Apparatus for determining indicator of presence of HRD in genome of subject

Abstract

The present invention relates to a device for determining the presence of a homologous recombination defect (HRD) index in a genome of a subject, said device being configured to: receive superficial WGS data and non-superficial sequencing data relating to a set of genes in the genome of the subject, obtain at least one first parameter from the superficial WGS data, and obtaining at least one second parameter from the non-superficial sequencing data and determining an HRD index indicative of the presence of HRD in the genome of the subject by applying an HRD predictive machine learning model to the obtained at least first and second parameters.

Description

Technical Field

[0001] The present invention relates to methods and apparatus for detecting homologous recombination deficiency in the genome of a subject. Background Art

[0002] Homologous recombination (HR) is a natural DNA repair mechanism. This mechanism can be disrupted by inactivation of genes in the homologous recombination repair (HRR) pathway. This inactivation is known as homologous recombination deficiency (HRD) and leads to the accumulation of numerous genomic alterations. This defect has been associated with a variety of tumor types, including breast, ovarian, prostate, and pancreatic cancers.

[0003] HRD tumors have been found to be more sensitive to some treatments, such as poly (adenosine diphosphate [ADP]-ribose) polymerase (PARP) inhibitor (PARPi)-based therapy for patients with ovarian cancer. Therefore, HRD identification is important for patients with some types of cancer, including ovarian cancer, as it can select appropriate treatments for the patients.

[0004] There are tests in the prior art to determine the HRD status of a subject (positive, i.e., with HRD, or negative, i.e., without HRD). In particular, by using single nucleotide polymorphism (SNP) sequencing data based on a custom hybridization capture panel to estimate three indicators, it can be determined whether a woman with ovarian cancer has HRD. These indicators are large scale transitions (LST), loss of heterozygosity (LoH), and trans-telomeric allelic imbalance (TAI), all of which increase when the homologous recombination repair mechanism is disrupted. Through these indicators, the characteristics of HRD in the genome can be quantified. The main disadvantage of this method is that it requires the use of specific capture panel probes to target SNPs and enrich these regions to a high depth (average 50X). Therefore, this method is time-consuming and costly.

[0005] Another existing solution is to estimate the probability of HRD based on deep learning methods originally designed for the field of computer vision. This solution uses shallow WGS data as input to its deep learning method. It is reviewed that shallow WGS is a shotgun sequencing-based technique that is used to obtain whole genome sequences at very low depths and is generally more cost-effective than high-depth WGS. The main disadvantage of this approach is that the deep learning model must be trained on a very large dataset, which is not easily available.

[0006] The object of the present invention is to propose a new method for identifying HRD in a subject, which overcomes the disadvantages of the above-mentioned existing methods.

[0007] The present invention proposes a new method that is more sensitive and enables faster and more economical diagnosis of HRD patients. Summary of the invention

[0008] Therefore, the present invention relates to a computer-implemented method and apparatus for automatically identifying the HRD status of a subject by using a machine learning method, which takes several kinds of gene markers as input. These gene markers can be obtained from both shallow WGS data and panel data. Panel data refers to data from a panel test, that is, data from a set of regions known to be associated with the development of a given disease (e.g., regions including genes associated with ovarian cancer (e.g., BRCA1 or BRCA2)). Typically, mutations that cause the disease are sought in this set of genes. According to the present invention, gene markers include a measure of quantifying genomic instability associated with HRD, a measure of genetic variation identified from panel data, and / or amplification levels of a set of genes known to be HRD markers. Then, all of these inputs are used for a machine learning classifier that is trained to predict whether a sample is HRD positive or negative.

[0009] Therefore, one aspect of the present invention relates to an apparatus for determining a homologous recombination defect (HRD) index (ie, an indicator) indicating the presence of a homologous recombination defect (HRD) in the genome of a subject. In one or more embodiments, the apparatus may include:

[0010] • an input interface for receiving first sequencing data from a shallow whole genome sequencing (WGS) process of a subject's genome and second sequencing data from a non-shallow sequencing process of a set of fragments in the subject's genome, the set of fragments comprising a set of genes referred to as a gene panel set;

[0011] • At least one processor configured to:

[0012] o Obtain at least one first parameter according to the first sequencing data, the at least one first parameter being among the following parameters:

[0013] a corresponding first measurement result representing the number of large-scale genomic alterations (LGA) in the subject's genome;

[0014] a second measurement indicating the number of fragments in the subject's genome that have partial deletions of chromosomes; and

[0015] at least one index, referred to as the CNV index, representing the amplification level of genes in a predefined gene set, referred to as the CNV gene set;

[0016] o obtaining at least one index from the second sequencing data, the index being referred to as the panel index, representing somatic variation on genes of the gene panel set;

[0017] o Determining the HRD index representing the presence of HRD in the subject's genome based on a trained HRD prediction machine learning (ML) model applied to at least the first parameter and at least one panel indicator.

[0018] "Sequencing data from a non-shallow sequencing process" refers to sequencing data obtained by a standard sequencing process that is not a shallow process, and the average depth of the standard sequencing process is usually 50X to 500X, such as a next-generation sequencing (NGS) process. In addition, "second sequencing data from a non-shallow sequencing process of a set of fragments in the subject's genome" means that these non-shallow sequencing data are obtained by sequencing only a set of fragments in the subject's genes instead of the entire genome. Therefore, the non-shallow sequencing process is not a WGS process.

[0019] The "first measurement result indicating the number of large-scale state transitions (LST) in the subject's genome" may also be referred to as large-scale genomic alterations (LGA). The "second measurement result indicating the number of fragments with partial chromosome deletions in the subject's genome" may also be referred to as an indicator of loss of heterozygosity (LOH) levels. These measurements are known to those skilled in the art.

[0020] The above-mentioned apparatus enables to determine the HRD index representing the presence of HRD in the subject's genome based on shallow WGS data and non-shallow sequencing data of only a set of fragments of the genome rather than the entire genome. Therefore, HRD determination according to the present invention is more efficient in terms of time and cost than the prior art.

[0021] The measurements and / or metrics may be received by a device (eg, from a sequencing device) or determined by a processor of the device based on received sequencing data.

[0022] According to other advantageous aspects of the invention, the device comprises one or more of the features described in the following embodiments, which features may be used alone or in any possible combination.

[0023] According to one embodiment, the trained HRD prediction machine learning model has been obtained by training the machine learning model using a training data set, the training data set being obtained from first sequencing data of a shallow whole genome sequencing (WGS) process of genomes of multiple subjects, and second sequencing data of a non-shallow sequencing process of a set of fragments in the genomes of the multiple subjects, the set of fragments comprising a gene set referred to as a gene panel set; wherein the training data set comprises a plurality of training data subsets, each subset being associated with one of the multiple subjects, and comprising:

[0024] • obtaining at least one corresponding first parameter based on the first sequencing data, wherein the at least one corresponding first parameter is among the following parameters:

[0025] o a respective first measurement representing a number of large-scale state transitions (LSTs) in the genome of the subject of the plurality of subjects;

[0026] o a second measurement result representing the number of fragments in the genome of the subject among the plurality of subjects that have a partial deletion of a chromosome; and

[0027] o at least one CNV indicator representing the amplification level of genes in a predefined gene set referred to as the CNV gene set;

[0028] • At least one corresponding panel indicator obtained from the second sequencing data, the at least one corresponding panel indicator representing somatic variations on genes of the gene panel set.

[0029] In one or more embodiments, at least one processor is further configured to obtain a first measurement result, a second measurement result, and at least one CNV indicator, and wherein the HRD prediction ML model is applied to the first measurement result, the second measurement result, and at least one CNV indicator and at least one panel indicator. In addition, due to the combined use of at least one panel indicator and at least one CNV indicator, the present invention is able to distinguish patients who are BRAC1 negative but have high instability, thereby treating these patients as HRD positive.

[0030] According to these embodiments, all measurements and indicators are used as input to the HRD prediction ML model.

[0031] In one or more embodiments, the gene panel set and the CNV gene set may be mutually exclusive.

[0032] "Mutual exclusion" means that the two sets have no common genes. Advantageously, the panel set and the CNV set advantageously include genes that are implicit in the HRD process. Defining or limiting these sets (e.g., these sets are mutually exclusive) reduces redundancy in data processing, thereby making the determination of HRD more cost- and time-effective.

[0033] In one or more embodiments, the CNV gene set may include the following genes: AKT1, BARD1, CCNE1, EMSY, ESR1, H2AX, MRE11, PTEN, RAD51B, RAD52, RAD54.

[0034] In one or more embodiments, the gene panel set may include the BRCA1 gene and / or the BRCA2 gene.

[0035] All of these genes are known to play a role in the HRD mechanism.

[0036] In one or more embodiments, a first measurement representing the number of LGAs in the subject's genome may be determined based on the number of pairs of adjacent fragments, each fragment in each pair being at least 10 Mb long, the fragments in each pair having different copy numbers (CN).

[0037] In one or more embodiments, based on the number of genomic fragments that are at least 10 Mb long and have a copy number of 0.5 to 1.5, a second measurement result representing the number of fragments with partial chromosome deletions in the subject's genome can be determined.

[0038] In one or more embodiments, the at least one CNV indicator may include at least one of the following:

[0039] • For each gene in the CNV gene set, the corresponding copy number of the gene;

[0040] • For each gene in the CNV gene set, a binary value indicating whether a copy number variation (CNV) of that gene was detected;

[0041] • For each gene in the CNV gene set, the ratio between the copy number of the gene and the average copy number of the gene in all CNV gene sets;

[0042] • The number of genes in the CNV gene set where CNVs were detected.

[0043] In one or more embodiments, at least one panel metric may include a counter representing the number of genes in the gene panel set that have somatic variation.

[0044] In one or more embodiments, the subject suffers from a disease, and the at least one processor may be further configured to:

[0045] • determining genes having a corresponding pathogenicity level for the disease above a predefined pathogenicity threshold based on at least one classification of somatic variation; and the gene panel set may include the determined genes.

[0046] In one or more embodiments, the genes of the gene panel set can be divided into several categories of pathogenicity levels, wherein, for each category of pathogenicity levels, at least one panel indicator includes a corresponding counter that represents the number of genes with somatic variants in each category of pathogenicity levels.

[0047] In one or more embodiments, the HRD index representing the presence of HRD in the subject's genome may be a binary variable indicating whether HRD exists in the subject's genome.

[0048] In one or more embodiments, the at least one processor may be further configured to determine a probability of the presence of HRD in the subject's genome based on a trained HRD prediction ML model applied to at least the first parameter and at least one panel indicator.

[0049] In one or more embodiments, the trained HRD prediction ML model may be a regression model.

[0050] For example, the trained HRD prediction ML model can be a regression model based on elastic net regularization. Of course, other machine learning models can also be used.

[0051] In one or more embodiments, the apparatus may include another processor configured to train a machine learning model based on a training data set obtained from a first sequencing data of a shallow whole genome sequencing (WGS) process of genomes of a plurality of subjects and a second sequencing data of a non-shallow sequencing process of a set of fragments in the genomes of the plurality of subjects, the set of fragments including a set of genes referred to as a gene panel set. The training data set may include a plurality of subsets of training data, each subset being associated with each subject in the plurality of subjects, and including:

[0052] • at least one corresponding first parameter determined based on the first sequencing data, wherein the at least one corresponding first parameter is among the following parameters:

[0053] o a corresponding first measurement representing a number of large-scale state transitions (LSTs) in a genome of the subject;

[0054] o a second measurement indicating the number of fragments in the subject's genome that have partial chromosome deletions; and

[0055] o at least one index, referred to as the CNV index, representing the amplification level of genes in a predefined gene set, referred to as the CNV gene set;

[0056] • At least one corresponding index, which is referred to as a panel index and is determined from the second sequencing data, the at least one corresponding index representing somatic variation on genes of the gene panel set.

[0057] In one or more embodiments, the HRD prediction ML machine learning model can be trained in a fully supervised manner.

[0058] Another aspect of the present invention relates to a computer-implemented method for determining a homologous recombination defect (HRD) index (ie, an indicator) indicating the presence of a homologous recombination defect (HRD) in the genome of a subject. In one or more embodiments, the method may include:

[0059] • receiving as input first sequencing data from a shallow whole genome sequencing (WGS) process of the subject's genome and second sequencing data from a non-shallow sequencing process of a set of fragments in the subject's genome, the set of fragments comprising a set of genes referred to as a gene panel set;

[0060] • obtaining at least one first parameter according to the first sequencing data, the at least one first parameter being among the following parameters:

[0061] • A first measure representing the number of large-scale genomic alterations (LGAs) in the subject’s genome;

[0062] • a second measurement representing the number of fragments in the subject's genome that are partially deleted from the chromosome; and

[0063] • at least one CNV indicator representing the gene amplification level in a predefined set of genes referred to as the CNV gene set;

[0064] - obtaining at least one panel indicator from the second sequencing data, the indicator representing somatic variation on genes of the gene panel set;

[0065] - determining said HRD index indicative of the presence of HRD in the subject's genome based on a trained HRD prediction machine learning (ML) model applied to at least the first parameter and at least one panel indicator.

[0066] One aspect of the present invention relates to a device for training a machine learning model to obtain an HRD prediction model. It is noteworthy that the device for obtaining an HRD prediction model of the present invention is used to determine an HRD index indicating the presence of homologous recombination defects in the genome of the subject under study, and the device comprises:

[0067] - at least one input configured to receive first sequencing data from a shallow whole genome sequencing (WGS) process of genomes of a plurality of subjects and second sequencing data from a non-shallow sequencing process of a set of fragments in the genomes of the plurality of subjects, the set of fragments comprising a set of genes referred to as a gene panel set;

[0068] - at least one processor configured to:

[0069] • generating a training data set using the first sequencing data and the second sequencing data of the plurality of subjects, the training data set comprising a plurality of training data subsets; wherein generating the training data set comprises generating at least one of the training data subsets for each subject in the plurality of subjects, the training data subset comprising:

[0070] • at least one first parameter obtained according to the first sequencing data, wherein the at least one first parameter is among the following parameters:

[0071] o a first measurement representing a number of large-scale genomic alterations (LGAs) in the genome of the subject;

[0072] o a second measurement indicating the number of fragments in the subject's genome that have partial chromosome deletions; and

[0073] o at least one CNV indicator representing the amplification level of genes in a predefined gene set referred to as the CNV gene set;

[0074] • at least one corresponding panel indicator obtained from the second sequencing data, the at least one corresponding panel indicator representing somatic variation on genes of the gene panel set;

[0075] - using the training data set to train a machine learning model to obtain the trained HRD prediction model, the trained HRD prediction model being configured to receive as input at least one first parameter and at least one panel index of the subject under study, and to provide as output the HRD index indicating the presence of homologous recombination deficiency (HRD) in the genome of the subject under study.

[0076] In one embodiment, the apparatus for training is configured to generate a training data subset, which also includes information representing the true HRD status of the subject (i.e., HRD positive or negative). The true status can be obtained according to standard methods of the prior art. In this embodiment, the apparatus for training is also configured to train the HRD prediction model in a supervised manner.

[0077] Furthermore, the present disclosure relates to a computer program comprising software code adapted to perform a method for determining an HRD index or a method for training in accordance with any of the above-mentioned execution modes, when the program is executed by a processor.

[0078] The present disclosure also relates to a non-transitory program storage device readable by a computer, the storage device tangibly embodying an instruction program executable by the computer to perform a method for determining an HRD index or a method for training consistent with the present disclosure.

[0079] definition

[0080] In the present invention, the following terms have the following meanings.

[0081] "Whole genome sequencing (WGS)" or "whole genome sequencing", "complete genome sequencing" or "total genome sequencing" refers to the process of determining the entire or nearly entire DNA sequence of an organism's genome in one go.

[0082] "Shallow WGS" refers to a WGS process at very low depth, for example, an average depth of 0.1X to 1X (preferably 0.3X to 1X).

[0083] "Panel data" refers to sequencing data obtained by traditional sequencing methods (e.g., high-depth sequencing) in specific regions of the genome. In the present disclosure, traditional sequencing methods refer to non-shallow sequencing processes. For example, these specific regions of the genome are regions containing genes known to be associated with a given pathology. Therefore, the term "panel data" is in contrast to "WGS data", which indicates sequencing data representing the entire genome.

[0084] The term "copy number" refers to the number of copies of a particular DNA fragment or gene present in the genome of an organism. It represents the duplication or loss of certain DNA regions within a chromosome or on multiple chromosomes. The copy number itself is a dimensionless quantity that indicates the relative increase or decrease in the number of copies of a particular DNA fragment. For example, a copy number of 2 would indicate that there are two copies of a particular DNA fragment, while a copy number of 1 or 3 would indicate a loss or increase of one copy compared to a reference. It is worth noting that the copy number is a relative measurement, not an absolute count of DNA copies. The reference copy number is usually based on a normal or expected baseline, which can vary due to the specific analysis or research context.

[0085] The term "processor" should not be construed as limited to hardware capable of executing software, but rather refers generally to a processing device, which can include, for example, a computer, a microprocessor, an integrated circuit, or a programmable logic device (PLD). A processor may also include one or more graphics processing units (GPUs), whether for computer graphics and image processing or other functions. In addition, instructions and / or data capable of performing related and / or resulting functions may be stored on any processor-readable medium, such as an integrated circuit, a hard disk, a CD (compact disk), an optical disk (such as a DVD (digital versatile disk)), a RAM (random access memory), or a ROM (read-only memory). It is noteworthy that instructions may be stored in hardware, software, firmware, or any combination thereof.

[0086] “Machine learning (ML)” traditionally means that a computer algorithm automatically improves through experience, with the ability to adjust the parameters of a computer model based on training data by reducing the gap between the expected output extracted from the training data and the estimated output calculated by the computer model.

[0087] A "dataset" is a collection of data used to build ML mathematical models in order to make data-driven predictions or decisions. In "supervised learning" (i.e., inferring functions from known input-output examples in the form of labeled training data), three types of ML datasets (also called ML sets) are often used for three tasks: "training" (i.e., fitting parameters), "validation" (i.e., tuning ML hyperparameters (parameters used to control the learning process)), and "testing" (i.e., checking whether the latter model provides satisfactory results, independent of the training dataset used to build the mathematical model). BRIEF DESCRIPTION OF THE DRAWINGS

[0088] The present disclosure will be better understood, and other particular features and advantages will emerge, from a reading of the following description of particular and non-limiting illustrative embodiments, which description refers to the accompanying drawings, in which:

[0089] Figure 1 is an example of a flow chart representing a method for detecting the presence of HRD in a subject according to an embodiment of the present invention;

[0090] Figure 2 An example of an apparatus for detecting the presence of HRD in a subject according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0091] In interpreting the specification and its related claims, expressions such as "include," "comprising," "incorporating," "containing," "is," and "having" should be interpreted in a non-exclusive manner, i.e., interpreted as allowing for the presence of other items or components that are not explicitly defined.

[0092] This specification explains the principles of the present disclosure. Therefore, it should be understood that although not explicitly described or shown herein, those skilled in the art will be able to conceive of various arrangements that embody the principles of the present disclosure and are included within the scope of the present disclosure.

[0093] All examples and conditional language described herein are intended for educational purposes to help readers understand the principles of the present disclosure and the concepts contributed by the inventors to further the development of the art, and should be understood as not being limited to such specifically recited examples and conditions.

[0094] In addition, all the principles, aspects and embodiments of the present disclosure listed herein and all descriptions of their specific examples are intended to cover both their structural and functional equivalents. In addition, such equivalents include both currently known equivalents and equivalents developed in the future, that is, no matter how the structure, the elements of any development perform the same function.

[0095] Thus, for example, those skilled in the art will appreciate that the block diagrams presented herein may represent conceptual illustrations of illustrative circuitry embodying the principles of the present disclosure. Similarly, it will be appreciated that any flow chart, flowchart, or the like process representation may be substantially represented in a computer-readable medium and thus executed by a computer or processor, whether or not such a computer or processor is explicitly shown.

[0096] The functions of the various elements shown in the figures may be provided using dedicated hardware as well as hardware capable of executing software associated with appropriate software. When provided by a processor, the functions may be provided by a single dedicated processor, a single shared processor, or a plurality of individual processors, wherein some processors may be shared.

[0097] It should be understood that the elements shown in the drawings can be implemented by various forms of hardware, software or a combination thereof. Preferably, these elements are implemented with a combination of hardware and software on one or more appropriately programmed general-purpose devices, which may include a processor, a memory and an input / output interface.

[0098] Although the presently described devices for determining an HRD index and for training are general and have a variety of functions that may be performed alternatively or in any cumulative manner, other embodiments within the scope of the present disclosure include devices having only some of the existing functions.

[0099] Each of the devices for determining the HRD index and for training is advantageously a device or a physical part of a device that is designed, configured and / or adapted to perform the above functions and produce an effect or result. In alternative embodiments, any of the devices for determining the HRD index and the devices for training are implemented as a set of devices or physical parts of a device, whether grouped in the same machine or in different, possibly remote machines. For example, the devices for determining the HRD index and / or the devices for training may have functionality distributed on a cloud infrastructure and available to users as a cloud-based service, or have remote functionality accessible via an API.

[0100] Figure 1 is an example of a flow chart representing a method for detecting the presence of HRD in a subject according to an embodiment of the present invention.

[0101] In step 110, DNA sequencing data for a subject with a pathology associated with a tumor (e.g., sequencing data for a woman with ovarian cancer) is received. The sequencing data may generally include shallow WGS data 112 and panel data 114. Shallow WGS data 112 corresponds to sequencing data obtained by a low-depth WGS process (e.g., a process with an average depth of at most 1X or 1.5X). Panel data 114 corresponds to data obtained by sequencing a specific portion of the genome (i.e., a specific DNA portion) by a classical sequencing process (typically with an average depth of 50X to 500X) (e.g., a next-generation sequencing (NGS) process), the specific portion including pre-selected genes of interest for a cancer tumor type.

[0102] The shallow WGS data 112 (i.e., first sequencing data) is used to obtain at least one first parameter. According to one or more embodiments, two metrics may be determined (step 120) based on the shallow WGS data 112: a first metric that quantifies the chromosome breakage rate in the genome, and a second metric that quantifies the loss of heterozygosity rate in the genome.

[0103] In one or more embodiments, the first metric to quantify the chromosome breakage rate in the genome may be LGA, where LGA represents the number of large-scale state transitions (LST) of the genome. It is recalled that LGA is defined as chromosome breaks of at least 10 megabases (Mb) between adjacent regions. In an embodiment, the first metric to quantify the chromosome breakage rate in the genome may be a SNP (large-scale genomic alteration) measurement result. In an embodiment, LGA is determined by counting the number of genomic breaks, each genomic break corresponding to two adjacent genomic fragments, the two adjacent genomic fragments being at least 10 Mb long, having different copy numbers (CN), or equivalently, having different CN ratios (CN ratio is defined as the CN of the genomic fragments of the sample under examination divided by the tumor ploidy). To this end, a value indicating the minimum CN ratio difference between two consecutive genomic fragments can be calculated based on the shallow WGS data, and if the value is within a predefined range (e.g., 0.05 to 0.4), it can be considered that there is a genomic break. In an embodiment, the number of genomic breaks can be determined after filtering out genomic fragments shorter than 3 Mb.

[0104] It is worth noting that the CN of a genomic region determined based on sequencing data is actually an estimated value of the CN of the genomic region. In this specification, the term "CN" can be used to specify an estimated CN.

[0105] In one or more embodiments, a second metric to quantify the rate of loss of heterozygosity in a genome can be a pLOH (pseudo loss of heterozygosity) measurement. Loss of heterozygosity (LOH) corresponds to a loss of contribution to a cell by one parent, which may be caused by direct loss, loss caused by unbalanced rearrangement, gene conversion, mitotic recombination, or partial chromosome loss. The number of genomic regions where LOH occurs cannot be determined based on shallow WGS data by standard methods known to those skilled in the art. Therefore, the present invention proposes to approximate this number by a pLOH measurement, which is defined as the number of genomic fragments in which a partial chromosome loss is detected. For example, the pLOH measurement can be at least 10 Mb, and the number of genomic fragments with an estimated CN of about 1 (e.g., between 0.5 and 1.5) or a CN ratio of about 0.5 (e.g., between 0.25 and 0.75).

[0106] Thus, in one or more embodiments, determining 120 the two metrics may include a preliminary step of determining a global CN ratio profile for the genome based on the shallow WGS data. In an embodiment, this determination may be performed only when the average depth of the shallow WGS data is greater than a predefined threshold (e.g., 0.3X) to ensure sufficient quality of the data to be analyzed.

[0107] In addition, the shallow WGS data can be used to obtain at least one CNV indicator. In this case, the WGS data can be used to determine (step 130) whether there is a copy number variation (CNV) in a predefined first set of genes (also referred to as a CNV gene set) by estimating the CN of these genes in the predefined first set of genes. It is recalled that CNV refers to the CN of a gene or a portion of a gene in the genome that varies from subject to subject. Such variation can generally be caused by the loss or gain of a DNA fragment. Many CNVs in CNV have no effect on health, but some CNVs may be associated with disease.

[0108] In the present invention, the first gene set for which CNV determination (step 130) is performed corresponds to genes associated with the pathology being studied. For example, in the context of a tumor associated with the presence of HRD (e.g., in ovarian cancer, breast cancer, prostate cancer, or pancreatic cancer), the first gene set may include up to 11 genes selected from AKT1, BARD1, CCNE1, EMSY, ESR1, H2AX, MRE11, PTEN, RAD51B, RAD52, and RAD54. The role of all of these genes in the HRD process is well known. Of course, the first gene set is likely to evolve as genetic knowledge evolves. In addition, some genes may be more or less relevant depending on the sex of the individual and / or the location of the tumor (e.g., breast versus ovary). In some embodiments of the present invention, CNV determination (step 130) is performed for all of these eleven genes. In an alternative embodiment, CNV determination (step 130) is performed only for a subset of the first gene set, the subset comprising at least two genes, at least three genes, or at least four genes. In some embodiments, CNV determination is performed for at least CCNE1 and RAD51B. In one example, CNV determination is performed for gene CCNE1 and gene RAD51B (step 130), which are advantageously shown to provide better performance in identifying HRD relative to other genes in the first gene set.

[0109] In one or more embodiments, for genes of a predefined first gene set (all predefined first sets or only a subset of the predefined first set), step 130 may include determining a CNV indicator associated with the amplification level of the corresponding gene. For example, for each gene, the CNV indicator may be the CN of the gene, or a binary value indicating whether the gene has a CNV, or a ratio between the CN of the gene in the sample and the average CN of the sample genome. Alternatively, the CNV indicator may be the number of genes in which CNVs are detected in the first gene set (i.e., the CNV gene set) or a subset of the first gene set.

[0110] In one or more embodiments, at least one panel indicator can be determined based on the panel data 114. It is noteworthy that in order to calculate the panel indicator, the variation present on the second gene set can be determined (step 140) based on the panel data 114 (i.e., second sequencing data from a non-shallow sequencing process of a set of fragments in the subject's genome, the set of fragments including the gene panel set). The second gene set (also referred to as the "gene panel set") corresponds to genes belonging to the portion where the panel data 114 has been obtained, i.e., the portion of the genome where the classical sequencing process has been performed. The second gene set includes genes associated with the cancer tumor type being studied. Advantageously, the genes of the first set and the second set are mutually exclusive, meaning that the second gene set includes genes that are not in the first gene set.

[0111] For example, in the context of tumors associated with the presence of HRD (e.g., in ovarian, breast, prostate, or pancreatic cancer), the second gene set may include BRCA1 and BRCA2, which play well-known roles in the HRR process. The second gene set may also include genes known to have a certain pathogenicity (e.g., pathogenicity probability above a predefined threshold) based on biological and clinical references or proprietary classifications of somatic variants, which can also provide clinical interpretation and annotations for somatic variants. These annotations can include, for example, the corresponding pathogenicity probability of the variant for a given pathology.

[0112] In one or more embodiments, the somatic variation identified in step 140 may be an insertion or deletion (Indel), a single nucleotide variation (SNV), and / or a multi-nucleotide variation (MNV). In addition, the somatic variation identified in step 140 may be a short variation, i.e., a variation involving a small length of DNA (e.g., less than 10 base pairs).

[0113] In one or more embodiments, the identification of somatic variants (step 140) may include one or more panel indicators indicating the presence of variants with pathogenicity levels within a predefined threshold. In particular, the identification of somatic variants may include determining the number of variants associated with predefined pathogenicity categories. For example, three pathogenicity categories may be defined: pathogenic, likely pathogenic, and variants of uncertain significance (VUS). This reclassification of pathogenicity is known to those skilled in the art, and a given variant may be determined to belong to one of the categories based on a reference or proprietary classification. Based on these categories, three corresponding panel indicators may be defined, each panel indicator representing the number of variants belonging to a corresponding category in the second gene set. In an embodiment, the gene BRCA1 and the gene BRCA2 may be considered different from other genes in the second gene set, and the fourth panel indicator may be defined as the number of variants in these BRCA1 and BRCA2 genes. Alternatively, the fourth panel indicator and the fifth panel indicator may be defined corresponding to the number of variants in the BRCA1 gene and the BRCA2 gene, respectively. In some embodiments, these numbers may represent the number of variants classified as pathogenic or the number of variants classified as (pathogenic or likely pathogenic) in the BRCA1 gene and / or the BRCA2 gene, etc.

[0114] It is worth noting that step 120 , step 130 , and step 140 may be performed in parallel or sequentially in any order.

[0115] Once the metrics / indicators have been obtained in steps 120, 130, and 140, these metrics / indicators are provided as input data (step 150) to a trained HRD prediction machine learning (ML) model to obtain an HRD index representing the presence of homologous recombination deficiency (HRD) in the genome of the subject as an output. The output (i.e., the HRD index) can represent the predicted state 155 of the subject regarding HRD (e.g., HRD positive or HRD negative). Alternatively or in addition, the output of the HRD prediction ML model can be the probability that the subject has HRD. In an embodiment, if the probability of the presence of HRD is between 0.5 plus or minus a predefined threshold (e.g., 0.2), the result can be considered a "failure" because the probability is too close to 0.5 to confirm or exclude the presence of HRD.

[0116] It is worth noting that Figure 1The method for determining the presence of HRD in a subject may be performed based on only some of the metrics / indicators described above. For example, in some embodiments, LGA and / or pLOH may not be used as input to the HRD prediction ML model. Therefore, in these embodiments, step 120 is only partially performed or skipped entirely. In other embodiments, only one or two of the metrics determined in step 120 based on the shallow WGS data 112 and one or more panel indicators determined in step 140 based on the panel data 114 are used. In such embodiments, step 130 is skipped.

[0117] In one or more embodiments, the HRD prediction ML model may have been previously trained in a fully supervised manner based on a training data set obtained from a set of sequencing data corresponding to a tumor sample whose true status (HRD positive or negative) is known for the subject. For example, the true status may correspond to the result of a reference HRD test that is considered a gold standard in the prior art. In this embodiment, in addition to at least one first parameter and at least one panel indicator, each training data subset of the training data set (associated with each subject in the plurality of subjects) also includes information representing the true HRD status of the subject (i.e., HRD positive or negative).

[0118] In one or more embodiments, the HRD prediction ML model may be associated with a regression algorithm, such as ridge regression (based on L2 norm), Lasso regression (based on L1 norm), or elastic net regularization (combining the L1 penalty and L2 penalty of the Lasso and ridge algorithms). Other ML algorithms may be used, including regression algorithms or classification algorithms, such as ensemble models, support vector machines, neural networks, etc. Such ML models are known and will not be described in detail in this specification.

[0119] As described above, the HRD prediction ML model can be trained in a preliminary stage based on a set of sequencing data corresponding to a tumor sample whose true state can be known according to standard methods in the prior art. The sequencing data may typically include shallow WGS data and panel data. In some embodiments, filtering of data quality may be applied to retain only samples with sufficient depth (e.g., at least 0.1X for shallow WGS data) and / or a sufficient percentage of correctly mapped reads (e.g., at least 50% for both shallow WGS data and panel data). In addition, when the state of a sample cannot be obtained by the gold standard, for example due to low tumor cell or DNA yield, the sample may be discarded and therefore cannot be used to train the ML model.

[0120] For example, an ML model can be trained by using two training datasets, a set of 77 samples from the XTH-HS1 Agilent DNA kit and a set of 94 samples from the XTH-HSS2 Agilent DNA kit.

[0121] Metrics and indicators can then be determined from the sequencing data as described in detail above (i.e., in a similar manner to the run phase) to be used as training data for the ML model. Due to the nature of the model and the differences in the inputs, normalization can be performed on the data set to account for differences in variance. Before learning the ML model, each input group can be set to zero mean and unit variance. The ML model can then be trained using an optimization algorithm (e.g., the SAGA algorithm).

[0122] To evaluate the ability of the above methods to detect HRD in subjects, a leave-one-out cross validation approach was used. The ML model was trained using training data from 63 women with ovarian cancer. The training data included all of the metrics / indicators mentioned above. Of the 63 women, 56 obtained results similar to the gold standard, which indicates that Figure 1 The accuracy of the new method is about 89%. It is worth noting that the gold standard itself may eventually produce false positive or false negative results. Therefore, among the 7 patients who obtained different HRD results between the new method and the gold standard, some patients may actually correspond to the gold standard error, which will further improve the accuracy of the new method for HRD determination.

[0123] Figure 2 An example of an apparatus for detecting the presence or absence of HRD in a subject according to an embodiment of the present invention is shown.

[0124] In these embodiments, the device 200 includes a computer including a memory 201 for storing program instructions that can be loaded into the circuit and adapted to cause the circuit 202 to execute when the circuit 202 executes the program instructions. Figure 1 The memory 201 may also store data and useful information for executing the above-mentioned steps of the present invention, in particular, a training ML model for predicting the HRD status of a subject.

[0125] For example, circuit 202 may be:

[0126] • a processor or processing unit adapted to interpret instructions in a computer language, which processor or processing unit may comprise, be associated with or be attached to a memory containing instructions, or

[0127] • a processor / processing unit associated with a memory, the processor or processing unit being adapted to interpret instructions in a computer language and the memory containing said instructions, or

[0128] • an electronic card in which the steps of the invention are described in silicon, or

[0129] • Programmable electronic chips, for example, FPGA chips (for «Field Programmable Gate Array»).

[0130] The computer may further include an input interface 203 for receiving sequencing data of a subject and an output interface 204 for providing the HRD status of the subject.

[0131] To facilitate interaction with the computer, a screen 205 and a keyboard 206 may be provided and connected to the computer circuitry 202 .

[0132] also, Figure 1 The flowchart shown in can represent all or part of the steps of a program that can be executed by a processor located in a transmitter. Figure 1 A flowchart may correspond to a general algorithm of a computer program within the meaning of the present invention.

[0133] Those skilled in the art will readily appreciate that the various parameters disclosed in the specification may be modified without departing from the scope of the invention, and the various embodiments of the present disclosure may be combined. Of course, the present invention is not limited to the exemplary embodiments described above. The present invention can be extended to other variants.

Claims

1. A device for determining an HRD index indicating the presence of homologous recombination deficiency (HRD) in the genome of a subject, the device comprising: • an input interface for receiving first sequencing data from a shallow whole genome sequencing (WGS) process of the subject's genome and second sequencing data from a non-shallow sequencing process of a set of fragments in the subject's genome, the set of fragments comprising a set of genes referred to as a gene panel set; • At least one processor configured to: o Obtain at least one first parameter according to the first sequencing data, wherein the at least one first parameter is among the following parameters: a first measurement representing a large-scale genomic alteration (LGA) in the genome of the subject; a second measurement result representing the number of fragments in the genome of the subject that have partial chromosome deletions; as well as at least one CNV indicator representing the amplification level of genes in a predefined gene set referred to as the CNV gene set; o obtaining at least one panel indicator according to the second sequencing data, wherein the at least one panel indicator represents somatic variation on genes of the gene panel set; as well as o determining the HRD index indicative of the presence of HRD in the subject's genome based on a trained HRD prediction machine learning model applied to at least a first parameter and at least one panel indicator, The trained HRD prediction machine learning model has been obtained by training the machine learning model using a training data set, wherein the training data set is obtained from a first sequencing data of a shallow whole genome sequencing WGS process from a plurality of subject genomes, and a second sequencing data of a non-shallow sequencing process from a set of fragments in the genomes of the plurality of subjects, the set of fragments comprising a gene set referred to as a gene panel set; wherein the training data set comprises a plurality of training data subsets, each subset being associated with one of the plurality of subjects, and comprising: • obtaining at least one corresponding first parameter according to the first sequencing data, wherein the at least one corresponding first parameter is among the following parameters: o a respective first measurement representing a number of large-scale state transitions LST in the genome of the subject of the plurality of subjects; o a second measurement result representing the number of fragments in the genome of the subject among the plurality of subjects that have a partial deletion of the chromosome; and o at least one CNV indicator representing the amplification level of genes in a predefined gene set referred to as the CNV gene set; • At least one corresponding panel indicator obtained from the second sequencing data, wherein the at least one corresponding panel indicator represents somatic variations on genes in the gene panel set.

2. The device according to claim 1, wherein: At least one processor is configured to obtain the first measurement result, the second measurement result, and the at least one CNV indicator, and wherein the HRD prediction machine learning model is applied to the first measurement result, the second measurement result, the at least one CNV indicator, and the at least one panel indicator.

3. The device according to any one of claims 1 and 2, wherein: The gene panel set is mutually exclusive with the CNV gene set.

4. The device according to any one of the preceding claims, wherein: The CNV gene set includes at least one of the following genes: AKT1, BARD1, CCNE1, EMSY, ESR1, H2AX, MRE11, PTEN, RAD51B, RAD52, RAD54.

5. Apparatus according to any one of the preceding claims, wherein: The gene panel set includes the BRCA1 gene and / or the BRCA2 gene.

6. Apparatus according to any one of the preceding claims, wherein: The first measurement result representing a large-scale genomic alteration LGA in the genome of the subject is determined based on the number of pairs of adjacent fragments, wherein each fragment in each pair is at least 10 Mb long and the fragments in each pair have different copy numbers CN.

7. Apparatus according to any one of the preceding claims, wherein: The second measurement result representing the number of fragments in the subject's genome in which a partial deletion of a chromosome occurs is determined based on the number of genomic fragments that are at least 10 Mb long and have a copy number of 0.5 to 1.

5.

8. Apparatus according to any one of the preceding claims, wherein: The at least one CNV indicator includes at least one of the following: • for each gene of the CNV gene set, the corresponding copy number of the gene; • for each gene of the CNV gene set, a binary value indicating whether a copy number variation CNV of the gene is detected; • for each gene of the CNV gene set, a ratio between the copy number of the gene and the average copy number of the gene in all the CNV gene sets; • The number of genes in the CNV gene set where CNVs were detected.

9. Apparatus according to any one of the preceding claims, wherein: The at least one panel metric comprises a counter representing a number of genes in the gene panel set that have somatic variation.

10. Apparatus according to any one of the preceding claims, wherein: The subject suffers from a disease, wherein the at least one processor is further configured to: • Identify genes whose corresponding pathogenicity level for said disease is above a predefined pathogenicity threshold based on at least one classification of somatic variants; Wherein, the gene panel set includes the determined genes.

11. Apparatus according to any one of the preceding claims, wherein: The genes of the gene panel set are divided into several categories of pathogenicity levels, wherein, for each category of pathogenicity levels, the at least one panel indicator includes a corresponding counter, which represents the number of genes with somatic mutations in each category of pathogenicity levels.

12. Apparatus according to any one of the preceding claims, wherein: The HRD index representing the presence of HRD in the subject's genome is a binary variable indicating whether HRD exists in the subject's genome.

13. Apparatus according to any one of the preceding claims, wherein: The at least one processor is also configured to determine a probability of the presence of HRD in the subject's genome based on the HRD prediction machine learning model applied to the at least first parameter and the at least one panel indicator.

14. Apparatus according to any one of the preceding claims, wherein: The HRD prediction machine learning model is a regression model.

15. Apparatus according to any one of the preceding claims, wherein: The HRD prediction machine learning model is trained in a fully supervised manner.

16. A device for obtaining an HRD prediction model, wherein the HRD prediction model is used to determine an HRD index indicating the presence of homologous recombination defect (HRD) in the genome of a subject under study, the device comprising: - at least one input configured to receive first sequencing data from a shallow whole genome sequencing (WGS) process of genomes of a plurality of subjects and second sequencing data from a non-shallow sequencing process of a set of fragments in the genomes of the plurality of subjects, the set of fragments comprising a set of genes referred to as a gene panel set; - at least one processor configured to: • generating a training data set using the first sequencing data and the second sequencing data of the plurality of subjects, the training data set comprising a plurality of training data subsets; wherein generating the training data set comprises generating at least one of the training data subsets for each subject in the plurality of subjects, the training data subsets comprising: o At least one first parameter obtained according to the first sequencing data, wherein the at least one first parameter is among the following parameters: a corresponding first measurement representing the number of large-scale genomic alterations (LGAs) in the genome of the subject; a second measurement result representing the number of fragments in the genome of the subject in which a partial deletion of a chromosome occurs; and at least one CNV indicator representing the amplification level of genes in a predefined gene set referred to as the CNV gene set; o at least one corresponding panel indicator obtained from the second sequencing data, the at least one corresponding panel indicator representing somatic variations on genes of the gene panel set; - using the training data set to train a machine learning model to obtain the trained HRD prediction model, wherein the trained HRD prediction model is configured to receive at least one first parameter and at least one panel indicator of the subject under study as input, and provide the HRD index representing the presence of homologous recombination deficiency HRD in the genome of the subject under study as output.

Images