In some embodiments, the machine learning classifier comprises a Random Forest, a neural network, a support vector machine, a logistic regression, a naive Bayes or a perceptron. In a second aspect, the present invention provides a method for detecting cancer in a subject, comprising: a) providing a cfDNA-containing sample obtained from the subject; b) performing methylation sequencing of the cfDNA or of a library generated from the cfDNA to obtain methylation sequencing reads, at least a portion of said sequencing reads mapping to at least 10 differentially methylated blocks of homogeneous methylation ("DMBs"); c) analysing said sequencing reads to generate multimodal data for the subject, wherein said multimodal data comprises at least two feature types, said feature types comprising: a first feature type comprising a metric of methylation of each of said at least 10 DMBs and a second feature type selected from: a metric of somatic copy number alterations of at least 10 non-overlapping bins of some or all or the genome of the subject; an accumulated signal in tissue-specific hypomethylated blocks feature; a fragment length ratio feature; and a fragment 5'-end sequence motif group feature; and d) carrying out the computer-implemented method of the first aspect of the invention using at least the multimodal data from step c) as input data, wherein cancer is considered to have been detected when the multimodal machine learning classifier classifies the subject into the cases class and is considered not to have been detected when the multimodal machine learning classifier classifies the subject into the controls class . In some embodiments, the method of the second aspect of the invention comprises : a) providing a cfDNA-containing sample obtained from the subject; b) performing methylation sequencing of the cfDNA or of a library generated from the cfDNA to obtain methylation sequencing reads, at least a portion of said sequencing reads mapping to the at least 10 DMBs; c) analysing said sequencing reads to generate a metric of methylation of each of the DMBs and a metric of somatic copy number alterations of at least 10 non-overlapping bins of the genome of the subject, thereby generating the multimodal data for the subject; and d) carrying out the computer-implemented method of the first aspect of the invention using at least the multimodal data from step c) as input data, wherein cancer is considered to have been detected when the multimodal machine learning classifier classifies the subject into the cases class and is considered not to have been detected when the multimodal machine learning classifier classifies the subject into the controls class . In some embodiments the methylation sequencing comprises shallow whole methylome sequencing (sWMS). In particular, the methylation sequencing coverage may be in the range 0. lx to 5x, such as 0.5x to l.Ox. In some embodiments, the cfDNA-containing sample may comprise a plasma sample, a blood sample, a urine sample, a tear sample, a saliva sample, a final needle aspirate (FNA) sample, a cerebrospinal fluid (CSF) sample or a fecal sample. In particular, the sample may comprise a blood spot sample, such as a dried blood spot sample or a pin-prick blood sample. In some embodiments, the sample may contain no more than 100 pg, no more than 50 pg or no more than 20 pg of cfDNA. In some embodiments, the cfDNA-containing sample may be subjected to a single round of size selection prior to sequencing and / or prior to library preparation. The size selection step may comprise use of magnetic beads, such as SPRIselect beads. In some embodiments, the methylation sequencing comprises an enzymatic methyl-seq (EM-seq). In particular, the EM-seq may comprise the NEBNext Enzymatic Methyl-seq protocol modified to have a bead incubation time of between 7 minutes and 15 minutes, such as 10 minutes and / or modified to employ a number of PCR cycles in the range 15-20, such as 18 cycles. In some embodiments of the first or second aspect of the present invention, the method is for: early detection of cancer; detection of minimal residual disease (MRD) or cancer recurrence following treatment, such as following surgical treatment; monitoring cancer progression or remission; monitoring response to an anti-cancer therapy; and / or assessing the stage and / or severity of cancer. In some embodiments of the first or second aspect of the present invention, the method is carried out for each of a plurality of subjects, and wherein the method is for stratifying the subjects by cancer risk. For example, the plurality of subjects may be stratified into risk groups (e.g. low, medium and high) or may be stratified according to relative risk (e.g. into risk tertiles, risk quartiles or similar). In some embodiments according to the first aspect of the present invention, said provided multimodal data derived from the cell-free DNA (cfDNA)-containing sample obtained from the subject is data that was derived as defined in connection with the second aspect of the present invention. In particular, the multimodal data may have been derived from the sample by methylation, such as shallow whole methylome sequencing. In particular, the methylation sequencing coverage may be in the range 0. lx to 5x, such as 0.5x to l.Ox. The provided multimodal data may have been derived from a plasma sample, a blood sample, a urine sample, a tear sample, a saliva sample, a final needle aspirate (FNA) sample, a cerebrospinal fluid (CSF) sample or a fecal sample. The provided multimodal data may have been derived from a blood spot sample, such as a dried blood spot sample or a pin-prick blood sample. The provided multimodal data may have been derived from a sample containing no more than 100 pg, no more than 50 pg or no more than 20 pg of cfDNA. The provided multimodal data may have been derived from a cfDNA-containing sample that was subjected to a single round of size selection prior to sequencing and / or prior to library preparation. The size selection may have been carried out using magnetic beads, such as SPRIselect beads. The provided multimodal data may have been derived from the sample by enzymatic methyl-seq (EMseq) , such as the NEBNext Enzymatic Methyl-seq protocol modified to have a bead incubation time of between 7 minutes and 15 minutes, such as 10 minutes and / or modified to employ a number of PCR cycles in the range 15-20, such as 18 cycles. In a third aspect, the present invention provides a system comprising one or more data processors and a non-transitory computer readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to carry out the method of the first aspect of the invention on multimodal data provided to the one or more data processors. The system may be for use in a method of the first aspect of the invention. In some embodiments, the computer readable storage medium further comprises one or more learned numerical parameters of the machine learning classifier. In a fourth aspect, the present invention provides a computer readable storage medium containing instructions that, when executed on one or more data processors, cause the one or more data processors to carry out the method of the first aspect of the invention on multimodal data provided to the one or more data processors. In some embodiments, the computer readable storage medium further comprises one or more learned numerical parameters of the machine learning classifier. In a fifth aspect, the present invention provides a method for selecting a subject for cancer treatment, comprising: a) carrying out the method of the first or second aspect of the invention, wherein the subject is predicted to have cancer; and b) selecting the subject to receive a therapeutically effective amount of an anti-cancer agent. In a sixth aspect, the present invention provides a method of treatment of cancer in a subject, comprising: a) carrying out the method of the first or second aspect of the invention, wherein the subject is predicted to have cancer; and b) administering a therapeutically effective amount of an anticancer agent to the subject. The invention includes the combination of the aspects and preferred features described except where such a combination is clearly impermissible or expressly avoided. 5 Summary of the Figures Embodiments and experiments illustrating the principles of the invention will now be discussed with reference to the accompanying figures in which: Figure 1. Schematic showing an embodiment of a system for predicting whether a subject has a cancer according to the present disclosure . Figure 2. Flow diagram showing, in schematic form, a method of predicting whether a subject has a cancer according to the present disclosure . Figure 3. Overview of Sodium Bisulfite Conversion and EM-seq: Left -sodium bisulphite method, Right - EM-seq method. Figure 4. Libraries traces with and without size selection of input cfDNA: Libraries were prepared with input cfDNA either with 1 or 2 or no round of size selection. The libraries from 1 round of size selection are more representative of cfDNA in Tapestation traces. Figure 5. Copy Number Aberrations across plasma dilutions: Copy number aberration profiles across serial plasma dilution from a patient with metastatic prostate cancer. The copy number profiles are similar across plasma dilutions. Figure 6. Modelling scheme for training and test datasets. Figure 7. Model 1 performance. The heatmap shows signal distribution across samples and features. Top annotations show true sample class (Type), class predicted by the classification model presented herein in Example 3 (Predicted), class predicted by ichorCNA, a copy-number-only based tool (Predicted ichor), and tumour fraction estimated from TP53 mutation with ddPCR analysis (tp53 TF). Figure 8. Accumulated signal in the 716 panel-derived ovarian cancer-specific hypermethylated DMBs. Figure 9. Accumulated signal in tissue-specific hypomethylated blocks. Figure 10. Median absolute deviation (MAD) of copy numbers. Figure 11. Genome-wide fragment length ratio. Figure 12. 4-mer 5'-end sequence motifs of cfDNA fragments. Figure 13. Cancer detection from sWMS of cfDNA samples with Model 2. Detailed Description of the Invention Aspects and embodiments of the present invention will now be discussed with reference to the accompanying figures. Further aspects and embodiments will be apparent to those skilled in the art. All documents mentioned in this text are incorporated herein by reference. Systems The systems and method described herein can 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 and carrying out a method according to the described embodiments. For example, a computer system can comprise one or more central processing units (CPU) and / or graphics processing units (GPU), input means, output means and data storage, which can 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 can comprise RAM, disk drives, solid-state disks or other computer readable media. The computer system can comprise 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 can consist of or comprise a cloud computer. As used herein, the term "computer readable storage medium" 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, magnetic tape; optical storage media such as optical discs or CD-ROMs; electrical storage media such as memory, including RAM, ROM and flash memory; hybrids and combinations of the above such as magnetic / optical storage media. Figure 1 shows an embodiment of a system for implementing the methods described herein. 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 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: methylation sequencing data, training data, parameters (such as e.g. parameters of a machine learning model used to predict whether cancer is present in a cfDNA sample, 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 of predicting whether a subject has a cancer, 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 predicting whether a subject has a cancer, as described herein. In such cases, the remote computing device may also be configured to send the result of the method of predicting whether a subject has a cancer 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 methylation sequencing data from patient samples, for example a blood sample (for example a dried blood spot sample or a pin-prick blood sample), a plasma sample, a urine sample, a tear sample, a saliva sample, a final needle aspirate (FNA) sample, a cerebrospinal fluid (CSF) sample or a fecal sample. In some embodiments, the sample may have been subject to one or more preprocessing steps such as DNA purification, fragmentation, labelling, library preparation, target sequence capture (such as e.g. exon capture and / or panel sequence capture). The sequence data acquisition means may be a next generation sequencer or an array reader. Methods The methods described herein are computer implemented unless context indicates otherwise. Indeed, the features of the data are such that the methods described herein are far beyond the capability of the human brain and cannot be performed as a mental act. The methods described herein can 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. Figure 2 is a flow diagram showing, in schematic form, a method of predicting whether a subject has a cancer according to the disclosure. At optional step 10, a cfDNA-containing sample is obtained from a subject. Optionally, the subject is suspected of having, or being at risk of having, a cancer. The sample may be a blood sample (for example a dried blood spot sample or a pin-prick blood sample), a plasma sample, a urine sample, a tear sample, a saliva sample, a final needle aspirate (FNA) sample, a cerebrospinal fluid (CSF) sample or a fecal sample. The sample may have not more than 100 pg, not more than 50 pg or not more than 20 pg of DNA. At optional step 12, the sample is subjected to a single round of size selection. At optional step 14, methylation sequencing is performed of the cfDNA or of a library generated from the cfDNA. Methylation sequencing can include shallow whole methylome sequencing. Methylation sequencing can include bisulfite pyrosequencing (BS-seq) or enzyme-based methods such as NEBNext Enzymatic Methyl-seq (EM-seq™, Vaisvila et al., 2021). The protocol for EM-seq can be found at https: / / www.neb.com / en-gb / products / e7120-nebnext-enzymatic-methyl-seq-kit, and is herein incorporated by reference in its entirety. Briefly, BS-seq uses the sodium bisulfite conversion of cytosine to uracil to differentiate unmethylated cytosines from protected, unaffected, 5-methylcytosine residues in subsequent next-generation sequencing. EM-seq instead uses an enzymatic conversion step to deaminate unmethylated cytosine into uracil prior to sequencing. Methylation can also be detected by methods including mass spectrometry, methylation-specific PCR, Hpall tiny fragment Enrichment by Ligation-mediated PCR Assay (HELP Assay) or Glal hydrolysis and Ligation Adapter Dependent PCR assay (GLAD-PCR assay). EM-seq can comprise the NEBNext Enzymatic Methyl-seq protocol, optionally modified to have a bead incubation time of between 7 minutes and 15 minutes, such as 10 minutes and / or modified to employ a number of PCR cycles in the range 15-20, such as 18 cycles. At step 16, multimodal data is provided comprising a metric of methylation of at least 10 DMBs and a metric of somatic copy number alterations of at least 10 non-overlapping bins of the genome of the subject. At optional steps 16a and 16b this comprises generating said metrics from methylation sequencing data, optionally obtained via one or all of steps 10 to 14. For example, the metric of methylation can comprise the fraction of methylated fragments summed over said at least 10 DMBs, and the metric of somatic copy number alterations may comprise the median absolute deviation (MAD) of copy numbers (e.g. trimmed median absolute deviation (t-MAD) from copy number neutrality (Mouliere et al., Science Translational Medicine, 2018, Vol 10, Issue 466, DOI: 10.1126 / scitranslmed.aat4921) or ichorCNA (Adalsteinsson et al., "Scalable whole-exome sequencing of cell-free DNA reveals high concordance with metastatic tumors." Nat Commun. 2017 Nov 6;8(1):1324. https: / / doi.org / 10.1038 / s41467-017-00965-y). DMBs can be a set of predetermined cancer-specific hypermethylated DMBs, optionally satisfying one or more of the following conditions: (i) the DMB block comprises at least 1, 2, 3, 4 or at least 5 CpGs, wherein all CpGs, if there is more than 1, share substantially the same level of methylation in a given sample; (ii) the DMB is no more than 5000 bp in length; (Hi) the median fraction of methylated fragments mapping to the DMB in a plurality of samples from controls is less than 0.005; and (iv) the 90th quantile fraction of methylated fragments mapping to the DMB in a plurality of samples from cancer cases is at least 0.1 greater than the 90th quantile fraction of methylated fragments mapping to the DMB in a plurality of samples from controls. At step 18, the multimodal data is input into a machine learning classifier. The machine learning classifier may be, for example, a Random Forest, a neural network, a support vector machine, a logistic regression, a naive Bayes or a perceptron. The machine learning classifier has been trained on a labelled training data set comprising corresponding multimodal data for each of: (i) a plurality of subjects known to have had the cancer the time of sample collection ("cases"); and (ii) a plurality of subjects known to not have the cancer at the time of sample collection ("controls"). Optionally, the metric of methylation comprises the fraction of methylated fragments summed over said at least 10 DMBs, and / or the metric of somatic copy number alterations comprises the median absolute deviation (MAD) of copy numbers, optionally wherein the at least 10 non-overlapping bins of the genome of the subject are each 1 Mb in size. The at least 10 DMBs may comprise cancer-specific hypermethylated DMBs, and may be selected from the DMBs set forth in Table 3 or Table 4. The multimodal data for both input and training can include additional features, such as (i) an accumulated signal in tissue-specific hypomethylated blocks feature; (ii) a fragment length ratio feature; and (iii) a fragment 5'-end sequence motif group feature, as further described herein. At step 20, the subject is classified by the classifier into the "cases" class or the "controls" class, i.e. cancer is, or is not, detected in the sample. In optional step 20a, the classifier outputs a numerical probability of cancer. In optional step 22, one or more results of this analysis may optionally be provided to a user through a user interface. In the context of the present disclosure, methylation of DNA refers to the biological process by which methyl groups are added to cytosine bases in a DNA molecule. Methylation of DNA in humans refers to the methylation of cytosine to form 5-methylcytosine (5mc) unless otherwise stated or indicated by context. Methylation of DNA may also refer to methylation of cytosine to form N4-methylcytosine (4mC), or methylation of adenine bases to form N6-methyladenine (6mA). In mammals, DNA methylation is essential for a number of developmental processes. DNA methylation in mammals is almost exclusively found in CpG dinucleotides, typically with methylation of cytosines on both strands. Typically around 75% of CpG dinucleotides are methylated in somatic cells (Tost, 2009). Abberant DNA methylation changes have been detected in several diseases, particularly cancer where genome-wide hypomethylation coincides with gene-specific hypermethylation. Cancerous cells are known to have distinct DNA methylation signatures from healthy tissue. In normal tissue, DNA methylation in promotor regions is a standard epigenetic mechanism to silence gene expression. Hypermethylation refers to the hypermethylation of DNA, an epigenetic process that can affect gene expression. Hypermethylation is defined as an increase in the epigenetic methylation of DNA relative to a reference methylation genome, such as MethBase, MethDB, or NGSmethDB, or a reference sample obtained from a healthy person. Deregulation of DNA hypermethylation is frequently observed in cancer. Hypomethylation refers to the hypomethylation of DNA. Hypomethylation is defined as a decrease in the epigenetic methylation of DNA relative to a reference methylation genome, such as MethBase, MethDB, or NGSmethDB, or a reference sample obtained from a healthy person. Deregulation of DNA hypomethylation is frequently observed in cancer. Hypomethylation is also observed in healthy persons in a tissuespecific manner. In particular, cell type-specific regions of hypomethylation (hypomethylated or unmethylated CpGs and blocks of CpGs) have been identified (Loyfer et al., 2023). As described, for example, in Loyfer et al., 2023, Slieker et al., 2013, and Varley et al., 2013, there are consistent differences in methylation of key genes between human cell lines and tissues. A "sample" as used herein may be a biological fluid, or an extract (e.g. a DNA extract obtained from the subject), from which genomic material can be obtained for methylomic analysis, such as methylation sequencing. The sample contains cell free DNA (cfDNA). The sample may include a plasma sample, a blood sample, a urine sample, a tear sample, a saliva sample, a final needle aspirate (FNA) sample, a cerebrospinal fluid (CSF) sample or a fecal sample, optionally a dried blood spot sample or a pin-prick blood sample, further optionally wherein the cfDNA sample has not more than 100 pg, not more than 50 pg or not more than 20 pg of DNA. Other sample types suitable for use in accordance with the present invention include fine needle aspirates, lymph nodes samples (e.g. aspirates or biopsies), surgical margins, bone marrow or other tissue from a tumour microenvironment, where traces of tumour DNA or cell-free tumour DNA may be found or expected to be found. 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). The sample may be derived from one or more of the above biological samples via a process of enrichment or amplification. A plurality of samples may be taken from a single patient, e.g. serially during a course of treatment. Moreover, a plurality of samples may be taken from a plurality of patients. The sample is preferably from a mammalian subject, more preferably from a human subject. Further, the sample may be transported and / or stored, and collection may take place at a location remote from the methylation sequencing 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 methylation sequencing data acquisition 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). "Patient" as used herein in accordance with any aspect of the present invention is intended to be equivalent to "subject" and specifically includes both healthy individuals and individuals having a disease or disorder (e.g. a proliferative disorder such as a cancer). Preferably, the patient is a human patient. In some cases, the patient is a human patient who has been diagnosed with, is suspected of having or has been classified as at risk of developing, a cancer. The cancer may be ovarian cancer, breast cancer, prostate cancer, gastrointestinal cancer (e.g. colorectal cancer, oesophagus cancer stomach cancer), endometrial cancer (uterus / womb cancer), kidney cancer (renal cell), lung cancer (small cell, non-small cell and mesothelioma), central nervous system cancer including brain cancer (gliomas, astrocytomas, glioblastomas), melanoma (including choroid melanoma and skin cancers), merkel cell carcinoma, clear cell renal cell carcinoma (ccRCC), carcinoma of unknown primary (CUP), lymphoid cancer (such as e.g. lymphoma), small bowel cancers (duodenal and jejunal), leukemia, pancreatic cancer, hepatobiliary tumours, germ cell cancers, bone / soft tissue cancer, head and neck cancers (such as e.g. adenoid cystic carcinoma, ACC), pancreatic cancer, cervical cancer (e.g. Cervical Squamous Cell Carcinoma and endocervical adenocarcinoma, CESC), liver cancer, bladder cancer (such as e.g. bladder carcinoma, BLCA), urinary tract cancer, neuroendocrine tumour (NET), thyroid cancer and sarcomas. For example, the cancer may be any cancer represented in The Cancer Genome Atlas (TCGA) such as LAML (Acute Myeloid Leukemia), ACC (Adrenocortical carcinoma), BLCA (Bladder Urothelial Carcinoma, LGG (Brain Lower Grade Glioma, BRCA (Breast invasive carcinoma), CESC (Cervical squamous cell carcinoma and endocervical adenocarcinoma), CHOL (Cholangiocarcinoma), LCML (Chronic Myelogenous Leukemia), COAD (Colon adenocarcinoma), ESCA (Esophageal carcinoma), GBM (Glioblastoma multiforme), HNSC (Head and Neck squamous cell carcinoma), RICH (Kidney Chromophobe), KIRC (Kidney renal clear cell carcinoma), KIRP (Kidney renal papillary cell carcinoma), LIHC (Liver hepatocellular carcinoma), LUAD (Lung adenocarcinoma), LUSC (Lung squamous cell carcinoma), DLBC (Lymphoid Neoplasm Diffuse Large B-cell Lymphoma), MESO (Mesothelioma), OV (Ovarian serous cystadenocarcinoma), PAAD (Pancreatic adenocarcinoma), PCPG (Pheochromocytoma and Paraganglioma), PRAD (Prostate adenocarcinoma), READ (Rectum adenocarcinoma), SARC (Sarcoma), SKCM (Skin Cutaneous Melanoma), STAD (Stomach adenocarcinoma), TGCT (Testicular Germ Cell Tumors), THYM (Thymoma), THCA (Thyroid carcinoma), UCS (Uterine Carcinosarcoma), UCEC (Uterine Corpus Endometrial Carcinoma), and UVM (Uveal Melanoma). The term "sequence data" refers to methylation sequence data unless context indicates otherwise. Methylation sequence data can include methylation levels for CpG sites in the region of interest, optionally down to single-CpG site resolution. Methylation sequence data can be obtained by or derived from next-generation sequencing methods. When NGS technologies are used, the sequence data may comprise a count or percentage of the number of sequencing reads that indicate methylation at a predetermined site. 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 indicate methylation at a predetermined site, for example by comparison to an appropriate control. Methylation sequencing can include shallow whole methylome sequencing. Methylation sequencing can include bisulfite pyrosequencing (BS-seq) or enzyme-based methods such as NEBNext Enzymatic Methyl-seq (EM-seq™). Briefly, BS-seq uses the sodium bisulfite conversion of cytosine to uracil to differentiate unmethylated cytosines from protected, unaffected, 5-methylcytosine residues in subsequent next-generation sequencing. EM-seq instead uses an enzymatic conversion step to deaminate unmethylated cytosine into uracil prior to sequencing. Methylation can also be detected by methods including mass spectrometry, methylation-specific PCR, Hpall tiny fragment Enrichment by Ligation-mediated PCR Assay (HELP Assay) or Glal hydrolysis and Ligation Adapter Dependent PCR assay (GLAD-PCR assay). EM-seq can comprise the NEBNext Enzymatic Methyl-seq protocol, optionally modified to have a bead incubation time of between 7 minutes and 15 minutes, such as 10 minutes and / or modified to employ a number of PCR cycles in the range 15-20, such as 18 cycles. Sequence data may be mapped to a reference sequence, for example a reference DNA methylation database such as MethBase, MethDB, or NGSmethDB, using methods known in the art. An "anti-cancer agent" as described herein may comprise a therapeutic agent, a tyrosine kinase inhibitor, a checkpoint inhibitor, a checkpoint stimulator, a chemotherapeutic agent, an immunotherapeutic agent, a platinum agent, an alkylating agent, a taxane, a nucleoside analog, an antimetabolite, a topoisomerase inhibitor, an anthracycline, a vinca alkaloid, or any combination thereof. 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. As used herein "a labelled training data set comprising corresponding multimodal data" refers to training examples (e.g. feature vectors and associated class labels) that match the features to be used as input features for the classification task. As an example, where the multimodal data comprises as feature types a metric of methylation of each of at least 10 differentially methylated blocks of homogeneous methylation ("DMBs") in a cell-free DNA (cfDNA)-containing sample obtained from the subject and a metric of somatic copy number alterations of at least 10 non-overlapping bins of some or all of the genome of the subject in a cell-free DNA (cfDNA)-containing sample obtained from the subject, the labelled training data set on which the machine learning classifier has been trained will have comprised for substantially all training examples the same feature types, i.e. the same metric of methylation of each of at least 10 DMBs and the same metric of somatic copy number alterations. The labelled training data set comprises training examples from each of (i) a plurality of subjects known to have had the cancer the time of sample collection ("cases"); and (ii) a plurality of subjects known to not have the cancer at the time of sample collection ("controls"). The features disclosed in the foregoing description, or in the following claims, or in the accompanying drawings, expressed in their specific forms or in terms of a means for performing the disclosed function, or a method or process for obtaining the disclosed results, as appropriate, may, separately, or in any combination of such features, be utilised for realising the invention in diverse forms thereof . While the invention has been described in conjunction with the exemplary embodiments described above, many equivalent modifications and variations will be apparent to those skilled in the art when given this disclosure. Accordingly, the exemplary embodiments of the invention set forth above are considered to be illustrative and not limiting. Various changes to the described embodiments may be made without departing from the spirit and scope of the invention. For the avoidance of any doubt, any theoretical explanations provided herein are provided for the purposes of improving the understanding of a reader. The inventors do not wish to be bound by any of these theoretical explanations . Any section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described. Throughout this specification, including the claims which follow, unless the context requires otherwise, the word "comprise" and "include", and variations such as "comprises", "comprising", and "including" will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps. It must be noted that, as used in the specification and the appended claims, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from "about" one particular value, and / or to "about" another particular value. When such a range is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when values are expressed as approximations, by the use of the antecedent "about," it will be understood that the particular value forms another embodiment. The term "about" in relation to a numerical value is optional and means for example + / - 10%. Units, prefixes, and symbols are denoted in their Systeme International des Unites (SI) accepted form. Examples EXAMPLES - Introduction Loyfer et al. 2023 performed the first genome-wide characterisation of the human cell methylome. The authors used paired-end, 150 bp-long reads at an average sequencing depth of 30x (6.62x or greater) on fluorescent activated cell sorter (FACS)-purified populations of 77 primary cell types obtained from freshly dissociated adult healthy tissues. The authors then coalesced methylation patterns across the entire genome into a total of 7,104,162 non-overlapping, continuous blocks of homogeneously methylated CpG sites, before using this information to investigate variation in methylation patterns across cell and tissue types. The authors found "distinctive changes between cell types in a block-like manner", which were used to facilitate development of methylation biomarkers to identify the cellular origin of circulating cfDNA fragments . Importantly, the work of Loyfer et al. was all performed in samples obtained from healthy human donors. Loyfer et al. did not investigate differences in differentially methylated CpG sites (DMCs) or differentially methylated blocks of homogenous methylation (DMBs) in tumour samples. In the present disclosure, the inventors investigate the presence of DMBs between healthy and cancerous tissue. They then use the identified DMBs in a method of predicting cancer in tumour DNA or circulating tumour DNA samples. EXAMPLE 1 - Panel Design The inventors first developed their own pipeline for the identification of differentially methylated blocks of homogenous methylation (DMBs) in a target cancer type. The concept for the pipeline was to use methylation sequencing to first identify single differentially methylated CpGs (DMCs) in tumour vs. healthy samples. The inventors then used the tendency of neighbouring CpG sites to share the same methylation states to segment the genome into DMBs. The DMBs that that contained at least one DMC between tumour and healthy samples were retained in the dataset, with the rationale that DNA methylation occurs in a segmental fashion resulting in a high likelihood that neighbouring CpGs will have a similar methylation state (Loyfer et al. 2023). The first step, panel design, included using Illumina 450K or EPIC array data to identify single differentially methylated CpGs (DMC) for a target cancer type. In the present disclosure, the chosen cancer type was ovarian, however the methods disclosed herein can be readily applied to any chosen cancer type. Methods The data used for analysis includes Illumina Methylation 450K idat files from The Cancer Genome Atlas (TCGA) legacy site of primary tumour and adjacent normal samples from 33 cancers, additional ovarian cancer data was from GSE72021 (n=221). Illumina 450K or EPIC idat files for cell types isolated from primary tissue were from GSE110555 (neutrophil, monocyte, B cell, CD8 T cells, NK cells all n=6, CD4 T cells n=5), GSE122126 (hepatocyte n=3, vascular endothelial cells n=2), and GES63409 (RBC progenitors n=25, WBC progenitors n=5). Ovarian cancer data was supplemented with the data series GSE72021. Cancer types are summarised in Table 1. Cancer type TCGA abbreviation Normal tissue Tumour tissue LUMP>=50%, not FFPE Bladder Urothelial Carcinoma BCLA 21 414 341 Breast invasive carcinoma BRCA 94 (all have LUMP50%) 773 725 Diffuse large B cell lymphoma DLBCL 0 48 48 Esophageal cancer ESCA 16 178 165 Head-Neck Squamous Cell Carcinoma HNSC 50 519 423 Acute myeloid leukemia LAML 0 194 194 Lung adenocarcinoma LUAD 32 462, 2 recurrent 390 Lung squamous cell carcinoma LUSC 42 362 296 Ovarian serous cystadenocarcinoma OV 0 10, 221 GSE72021 9, 210 GSE72021 Prostate adenocarcinoma PRAD 50 496 492 23 other cancer types ACC, CESC, CHOL, COAD, GBM, KICH, KIRC, KIRP, LGG, LIHC, MESO, PAAD, PCPG, READ, SARC, SKCM, STAD, TGCT, THCA, THYM, UCEC, UCS, UVM All 575 Table 1. Cancer types used in analysis and numbers of samples before and after filtering for LUMP>= 50%. ACC - Adrenocortical carcinoma; CESC - Cervical squamous cell carcinoma and endocervical adenocarcinoma; CHOL - Cholangiocarcinoma; 5 COAD - Colon adenocarcinoma; GBM - Glioblastoma multiforme; RICH -Kidney Chromophobe; KIRC - Kidney renal clear cell carcinoma; KIRP -Kidney renal papillary cell carcinoma; LGG - Brain Lower Grade Glioma; LIHC - Liver hepatocellular carcinoma; MESO - Mesothelioma; PAAD - Pancreatic adenocarcinoma; PCPG - Pheochromocytoma and Paraganglioma; READ - Rectum adenocarcinoma; SARC - Sarcoma; SKCM - Skin Cutaneous Melanoma; STAD - Stomach adenocarcinoma; TGCT - Testicular Germ Cell Tumors; THCA - Thyroid carcinoma; THYM - Thymoma; UCEC - Uterine Corpus Endometrial Carcinoma; UCS - Uterine Carcinosarcoma; UVM -Uveal Melanoma. The idat files were converted to [3 values using minfi as per Moss et al. 2018, incorporated herein by reference. [3 = 0 means fully unmethylated, and [3 = 1 means fully methylated. CpG sites represented by less than 3 beads, detection p-value >0.01, mapped to sex chromosome, or located at single nucleotide polymorphism (SNP) with MAE >0.01 were removed (n=5872 CpG). All data was normalized against an arbitrarily chosen reference sample of non-cancer associated fibroblasts (GSM2309161). Lastly, since this analysis includes samples analyzed with either Illumina 450K or EPIC array, the inventors looked to identify and remove sites with low reproducibility between the platforms. To this end, data was collected from samples analyzed on both platforms: 15 samples from GSE86833, 12 samples from GSE92580, and one sample from Moss et al., 2018 (hepatocytes). For each overlapping CpG, the inventors then calculated the median absolute error (MAE) between the 450K samples and the corresponding EPIC samples, and removed 37,747 CpGs with an MAE >0.05." For all samples, 386,047 CpG were carried forward in analysis. Samples that were found to have white blood cell contamination were removed by calculating the leukocyte unmethylation for purity (LUMP) score (Aran, Sirota, and Butte 2015) and retaining samples with LUMP >= 50%. LUMP estimations are the average methylation levels of 44 CpG sites divided by 0.85, the CpG sites identified by Aran, Sirota and Butte as being consistently unmethylated across 10 immune cells, but averagely methylated in 21 analysed cancer types. The inventors then conducted multiple comparisons between groups of cancer and normal samples, according to Table 2. For each CpG and each comparison (e.g NAI is to compare target cancer cells + target normal tissue vs Blood controls), the inventors calculated the Area Under the Curve (AUG) of the Receiver Operating Characteristic (ROC) curve using the R package pROC based on the p values of the two comparison groups and difference in the mean p values of the two comparison groups, namely Ap (i.e. (mean p of target cancer cells and target normal tissue) - (mean p of blood controls for NAI)). Blood controls included cell types that are known to be found in cell-free DNA (cfDNA) (blood cells (GSE110555, GSE63409); hepatocytes (GSE122126), and vascular endothelial cells (GSE122126)). CpGs were filtered based on AUG and Ap, where Ap is the difference in p value between Group 1 and Group 2 (Table 2). ID Group1 Group2 NAI Target cancer cells + target normal tissue Blood controls NA2 Target normal tissue Blood controls NA3 Target cancer cells Blood controls NB2 Target cancer cells Target normal tissue NB3 Target cancer cells Other cancers NB4 Target cancer cells + target normal tissue Other cancers + normal tissues NB5 Target normal tissue Other normal tissues Table 2. Groups used for comparisons to identify differentially methylated CpGs (DMCs) . EXAMPLE 2 - An optimised protocol for methylation sequencing Starting from the NEBNext Enzymatic Methyl-seq®, protocol ("EM-seq", New England Biolabs #E7120S / L), available from https: / / www.neb.com / en-gb / products / e7120-nebnext-enzymatic-methyl-seq-kit and incorporated herein by reference in its entirety, the inventors developed an optimised method for extracting methylation signals from low quantities of DNA, including DNA from Dried Blood Spots (DBS). Methods The optimised protocol uses an enzymatic conversion method initially developed by NEB (Vaisvila et al., 2021), and improves it to work with very low quantities (5-15pg) of cfDNA from DBS. Figure 3 provides a schematic comparison of Sodium Bisulfite Conversion and EM-seq. In brief, Sodium bisulfite treatment of DNA results in the deamination of cytosines into uracils, however the modified forms of cytosine (5mC and 5hmC) are not deaminated. Therefore, the preference of bisulfite to chemically deaminate cytosines enables the methylation status of cytosines to be determined. When bisulfite treated DNA is PCR amplified, uracils are replaced by thymines and the 5mC / 5hmC are replaced by cytosines. Once sequenced, unmethylated cytosines are represented by thymines and 5mC and 5hmC are represented by cytosines. By comparing sequences to nonconverted genomes the appropriate methylation status can be assessed. EM-seq is a two-step enzymatic conversion process to detect modified cytosines. The first step uses TET2 and an oxidation enhancer to protect modified cytosines from downstream deamination. TET2 enzymatically oxidizes 5mC and 5hmC through a cascade reaction into 5-carboxycytosine [5-methylcytosine (5mC) =>5-hydroxymethylcytosine (5hmC) => 5-formylcytosine (5fC) => 5- carboxycytosine (5caC)]. This protects 5mC and 5hmC from deamination. 5hmC can also be protected from deamination by glucosylation to form 5ghmc using the oxidation enhancer. The second enzymatic step uses APOBEC to deaminate C but does not convert 5caC and 5ghmC. The resulting converted sequence can be analyzed like bisulfite-treated DNA. Typical aligners used to analyze data include but are not limited to Bismark and BWAMeth. The standard workflow enabled by the NEB protocol enables methylation detection from inputs ranging between 10 ng-200 ng. The inventors implement a number of optimised steps to facilitate the use of EM-seq with very low quantities (5-15 pg) of cfDNA from DBS. First, the inventors implemented a single round of size selection, in contrast to two rounds of size selection previously used for extraction of cfDNA from blood spots. The inventors tested library preparation with two rounds of size selection, one round, and no size selection. Size selection was implemented using SPRIselect beads from Beckman Coulter (AMPure XP Bead-Based Reagent; catalog number: A63881). The size selection protocol is a modified version of the standard manufacturer's protocol (SPRIselect User Guide B24965AA October 2012). The key steps of the modified protocol are as follows: 1.Thoroughly shake the BEADS bottle to resuspend the SPRI beads. Add the required volume of BEADS for the desired ratio to the sample. Volume of sample * ratio = volume of BEADS. Example: 50 pL * 0.7x ratio = 35 pL of BEADS 2.Mix the total reaction volume by pipetting 10 times and incubate at RT for 1 minute. OR Vortex for 1 minute at an appropriate speed until homogenous (depending on labware and total volume). Insufficient mixing of sample and BEADS will lead to inconsistent size selection results. Make sure to mix well. 3. Place the reaction vessel on an appropriate magnetic stand or plate and allow the SPRI beads to settle to the magnet. Settle times will vary; a higher initial sample volume, higher BEADS ratio or weaker magnets will require a longer settle time. 4.Transfer the clear supernatant, which contains the Right Side Size Selected sample, to a new reaction vessel. The reaction vessel with the remaining beads can be discarded. Care should be taken not to aspirate more than a trace amount of beads during this step as the undesired larger fragment sizes are associated with the beads. Significant bead transfer will cause tailing into the larger size range. 5.Add the required volume of BEADS, using the calculation below, to the supernatant from Step 4 above. This will bind the fragments in the supernatant to the new SPRI beads. Sample Volume pL * (1.8x - the initial ratio) = volume of BEADSExample: 50 pL * (1.8 - 0.7) = 55 pL of BEADS 6.Perform the following: a.Mix the total reaction volume by pipetting 10 times and incubate at RT for 1 minuteORvortex for 1 minute at an appropriate speed until homogenous (depending on labware and total volume). Insufficient mixing of sample and BEADS will lead to inconsistent size selection results . b.Place the reaction vessel on an appropriate magnetic stand or plate and allow the SPRI beads to settle to the magnet. Settle times will vary; a higher initial sample volume, higher BEADS ratio or weaker magnets will require a longer settle time. c.Remove and discard the clear supernatant. Care should be taken not to aspirate more than a trace amount of beads during this step, as the desired library is associated with the beads. Significant bead loss will result in reduced yield. 7.With the reaction vessel still on the magnet, add 180 pL of 85% ethanol (non-denatured) and incubate at RT for 30 seconds. Remove and discard the ethanol supernatant. Care should be taken not to aspirate more than a trace amount of beads during this step, as the desired library is associated with the beads. Significant bead loss will result in reduced yield. 8.To elute the sample: a.Remove the reaction vessel from the magnet and add >20 pL of molecular biology grade water or standard buffer solution such as Tris or TE . Elution volume should be large enough so that the liquid level is high enough for the beads to settle to the magnet. b.Mix the total elution volume by pipetting 10 times to resuspend the beads and incubate at RT for 1 minute OR vortex for 1 minute at an appropriate speed until homogenous (depending on labware and total volume). c.Place the reaction vessel on an appropriate magnetic stand or plate and allow the SPRI beads to settle to the magnet. Settle times will vary; a higher elution volume or weaker magnets will require a longer settle time. 9.Transfer the eluate (size selected sample) to an appropriate storage vessel. Secondly, the inventors increased the incubation time for all bead incubations from 5 minutes to 10 minutes, increased the number of PCT cycles from 8 to 18, and avoided bead carry-over which affects the methylation conversion. During the oxidation clean-up step, bead carryover was carefully avoided to maintain high conversion efficiency, as bead carryover was found to interfere with this process. The inventors implemented their optimised protocol on a serial dilution of plasma cfDNA from 1 ng to 15 pg. Results The inventors surprisingly found that the library from a single round of size selection is more representative of cfDNA peaks than with the standard two rounds of size selection. While no size selection resulted in a good library yield, the libraries as expected contained higher molecular weight genomic DNA. DNA size was assessed using Agilent Tapestation; results of the three libraries are shown as TapeStation traces in Figure 4. The inventors demonstrated the preparation of libraries was successful with as little as 15 pg DNA input. The copy number profiles of these libraries were consistent across serial dilutions (Figure 5). Compared to previous library preparation on plasma using the standard NEB protocol, the inventors noted a two-fold increase in library yield after optimisation. The modified optimisation facilitated the preparation of libraries from very small amounts of input DNA obtained from DBS. Discussion While Vaisvila et al. discuss the preparation of libraries from 100 pg genomic DNA, the authors did not specify the library yield, only showing that these libraries were similar to libraries made from higher input in terms of methylation signatures and patterns. By implementing their optimised methods, the inventors demonstrate library preparation from 15 pg or less cfDNA, and showed that the methylation profile, patterns of copy number aberrations (Figure 5), and fragmentomic features were similar to plasma cfDNA libraries at higher input concentrations. EXAMPLE 3 - A first machine learning classification model based on the targeted capture data and somatic copy number alterations The inventors used the modified methylation sequencing protocol described in Example 2 to process a set of cfDNA samples and generate shallow whole-methylome sequencing data. The sequencing data was then processed to identify DMBs specific to target cancer types, with a significant difference from healthy controls (feature 1). The inventors continued to calculate a median absolute deviation (MAD) of copy number for each sample (feature 2), and used the two features to train and test a random forest machine learning classification model. Methods The inventors collected cfDNA samples from patients with ovarian cancer (n = 39) and healthy controls (n = 41). The samples were processed using the modified NEB protocol as described in Example 2 to generate sequencing libraries. The inventors sequenced a portion of each library before target capture to do shallow whole-methylome sequencing (sWMS, mean coverage 0.5x, max lx). The libraries were then enriched using a target capture (TC) panel designed, as described in Example 1, to capture DMBs specific to ovarian and breast cancers before deep sequencing these regions (mean coverage llOOx). After sequencing, adapter trimmed reads were aligned to human genome assembly GRCh37 (NCBI RefSeq assembly GCF 000001405.13) using bwa-meth (Pedersen et al., 2014). Then from these TC alignments, the inventors extracted the counts of methylated and unmethylated cfDNA fragments overlapping each DMB in each sample using wgbstools (Loyfer et al., 2024). Next, the reference genome was divided into equal 1 Mb nonoverlapping bins and the copy number value per bin was calculated using QDNAseq R package (Scheinin et al., 2014) . Then, a median absolute deviation (MAD) of copy numbers was calculated for each sample. Further, the inventors used DeSeq2 R package (Love et al., 2014) to analyse counts of methylated fragments over DMBs comparing cancer cases vs controls. Then the set was randomly split into training (n = 59) and test (n = 21) subsets, accounting for the balance of classes (Figure 6). The inventors used fractions of methylated cfDNA fragments in the DMBs and MAD values for training set samples, as the two features to build and tune a random forest (RF) machine learning (ML) classification model. The models were built within tidymodels framework which provides an R interface for interaction with various packages. The Random Forest models are built with the ranger package (Wright and Ziegler, 2017). The inventors did lOx cross-validation on the Training Set to tune the model and select the one with the best performance. Hyperparameters of the final Model 1 are: 1) number of trees = 1000; 2) number of randomly selected predictors = 14; 3) minimal node size = 5 25. Predicted classes with corresponding probabilities from Model 1 were compared with ground truth sample types, sample type predicted by ichorCNA, a copy-number based tool (Adalsteinsson et al., 2017), and tumour fraction estimated from TP53 mutation with ddPCR analysis. The model was validated with 10-fold cross validation, and final testing 10 with the hold-out test set. Results The inventors identified 73 DMBs with significant difference between the two groups of cancer cases and controls (p <0.00001 and log2 fold 15 change of 7 or higher). The hgl9 genomic coordinates (chromosome, start and end position) of each of the 73 DMBs are set forth in Table 3 . DMB hgl9 genomic coordinates 1 chrl:10571475-10571726 2 chrl:110610858-110610955 3 chrl:110611373-110611555 4 chrl:110612002-110612071 5 chrl:115641978-115642432 6 chrl:116044052-116044437 7 chrl:119529823-119530007 8 chrl:119530443-119530682 9 chrl:119531998-119532085 10 chrl:119532116-119532234 11 chrl:119535319-119535640 12 chrl:119535908-119535963 13 chrl:119536218-119536298 14 chrl:151694320-151694360 15 chrl:151810772-151811181 16 chrl:151811354-151811523 17 chrl:155505966-155506057 18 chrl:156130726-156131011 19 chrl:156215470-156215847 20 chrl:156357771-156357971 21 chrl:160053850-160054190 22 chrl:160681530-160682009 23 chrl:161275248-161275683 24 chrl:161275765-161276205 25 chrl:165322069-165322122 26 chrl:165322127-165322260 27 chrl:165323668-165323733 28 chrl:165326220-165326434 DMB hgl9 genomic coordinates 29 chrl:16553470-16553557 30 chrl:166853526-166853620 31 chrl:169396363-169396980 32 chrl:170629353-170629520 33 chrl:171810257-171810377 34 chrl:180202465-180202549 35 chrl:18956827-18956954 36 chrl:192544716-192544893 37 chrl:197888457-197888513 38 chrl:197888847-197889067 39 chrl:200842867-200842921 40 chrl:203830112-203830201 41 chrl:20810427-20810652 42 chrl:209929907-209930128 43 chrl:210419938-210419995 44 chrl:212838634-212839041 45 chrl:215255076-215255320 46 chrl:217312926-217313067 47 chrl:218098448-218098581 48 chrl:235813969-235814135 49 chrl:235814145-235814312 50 chrl:245474268-245474762 51 chrl:32237785-32238050 52 chrl:35351185-35351617 53 chrl:35351627-35351760 54 chrl:36042904-36043173 55 chrl:38412519-38412745 56 chrl:42149528-42149892 DMB hgl9 genomic coordinates 57 chrl:44872851-44873293 58 chrl:46957013-46957158 59 chrl:46958685-46958811 60 chrl:4714008-4714117 61 chrl:47691700-47692027 62 chrl:47695461-47695671 63 chrl:47696185-47696312 64 chrl:47696332-47696567 65 chrl:47697960-47698033 Table 3 — Genomic coordinates of 73 identified DMBs DMB hgl9 genomic coordinates 66 chrl:47701164-47701300 67 chrl:47882676-47882863 68 chrl:47910843-47911020 69 chrl:50881107-50881192 70 chrl:53527382-53527601 71 chrl:54821886-54821992 72 chrl:62660606-62660691 73 chrl:63784717-63784900 After splitting into training (n = 59) and test (n = 21) subsets, the data was used to build and tune the classification model. Performance of the model is shown as a heatmap in Figure 7, comparing sample type predicted by Model 1 disclosed herein ('Predicted') against ground truth sample values ('Type'), sample type predicted by ichorCNA, a copy-number based tool ('Predicted ichor', Adalsteinsson et al., 2017), and against tumour fraction estimated from TP53 mutation with ddPCR analysis ( 'tp53 TF') . The sample type prediction by Model 1 was associated with a cancer probability (0-1) which may be expressed as a percentage . In 10-fold cross validation, mean area under the curve (AUC) of the final model was 0.944. The inventors then evaluated performance of the final model in the held-out test set, where AUC was 0.955. EXAMPLE 4 - A second machine learning classification model based on multimodal data from shallow whole methylome sequencing data (sWMS) The inventors developed a second ML classification model informed by TC data analysis but based on sWMS data only. The rationale is that the sWMS workflow is more straightforward and standardisable across laboratories, it does not require enrichment with capture probes and is less expensive. Method The inventors used five features to build a second, multi-modal classification model. 1. Accumulated signal in ovarian cancer-specific hypermethylated DMBs. The inventors identified 716 ovarian cancer-specific hypermethylated DMBs using the panel as described in Example 1. The hgl9 genomic coordinates (chromosome, start DMBs are set forth in Table 4. and end position) of each of the 716 DMB hgl9 genomic coordinate 1 chrl:10571475-10571726 2 chrl:110610858-110610955 3 chrl:110611373-110611555 4 chrl:110612002-110612071 5 chrl:115641978-115642432 6 chrl:116044052-116044437 7 chrl:119529823-119530007 8 chrl:119530443-119530682 9 chrl:119531998-119532085 10 chrl:119532116-119532234 11 chrl:119535319-119535640 12 chrl:119535908-119535963 13 chrl:119536218-119536298 14 chrl:151694320-151694360 15 chrl:151810772-151811181 16 chrl:151811354-151811523 17 chrl:155505966-155506057 18 chrl:156130726-156131011 19 chrl:156215470-156215847 20 chrl:156357771-156357971 21 chrl:160053850-160054190 22 chrl:160681530-160682009 23 chrl:161275248-161275683 24 chrl:161275765-161276205 25 chrl:165322069-165322122 26 chrl:165322127-165322260 27 chrl:165323668-165323733 28 chrl:165326220-165326434 29 chrl:16553470-16553557 30 chrl:166853526-166853620 31 chrl:169396363-169396980 32 chrl:170629353-170629520 33 chrl:171810257-171810377 34 chrl:180202465-180202549 35 chrl:18956827-18956954 36 chrl:192544716-192544893 37 chrl:197888457-197888513 38 chrl:197888847-197889067 39 chrl:200842867-200842921 40 chrl:203830112-203830201 41 chrl:20810427-20810652 42 chrl:209929907-209930128 43 chrl:210419938-210419995 44 chrl:212838634-212839041 45 chrl:215255076-215255320 46 chrl:217312926-217313067 47 chrl:218098448-218098581 48 chrl:235813969-235814135 49 chrl:235814145-235814312 50 chrl:245474268-245474762 51 chrl:32237785-32238050 52 chrl:35351185-35351617 53 chrl:35351627-35351760 54 chrl:36042904-36043173 55 chrl:38412519-38412745 DMB hgl9 genomic coordinate 56 chrl:42149528-42149892 57 chrl:44872851-44873293 58 chrl:46957013-46957158 59 chrl:46958685-46958811 60 chrl:4714008-4714117 61 chrl:47691700-47692027 62 chrl:47695461-47695671 63 chrl:47696185-47696312 64 chrl:47696332-47696567 65 chrl:47697960-47698033 66 chrl:47701164-47701300 67 chrl:47882676-47882863 68 chrl:47910843-47911020 69 chrl:50881107-50881192 70 chrl:53527382-53527601 71 chrl:54821886-54821992 72 chrl:62660606-62660691 73 chrl:63784717-63784900 74 chrl:63785287-63785398 75 chrl:63785507-63785554 76 chrl:63785782-63785975 77 chrl:63795884-63796035 78 chrl:66999481-66999710 79 chrl:67218004-67218330 80 chrl:75595870-75596070 81 chrl:75601945-75602239 82 chrl:75602283-75602413 83 chrl:76082539-76082748 84 chrl:79472261-79472508 85 chrl:87617229-87617417 86 chrl:87617693-87617917 87 chrl:88921473-88921698 88 chrl:89663866-89664034 89 chrl:91182528-91182648 90 chrl:91182677-91182877 91 chrl:91185366-91185442 92 chrl:91192311-91192467 93 chrl:91194806-91195165 94 chrl:92952467-92952609 95 chrl:997769-997986 96 chrlO:101290162-101290438 97 chrlO:102589365-102589472 98 chrlO:102590209-102590330 99 chrlO:102894117-102894392 100 chrlO:102894802-102895055 101 chrlO:102899256-102899588 102 chrlO:102997363-102997418 103 chrlO:102998592-102998793 104 chrlO:103043756-103043992 105 chrlO:103043997-103044282 106 chrlO:103536314-103536399 107 chrlO:110226368-110226416 108 chrlO:111216824-111217035 109 chrlO:11187452-11187765 110 chrlO:118891628-118891717 DMB hgl9 genomic coordinate 111 chrlO 118892211-118892310 112 chrlO 118894110-118894311 113 chrlO 118899718-118899869 114 chrlO 119291989-119292069 115 chrlO 119307931-119308210 116 chrlO 119310391-119310810 117 chrlO 119494602-119494700 118 chrlO 124902840-124902882 119 chrlO 129534924-129535139 120 chrlO 129535159-129535349 121 chrlO 134598304-134598353 122 chrlO 16562002-16562202 123 chrlO 22541945-22542121 124 chrlO 22634161-22634219 125 chrlO 22634324-22634450 126 chrlO 22765821-22765856 127 chrlO 28035850-28035951 128 chrlO 43393673-43393795 129 chrlO 49842443-49842768 130 chrlO 60552724-60552918 131 chrlO 72336097-72336290 132 chrlO 8084623-8085073 133 chrlO 8089664-8089813 134 chrlO 81370414-81370783 135 chrlO 88149112-88149283 136 chrlO 94450826-94450867 137 chrlO 94451625-94451654 138 chrlO 94834668-94834873 139 chrlO 98479828-98479992 140 chrll 102702269-102702390 141 chrll 114186909-114187080 142 chrll 115530919-115531229 143 chrll 118084920-118085215 144 chrll 119613006-119613247 145 chrll 121460846-121461086 146 chrll 128565519-128566010 147 chrll 132813596-132814004 148 chrll 20177768-20177884 149 chrll 20181676-20181769 150 chrll 20618054-20618246 151 chrll 31822261-31822312 152 chrll 31824237-31824375 153 chrll 31825792-31825834 154 chrll 31826558-31826674 155 chrll 31826928-31827088 156 chrll 31841506-31841530 157 chrll 31843855-31843937 158 chrll 31846826-31846850 159 chrll 31848519-31848744 160 chrll 32461212-32461422 161 chrll 36509698-36509911 162 chrll 44325196-44325306 163 chrll 44330903-44331093 164 chrll 44332604-44332706 165 chrll 44332972-44333121 166 chrll 58342886-58343319 167 chrll 60047979-60048222 168 chrll 61062685-61063044 169 chrll 627008-627486 DMB hg!9 genomic coordinate 170 chrll 64479011-64479152 171 chrll 65816428-65816551 172 chrll 69704651-69704876 173 chrll 71952337-71952419 174 chrll 7487141-7487182 175 chrll 77160107-77160348 176 chrll 86085694-86085758 177 chrll 86085830-86085933 178 chrll 8615304-8615357 179 chrll 9622922-9623179 180 chrl2 10103711-10103978 181 chrl2 102036375-102036461 182 chrl2 108169020-108169182 183 chrl2 122096638-122096676 184 chrl2 130388743-130388889 185 chrl2 15113093-15113499 186 chrl2 25204975-25205469 187 chrl2 47224761-47224998 188 chrl2 49390661-49390926 189 chrl2 49391281-49391438 190 chrl2 52401356-52401554 191 chrl2 52652205-52652424 192 chrl2 54132189-54132233 193 chrl2 54398725-54398946 194 chrl2 54409139-54409337 195 chrl2 54409450-54409737 196 chrl2 54954086-54954558 197 chrl2 57618852-57619003 198 chrl2 58021140-58021347 199 chrl2 58021713-58021918 200 chrl2 62584133-62584514 201 chrl2 65066729-65066844 202 chrl2 6664948-6665388 203 chrl2 8025451-8025667 204 chrl2 8088631-8088955 205 chrl2 8276276-8276592 206 chrl2 85667288-85667407 207 chrl2 95942900-95943032 208 chrl2 9913086-9913396 209 chrl2 99139511-99140000 210 chrl3 100608166-100608264 211 chrl3 100640958-100641410 212 chrl3 112711814-112712099 213 chrl3 112721028-112721488 214 chrl3 112722275-112722401 215 chrl3 112722949-112723164 216 chrl3 112759112-112759515 217 chrl3 28498878-28499034 218 chrl3 28502394-28502560 219 chrl3 46402408-46402608 220 chrl3 46756193-46756537 221 chrl3 51417647-51417687 222 chrl3 53312966-53313269 223 chrl3 53422629-53422739 224 chrl3 58207803-58207949 225 chrl3 72650440-72650541 226 chrl3 79170042-79170239 227 chrl3 79170249-79170386 228 chrl3 79175801-79176291 DMB hgl9 genomic coordinate 229 chrl3 84453564-84453707 230 chrl3 95202330-95202504 231 chrl3 96204794-96205242 232 chrl4 24804124-24804396 233 chrl4 26673957-26674086 234 chrl4 29244278-29244338 235 chrl4 33401961-33402181 236 chrl4 36973642-36973853 237 chrl4 37123744-37123793 238 chrl4 37130122-37130330 239 chrl4 37136283-37136299 240 chrl4 38724435-38724686 241 chrl4 50527872-50528067 242 chrl4 52536091-52536329 243 chrl4 57274658-57274764 244 chrl4 57276039-57276185 245 chrl4 57276257-57276338 246 chrl4 57278563-57278718 247 chrl4 57278729-57278791 248 chrl4 60952348-60952800 249 chrl4 60973356-60973679 250 chrl4 60976285-60976482 251 chrl4 60977645-60977899 252 chrl4 61109911-61110044 253 chrl4 61118751-61118795 254 chrl4 70653714-70654140 255 chrl4 95233855-95234127 256 chrl4 95237623-95237746 257 chrl4 95239506-95239645 258 chrl4 95239698-95239791 259 chrl4 95241907-95242107 260 chrl5 32607544-32607958 261 chrl5 37174376-37174662 262 chrl5 37180615-37180642 263 chrl5 37403153-37403444 264 chrl5 45996608-45996677 265 chrl5 52408383-52408547 266 chrl5 53075377-53075451 267 chrl5 72072589-72073004 268 chrl5 73611218-73611506 269 chrl5 76627568-76627615 270 chrl5 79502065-79502138 271 chrl5 85360569-85360678 272 chrl5 85360691-85360768 273 chrl5 89952454-89952506 274 chrl5 96907756-96907984 275 chrl6 11455815-11456239 276 chrl6 21294883-21294982 277 chrl6 29887967-29888165 278 chrl6 31580254-31580612 279 chrl6 31580911-31581117 280 chrl6 51183533-51183770 281 chrl6 54404652-54404956 282 chrl6 57571379-57571605 283 chrl6 68823690-68823828 284 chrl6 85041486-85041828 285 chrl7 1881115-1881196 286 chrl7 29636688-29637105 287 chrl7 29647977-29648324 DMB hg!9 genomic coordinate 288 chrl7 29648447-29648812 289 chrl7 32569640-32569861 290 chrl7 35299596-35299878 291 chrl7 35299913-35300021 292 chrl7 36103092-36103590 293 chrl7 36105064-36105518 294 chrl7 36714772-36715244 295 chrl7 42287754-42288095 296 chrl7 42988996-42989106 297 chrl7 4636643-4636831 298 chrl7 46711051-46711447 299 chrl7 46827626-46827792 300 chrl7 47301386-47301842 301 chrl7 47547314-47547416 302 chrl7 48948717-48949122 303 chrl7 5000955-5001205 304 chrl7 5019493-5019652 305 chrl7 56406035-56406295 306 chrl7 59529367-59529806 307 chrl7 59532142-59532218 308 chrl7 59532519-59532593 309 chrl7 70216139-70216403 310 chrl7 72270302-72270445 311 chrl7 7455982-7456224 312 chrl7 8481010-8481126 313 chrl8 12254452-12254729 314 chrl8 31738934-31739318 315 chrl8 31739358-31739453 316 chrl8 44618528-44618869 317 chrl8 44787490-44787603 318 chrl8 55019757-55020041 319 chrl8 70534817-70535073 320 chrl8 70536245-70536545 321 chrl8 74962515-74962853 322 chrl9 10397776-10397840 323 chrl9 11354122-11354279 324 chrl9 12978367-12978806 325 chrl9 12996417-12996796 326 chrl9 12997318-12997518 327 chrl9 13983550-13983726 328 chrl9 13983739-13984188 329 chrl9 15288310-15288701 330 chrl9 15343235-15343346 331 chrl9 15343350-15343454 332 chrl9 15344325-15344528 333 chrl9 2252969-2253455 334 chrl9 2253771-2253855 335 chrl9 2282439-2282730 336 chrl9 22990055-22990180 337 chrl9 2302467-2302599 338 chrl9 35068455-35068655 339 chrl9 38886319-38886635 340 chrl9 40993862-40994097 341 chrl9 4104865-4105239 342 chrl9 42828138-42828485 343 chrl9 44324693-44324827 344 chrl9 45889046-45889210 345 chrl9 45898697-45898908 346 chrl9 46379859-46380198 DMB hgl9 genomic coordinate 347 chrl9:48984019-48984319 348 chrl9:49936869-49936984 349 chrl9:50004210-50004626 350 chrl9:51231728-51231909 351 chrl9:51830195-51830382 352 chrl9:5338762-5339204 353 chrl9:54481859-54482102 354 chrl9:9473590-9473791 355 chr2:105458990-105459192 356 chr2:105459614-105459841 357 chr2:113463877-113464049 358 chr2:113594343-113594612 359 chr2:119566173-119566386 360 chr2:119594493-119594777 361 chr2:119599689-119599831 362 chr2:119602802-119602863 363 chr2:121412134-121412214 364 chr2:130971054-130971317 365 chr2:132088620-132088868 366 chr2:157177008-157177075 367 chr2:157178165-157178259 368 chr2:157178647-157178754 369 chr2:162273268-162273352 370 chr2:162275746-162275926 371 chr2:162277174-162277301 372 chr2:162279621-162279890 373 chr2:162280483-162280586 374 chr2:162282980-162283190 375 chr2:162283381-162283731 376 chr2:172972827-172972895 377 chr2:173204440-173204640 378 chr2:175208488-175208650 379 chr2:175208761-175208825 380 chr2:176948069-176948154 381 chr2:176993545-176993719 382 chr2:182325638-182326030 383 chr2:182451476-182451538 384 chr2:189046152-189046366 385 chr2:200335520-200335563 386 chr2:206551212-206551365 387 chr2:206890514-206890843 388 chr2:208491961-208492161 389 chr2:208989143-208989383 390 chr2:218767636-218767656 391 chr2:218770208-218770488 392 chr2:219736312-219736592 393 chr2:220299604-220299735 394 chr2:220313202-220313433 395 chr2:220349206-220349546 396 chr2:220349551-220349909 397 chr2:223163784-223163817 398 chr2:228736324-228736477 399 chr2:37308011-37308335 400 chr2:45029176-45029289 401 chr2:45158117-45158305 402 chr2:45160439-45160513 403 chr2:45168246-45168351 404 chr2:45171790-45171956 405 chr2:45231306-45231482 DMB hgl9 genomic coordinate 406 chr2:45231782-45231832 407 chr2:45231868-45231950 408 chr2:45231981-45232316 409 chr2:45232391-45232602 410 chr2:45240493-45240633 411 chr2:63275459-63275563 412 chr2:63280968-63281090 413 chr2:63281117-63281244 414 chr2:63282922-63283035 415 chr2:63283967-63284166 416 chr2:63286006-63286050 417 chr2:86263537-86263968 418 chr2:96990741-96990915 419 chr2:97136442-97136577 420 chr2:97193136-97193563 421 chr20:21378157-21378360 422 chr20:21492886-21492998 423 chr20:23015899-23016003 424 chr20:25062754-25062861 425 chr20:25129141-25129458 426 chr20:25129465-25129610 427 chr20:2801670-2801746 428 chr2 0:3052550-3053040 429 chr2 0:30639045-30639430 430 chr2 0:39126768-39127082 431 chr2 0:39319466-39319541 432 chr2 0:42544563-42544904 433 chr2 0:50158202-50158523 434 chr2 0:55202189-55202302 435 chr2 0:55499326-55499549 436 chr21:27011102-27011140 437 chr21:37752927-37752973 438 chr21:38065419-38065556 439 chr21:46352817-46353008 440 chr21:46694595-46695075 441 chr21:48026043-48026395 442 chr22:19754627-19754839 443 chr22:24549500-24549739 444 chr22:29709286-29709342 445 chr22:29876932-29876978 446 chr22:46044084-46044331 447 chr22:50438395-50438502 448 chr22:50986972-50987295 449 chr22:50987303-50987506 450 chr3:107317882-107318379 451 chr3:112693867-112694063 452 chr3:121903467-121903521 453 chr3:122296541-122296866 454 chr3:127633994-127634299 455 chr3:127794759-127794993 456 chr3:129693258-129693438 457 chr3:129693441-129693846 458 chr3:129693912-129694521 459 chr3:138679286-138679378 460 chr3:138679392-138679579 461 chr3:147105970-147106037 462 chr3:147106455-147106564 463 chr3:147111638-147111722 464 chr3:147127646-147127683 DMB hgl9 genomic coordinate 465 chr3:147141096-147141263 466 chr3:157813322-157813382 467 chr3:157814023-157814398 468 chr3:170303607-170303722 469 chr3:181437106-181437277 470 chr3:181437291-181437399 471 chr3:181441471-181441671 472 chr3:192228556-192228922 473 chr3:192444904-192445086 474 chr3:193506005-193506239 475 chr3:193996432-193996677 476 chr3:27063888-27064156 477 chr3:27771389-27771691 478 chr3:43221639-43221816 479 chr3:45076240-45076600 480 chr3:5137530-5137823 481 chr3:62356962-62357146 482 chr3:71834773-71834930 483 chr3:9177892-9178250 484 chr3:9904307-9904634 485 chr3:99594893-99595122 486 chr3:99595130-99595205 487 chr4:102711858-102712120 488 chr4:102712125-102712252 489 chr4:108642311-108642511 490 chr4:111550684-111550831 491 chr4:111558727-111558994 492 chr4:126238254-126238526 493 chr4:13526669-13526869 494 chr4:140656749-140657058 495 chr4:147559368-147559439 496 chr4:147561143-147561204 497 chr4:154713748-154713912 498 chr4:155662874-155663347 499 chr4:155663396-155663634 500 chr4:174439822-174440104 501 chr4:174440334-174440548 502 chr4:174452381-174452507 503 chr4:174456504-174456571 504 chr4:1811805-1811834 505 chr4:183062127-183062322 506 chr4:185937376-185937876 507 chr4:188916633-188916727 508 chr4:26030463-26030719 509 chr4:40999951-41000339 510 chr4:41882516-41882699 511 chr4:4859772-4859842 512 chr4:4859975-4860062 513 chr4:55990889-55990963 514 chr4:85424195-85424289 515 chr4:85424369-85424569 516 chr4:90815958-90816298 517 chr4:90816307-90816483 518 chr5:115151968-115152432 519 chr5:115152460-115152604 520 chr5:115299027-115299089 521 chr5:119799056-119799221 522 chr5:122431080-122431245 523 chr5:131792820-131793303 DMB hgl9 genomic coordinate 524 chr5:134363306-134363349 525 chr5:134363908-134364007 526 chr5:134827169-134827555 527 chr5:139080900-139081137 528 chr5:140787474-140787624 529 chr5:141931152-141931357 530 chr5:1446196-1446345 531 chr5:145719938-145720035 532 chr5:14676030-14676461 533 chr5:160972998-160973149 534 chr5:161274440-161274640 535 chr5:161495013-161495077 536 chr5:170740870-170740914 537 chr5:172672533-172672685 538 chr5:172672894-172673044 539 chr5:174158548-174158712 540 chr5:174158976-174159283 541 chr5:174159294-174159388 542 chr5:174159406-174159674 543 chr5:176023778-176024184 544 chr5:176107096-176107262 545 chr5:176107290-176107492 546 chr5:176433637-176434132 547 chr5:178957538-178958008 548 chr5:180075873-180075929 549 chr5:180596563-180596815 550 chr5:1876386-1876421 551 chr5:1878672-1878779 552 chr5:1879621-1879706 553 chr5:320831-320889 554 chr5:3596197-3596466 555 chr5:3596468-3596705 556 chr5:40681077-40681411 557 chr5:42993734-42994199 558 chr5:42994655-42994824 559 chr5:42995338-42995535 560 chr5:43017382-43017562 561 chr5:43017595-43017809 562 chr5:43019286-43019726 563 chr5:43019748-43019979 564 chr5:43040476-43040622 565 chr5:43040996-43041384 566 chr5:59189059-59189156 567 chr5:72595636-72595717 568 chr5:72740432-72740478 569 chr5:74231065-74231286 570 chr5:76034700-76035159 571 chr5:76145903-76146149 572 chr5:76924122-76924200 573 chr5:76937863-76938063 574 chr5:78407678-78407705 575 chr5:87441873-87441994 576 chr5:87980578-87981035 577 chr5:87981051-87981189 578 chr6:100050918-100051132 579 chr6:100903458-100903562 580 chr6:100903567-100903585 581 chr6:100903839-100904127 582 chr6:100904428-100904790 DMB hgl9 genomic coordinate 583 chr6:100905283-100905459 584 chr6:100911709-100911896 585 chr6:100912822-100913039 586 chr6:10390917-10391003 587 chr6:10391325-10391485 588 chr6:10393466-10393779 589 chr6:10421442-10421461 590 chr6:10421619-10421829 591 chr6:10421865-10422230 592 chr6:10422266-10422390 593 chr6:10424309-10424469 594 chr6:106429273-106429533 595 chr6:106433890-106434178 596 chr6:106441968-106442011 597 chr6:108145484-108145931 598 chr6:108488080-108488336 599 chr6:111928637-111928908 600 chr6:131602680-131603008 601 chr6:133561575-133561631 602 chr6:134210279-134210355 603 chr6:137814620-137814729 604 chr6:1385797-1386099 605 chr6:1392971-1393208 606 chr6:14661412-14661636 607 chr6:150286302-150286540 608 chr6:20023958-20024448 609 chr6:25056470-25056724 610 chr6:26223791-26224273 611 chr6:26224392-26224791 612 chr6:26577174-26577337 613 chr6:27235791-27235991 614 chr6:27463183-27463251 615 chr6:27513267-27513460 616 chr6:27525987-27526030 617 chr6:27647843-27647971 618 chr6:27648605-27648831 619 chr6:28175078-28175388 620 chr6:28175984-28176172 621 chr6:29795595-29795703 622 chr6:29943349-29943505 623 chr6:3053739-3053881 624 chr6:35479529-35479748 625 chr6:38682950-38683173 626 chr6:42145872-42145955 627 chr6:45631261-45631364 628 chr6:50674641-50675011 629 chr6:50818685-50818872 630 chr6:6003913-6004224 631 chr6:71938185-71938655 632 chr6:73329800-73330267 633 chr6:85473551-85474051 634 chr6:85476132-85476250 635 chr6:85477979-85478297 636 chr7:101006052-101006064 637 chr7:106507245-106507475 638 chr7:117119508-117119708 639 chr7:117119806-117120258 640 chr7:121956890-121956958 641 chr7:121957094-121957255 DMB hgl9 genomic coordinate 642 chr7:1270483-1270564 643 chr7:129423033-129423219 644 chr7:132339978-132340451 645 chr7:151078549-151078943 646 chr7:153584414-153584552 647 chr7:153584813-153584943 648 chr7:19146030-19146400 649 chr7:19184051-19184274 650 chr7:19184906-19185021 651 chr7:23715959-23716342 652 chr7:24323586-24324005 653 chr7:27204713-27204769 654 chr7:27204771-27205099 655 chr7:27205102-27205509 656 chr7:27260102-27260467 657 chr7:27291305-27291418 658 chr7:29283532-29283808 659 chr7:35296951-35297034 660 chr7:35301173-35301362 661 chr7:39649224-39649336 662 chr7:39649347-39649481 663 chr7:50347926-50348422 664 chr7:55087447-55087946 665 chr7:55088037-55088105 666 chr7:66096480-66096807 667 chr7:6703991-6704046 668 chr7:79083753-79083998 669 chr7:80080122-80080214 670 chr7:8480853-8480880 671 chr7:8481465-8481628 672 chr7:8482072-8482166 673 chr7:92533675-92533867 674 chr7:97361402-97361435 675 chr8:101118463-101118577 676 chr8:103135293-103135596 677 chr8:105598054-105598225 678 chr8:120651069-120651542 679 chr8:140714586-140714675 680 chr8:142428074-142428279 681 chr8:144871650-144872136 682 chr8:145068673-145068873 683 chr8:18244823-18244929 684 chr8:2127280-2127468 685 chr8:22562187-22562617 686 chr8:23563940-23564289 687 chr8:24771430-24771520 688 chr8:24772257-24772351 689 chr8:24772357-24772856 690 chr8:26721306-26721767 691 chr8:57358556-57358780 692 chr8:65290320-65290714 693 chr8:67873655-67873921 694 chr8:70981777-70982099 695 chr8:72470927-72470990 696 chr8:72471017-72471160 697 chr8:79427983-79428118 698 chr8:86350423-86350633 699 chr8:97157963-97158142 700 chr8:99440179-99440379 DMB hgl9 genomic coordinate 701 chr9:100616554-100616782 702 chr9:100864214-100864377 703 chr9:1042461-1042608 704 chr9:129377230-129377578 705 chr9:129377592-129378026 706 chr9:129380714-129381017 707 chr9:129382979-129383285 708 chr9:129384447-129384768 DMB hgl9 genomic coordinate 709 chr9:129388455-129388720 710 chr9:129389047-129389137 711 chr9:129393063-129393329 712 chr9:129439142-129439425 713 chr9:140051156-140051289 714 chr9:140348803-140348996 715 chr9:37002555-37002778 716 chr9:86886580-86886663 Table 4 — Genomic coordinates of 716 identified DMBs Accumulated signal from the 716 DMBs was calculated as a fraction of methylated fragments summed over all 716 DMBs (Figure 8). The target DMBs were identified from TC data based on the following conditions: (i) median fraction of methylated fragments in healthy samples <0.005 and (ii) the difference of fractions of methylated fragments between 90th quantile in cancer samples and 90th quantile in controls >0.1. For this analysis the inventors used a subset of ovarian cancer samples with tumour fraction >0.1. Notably, the 716 DMBs selected for Model 2 included all the 73 DMBs selected during the development of Model 1. 2. Accumulated signal in tissue-specific hypomethylated blocks. The accumulated signal was extracted as above for the following categories of tissue type: "Breast-Basal", "Breast-Luminal", "Lung", "Ovarian", "Prostate". The fraction of unmethylated reads that overlap the tissue-specific blocks of hypomethylation (# of unmethylated / (# of unmethylated + # of methylated) was determined for each plasma sample. The values were median normalized within-sample across tissue types and across samples to median value in healthy controls for each tissue type (Figure 9). The right-hand panels of Figure 9 (plasma samples from ovarian cancer patients) exhibit a higher median fraction of unmethylated reads that overlap the ovary-specific blocks of hypomethylation, indicating that an ovary-derived hypomethylation signal is elevated in cfDNA from patients with ovarian cancer. 3. Genome-wide measure of somatic copy number alterations (SCNAs). The inventors divided the reference genome into equal 1 Mb non-overlapping bins and calculated copy number values per bin with QDNAseq R package (Scheinin et al. 2014) . Then, a median absolute deviation (MAD) of copy numbers was calculated for each sample (Figure 10). 4. Genome-wide fragment length ratio. To extract cfDNA fragment lengths for each sample, the alignments were processed with the cfDNA- Pro R package (Wang et al., 2024). For each sample, the inventors calculated a ratio of the fragment count in range 75bp-160bp to the fragment count in range 167bp-225bp (Figure 11). 5. 4-mer 5'-end. sequence motifs of cfDNA fragments. The inventors used 5 the cfDNA-Pro R package to extract frequencies of 4-mer 5'-end sequence motifs of cfDNA fragments from sWMS data. The inventors identified two motif groups which were overrepresented (group 1) and underrepresented (group 2) in cancer samples based on t-test results with a threshold for adjusted P-value of 0.0001 (Figure 12). The motifs of group 1 and of group 2 are set forth below in Table 5. motif median_fraction_delta p.adj mo ti f_group CACG -0.000120802 0.000027965 groupl CAGC -0.000510876 2.22768E-08 groupl CAGG -0.000628835 0.0000912 groupl CCGG -0.00011951 6.6757E-06 groupl CCGT -7.13262E-05 0.000080157 groupl CT GA -0.000286108 1.63237E-05 groupl CTCC -0.000302811 0.000023836 groupl CTCG -0.00010652 1.79883E-05 groupl CTGC -0.000417404 0.00001452 groupl TCCC -0.000227501 0.000087249 groupl TGGC -0.000503776 7.7805E-07 groupl ACAA 0.000437898 1.1466E-06 group2 ACAT 0.000430917 5.0336E-06 group2 ACTA 0.000325332 3.5478E-06 group2 AGTA 0.000197566 0.0000585 group2 CATA 0.000424403 0.000069368 group2 GATA 0.000263735 2.988E-07 group2 GCAA 0.000311562 0.000063376 group2 GCAT 0.000276511 2.16672E-06 group2 GGTA 0.000372653 1.9228E-08 group2 GT TA 0.00024889 1.01598E-06 group2 TACA 0.000416919 5.5056E-07 group2 TATA 0.00109058 1.36192E-13 group2 TATC 0.000375095 7.0029E-08 group2 TATG 0.000442382 5.4864E-09 group2 TATT 0.000650676 2.1825E-07 group2 TCAT 0.00017773 1.63506E-05 group2 TCTA 0.000395373 1.1373E-09 group2 TGTA 0.000474649 0.00008464 group2 Table 5 — 4-mer 5'-end sequence motifs Model building. The inventors used the five features listed above to build a random forest classification model within the R tidymodels framework (Kuhn et al., 2020) . The dataset was split into training and test subsets in a 3:1 ratio accounting for the balance of classes. The inventors tuned model hyperparameters with grid search in 10-fold cross-validation based on the training subset. The models were built within tidymodels framework which provides an R interface for interaction with various packages. The Random Forest models are built with the ranger package (https: / / cran.r-project.org / web / packages / ranger / ranger.pdf). The inventors did crossvalidation on the training set to tune the model and select the one with the best performance. Hyperparameters of the final Model 2 are: 1) number of trees = 1000; 2) number of randomly selected predictors = 3; 3) minimal node size = 18. Results The best model ("sWMS model") was selected based on the highest mean value of AUC in cross-validation (AUC = 0.967). Model 2 performance was finally assessed on the test subset, where AUC was 0.991. Model 2 assigns each sample a score ranging from 0 to 1, which represents the probability of being attributed to the "cancer" class, i.e. the probability of cancer being present, which can also be represented as a percentage (Figure 13). References A number of publications are cited above in order to more fully describe and disclose the invention and the state of the art to which the invention pertains. Full citations for these references are provided below. The entirety of each of these references is incorporated herein. Aran, Dvir, Marina Sirota, and Atul J. Butte. 2015. "Systematic PanCancer Analysis of Tumour Purity." Nature Communications 6 (1): 8971. Loyfer, Netanel, Judith Magenheim, Ayelet Peretz, Gordon Cann, Joerg Bredno, Agnes Klochendler, Ilana Fox-Fisher, et al. 2023. "A DNA Methylation Atlas of Normal Human Cell Types." Nature 613 (7943): 35564 . Moss, Joshua, Judith Magenheim, Daniel Neiman, Hai Zemmour, Netanel Loyfer, Amit Korach, Yaacov Samet, et al. 2018. "Comprehensive Human Cell-Type Methylation Atlas Reveals Origins of Circulating Cell-Free DNA in Health and Disease." Nature Communications 9 (1): 5068. Scheinin, Ilari, Daoud Sie, Henrik Bengtsson, Mark A. van de Wiel, Adam B. Olshen, Hinke F. van Thuijl, Hendrik F. van Essen, et al. 2014. "DNA Copy Number Analysis of Fresh and Formalin-Fixed Specimens by Shallow Whole-Genome Sequencing with Identification and Exclusion of Problematic Regions in the Genome Assembly." Genome Research 24 (12): 2022-32. Vaisvila, Romualdas, V. K. Chaithanya Ponnaluri, Zhiyi Sun, Bradley W. Langhorst, Lana Saleh, Shengxi Guan, Nan Dai, et al. 2021. "Enzymatic Methyl Sequencing Detects DNA Methylation at Single-Base Resolution from Picograms of DNA." Genome Research, June. https: / / doi.org / 10.1101 / gr.266551.120. Brent S. Pedersen, Kenneth Eyring, Subhajyoti De, Ivana V. Yang, David A. Schwartz 2014. "Fast and accurate alignment of long bisulfite-seq reads." ArXiv >Quantitative Biology >Genomics. arXiv:1401.1129. Netanel Loyfer, Jonathan Rosenski, Tommy Kaplan 2024. "wgbstools: A computational suite for DNA methylation sequencing data representation, visualization, and analysis." bioRxiv 2024.05.08.593132. Love, M.I., Huber, W. &Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550 (2014). https: / / doi.org / 10.1186 / sl3059-014-0550-8 Adalsteinsson, V.A., Ha, G., Freeman, S.S. et al. Scalable whole-exome sequencing of cell-free DNA reveals high concordance with metastatic tumors. Nat Commun 8, 1324 (2017). https: / / doi.org / 10.1038 / s41467-017~ 0 0 9 65-y Haichao Wang, Paulius Mennea, Elkie Chan, Hui Zhao, Christopher G. Smith, Tomer Kaplan, Florian Markowetz, Nitzan Rosenfeld (2024) . cfDNAPro:An R / Bioconductor package to extract and visualise cell-free DNA biological features. R package version 1.7.1 https : / / github.com / hw538 / cfDNAPro Kuhn et al., (2020). Tidymodels: a collection of packages for modeling and machine learning using tidyverse principles. https : / / ww. tidymodels . org Heider, K., et al. Detection of ctDNA from Dried Blood Spots after DNA Size Selection. Clinical Chemistry 66, 697-705 (2020). For standard molecular biology techniques, see Sambrook, J., Russel, D.W. Molecular Cloning, A Laboratory Manual. 3 ed. 2001, Cold Spring Harbor, New York: Cold Spring Harbor Laboratory Press Tost, J. DNA Methylation: An Introduction to the Biology and the Disease-Associated Changes of a Promising Biomarker. Mol Biotechnol 44, 71-81 (2010). https: / / doi.org / 10.1007 / sl2033-009-9216-2 Varley KE, et al. Dynamic DNA methylation across diverse human cell lines and tissues. Genome Res. 2013;23:555-567. doi: 10.1101 / gr.147942.112. Slieker RC, et al. Identification and systematic annotation of tissuespecific differentially methylated regions using the Illumina 450k array. Epigenet. Chrom. 2013;6:1-12. Wright MN, Ziegler A (2017). "ranger: A Fast Implementation of Random Forests for High Dimensional Data in C++ and R." Journal of Statistical Software, 77(1), 1-17. doi:10.18637 / jss.v077.iO1

Claims

1. A computer-implemented method for predicting whether a subject has a cancer, comprising:providing multimodal data comprising at least two feature types, said feature types comprising a metric of methylation of each of at least 10 differentially methylated blocks of homogeneous methylation ("DMBs") in a cell-free DNA (cfDNA)-containing sample obtained from the subject and a metric of somatic copy number alterations of at least 10 non-overlapping bins of the genome of the subject in a cell-free DNA (cfDNA)-containing sample obtained from the subject;inputting the multimodal data into a multimodal machine learning classifier that has been trained on a labelled training data set comprising corresponding multimodal data for each of: (i) a plurality of subjects known to have had the cancer the time of sample collection ("cases"); and (ii) a plurality of subjects known to not have the cancer at the time of sample collection ("controls"); andcausing the multimodal machine learning classifier to classify the subject into the cases class and thereby as having cancer or the controls class and thereby not having cancer based on at least the inputted multimodal data, optionally wherein the classifier outputs a numerical probability of cancer.

2. The method of claim 1, wherein the metric of methylation comprises the fraction of methylated fragments summed over said at least 10 DMBs .

3. The method of claim 1 or claim 2, wherein the metric of somatic copy number alterations comprises the median absolute deviation (MAD) of copy numbers, optionally wherein the at least 10 non-overlapping bins of the genome of the subject are each 1 Mb in size.

4. The method of any one of the preceding claims, wherein the at least 10 DMBs comprise cancer-specific hypermethylated DMBs.

5. The method of claim 4, wherein each of said cancer-specific hypermethylated DMBs satisfies the following conditions:(i) the DMB block comprises at least 1, 2, 3, 4 or at least 5 CpGs, wherein all CpGs, if there is more than 1, share substantially the same level of methylation in a given sample;(ii) the DMB is no more than 5000 bp in length;(iii) the median fraction of methylated fragments mapping to the DMB in a plurality of samples from controls is less than 0.005; and(iv) the 90th quantile fraction of methylated fragments mapping to the DMB in a plurality of samples from cancer cases is at least 0.1 greater than the 90th quantile fraction of methylated fragments mapping to the DMB in a plurality of samples from controls.

6. The method of any one of the preceding claims, wherein the cancer is selected from: ovarian cancer, breast cancer, prostate cancer and oesophageal cancer.

7. The method of claim 6, wherein the cancer is ovarian cancer.

8. The method of claim 7, wherein the at least 10 DMBs are selected from the 716 DMBs set forth in Table 4.

9. The method of claim 7 or claim 8, wherein the at least 10 DMBs areselected from the 73 DMBs set forth in Table 3.

10. The method of any one of claims 7 DMBs comprise at least 20, 30, 40, 50, set forth in Table 3.to 9, wherein the at least 1060, 70 or all of the 73 DMBs11. The method of any one of claims 7 to 10, wherein the at least 10 DMBs comprise at least 20, 30, 40, 50, 100, 150, 200, 250, 300, 350, 400, 450, 500, 550, 600, 650, 700 or all 716 DMBs set forth in Table 4 .

12. The method of any one of the preceding claims, wherein the multimodal data further comprises at least one, at least two or three additional feature types selected from the group:(i) an accumulated signal in tissue-specific hypomethylated blocks feature;(ii) a fragment length ratio feature; and(iii) a fragment 5'-end sequence motif group feature.

13. The method of claim 12, wherein the accumulated signal in tissuespecific hypomethylated blocks feature comprises median-normalised fraction of unmethylated cfDNA fragments mapping to tissue-specific blocks of hypomethylation for each of one or more of the following tissue types: breast-basal, breast-luminal, lung, ovarian and prostate .

14. The method of claim 12 or claim 13, wherein the fragment length ratio feature comprises the ratio of: (i) the count of cfDNA fragments of length in the range 75-160 bp to (ii) the count of cfDNA fragments of length in the range 167-225 bp, said cfDNA fragment counts being determined from DNA sequencing data derived from the cfDNA-containing sample obtained from the subject.

15. The method of any one of claims 12 to 14, wherein the fragment 5'-end sequence motif feature comprises the frequency, optionally centred and scaled frequency, of a) cfDNA fragments having a 4-mer 5'-end sequence belonging to a group that is underrepresented in cancer samples and b) cfDNA fragments having a 4-mer 5'-end sequence belonging to a group that is overrepresented in cancer samples, further optionally wherein the threshold for adjusted p-value of t-test of underrepresented vs overrepresented is set at 0.0001.

16. The method of claim 15, wherein the 4-mer 5'-end sequence motifs overrepresented in cancer comprise the Group 1 4-mer sequences set forth in Table 5 and the 4-mer 5'-end sequence motifs underrepresented in cancer comprise the Group 2 4-mer sequences set forth in Table 5.

17. The method of any one of the preceding claims, wherein the machine learning classifier exhibits predictive performance measured as area under curve of the receiver operating characteristic curve (AUROC) of at least 0.60, 0.70, 0.80 or 0.90, as assessed on a test data set comprising labelled data for a plurality of subjects known to have said cancer at the time of sample collection and a plurality of subjects known not to have any cancer at the time of samplecollection, said test data set differing from said labelled training data set.

18. The method of any one of the preceding claims, wherein the machine learning classifier comprises a Random Forest, a neural network, a support vector machine, a logistic regression, a naive Bayes or a perceptron.

19. A method for detecting cancer in a subject, comprising:a) providing a cfDNA-containing sample obtained from the subject;b) performing methylation sequencing of the cfDNA or of a library generated from the cfDNA to obtain methylation sequencing reads, at least a portion of said sequencing reads mapping to at least 10 differentially methylated blocks of homogeneous methylation ("DMBs");c) analysing said sequencing reads to generate a metric of methylation of each of the DMBs and a metric of somatic copy number alterations of at least 10 non-overlapping bins of the genome of the subject, thereby generating multimodal data for the subject; andd) carrying out the computer-implemented method of any one of claims 1 to 18 using at least the multimodal data from step c) as input data, wherein cancer is considered to have been detected when the multimodal machine learning classifier classifies the subject into the cases class and is considered not to have been detected when the multimodal machine learning classifier classifies the subject into the controls class .

20. The method of claim 19, wherein the methylation sequencing comprises shallow whole methylome sequencing.

21. The method of claim 19 or claim 20, wherein the cfDNA-containing sample comprises a plasma sample, a blood sample, a urine sample, a tear sample, a saliva sample, a final needle aspirate (FNA) sample, a cerebrospinal fluid (CSF) sample or a fecal sample.

22. The method of claim 21, wherein the cfDNA sample comprises a blood spot sample, optionally a dried blood spot sample or a pin-prickblood sample, further optionally wherein the cfDNA sample has not more than 100 pg, not more than 50 pg or not more than 20 pg of DNA.

23. The method of any one of claims 19 to 22, wherein the cfDNA-containing sample is subjected to a single round of size selection.

24. The method of any one of claims 19 to 23, wherein the methylation sequencing comprises an enzymatic methyl-seq (EM-seq), optionally wherein said EM-seq comprises the NEBNext Enzymatic Methyl-seq protocol modified to have a bead incubation time of between 7 minutes and 15 minutes, such as 10 minutes and / or modified to employ a number of PCR cycles in the range 15-20, such as 18 cycles.

25. The method of any one of the preceding claims, wherein the method is for: early detection of cancer; detection of minimal residual disease (MRD) or cancer recurrence following treatment, such as following surgical treatment; monitoring cancer progression or remission; monitoring response to an anti-cancer therapy; and / or assessing the stage and / or severity of cancer.

26. The method of any one of the preceding claims, wherein the method is carried out for each of a plurality of subjects, and wherein the method is for stratifying the subjects by cancer risk.

27. A system comprising one or more data processors and a non-transitory computer readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to carry out the method of any one of claims 1 to 18 on multimodal data provided to the one or more data processors.

28. The system of claim 27, wherein said computer readable storage medium further comprises one or more learned numerical parameters of the machine learning classifier.

29. A computer readable storage medium containing instructions that, when executed on one or more data processors, cause the one or more data processors to carry out the method of any one of claims 1 to 18 on multimodal data provided to the one or more data processors.

30. A method of treatment of cancer in a subject, comprising:a) carrying out the method of any one of claims 1 to 26, wherein the subject is predicted to have cancer; and5 b) administering a therapeutically effective amount of an anticancer agent to the subject.s

Images