Methylation-based age prediction as feature for cancer classification

Abstract

Methods and systems for making covariant predictions from methylation features are disclosed. A system for identifying a feature set for a genomic region by training one or more regression to evaluate a covariance score for the genomic region. The system may select a feature set with the highest cue score, and may account for other selection criteria. The system trains an age prediction model using training samples with one or more reporting timing age tags. The system may further utilize timing age prediction to predict the likelihood of cancer in the test sample. To this end, the system may compare the predicted covariable value and / or tag to the reported value and / or tag. In one embodiment, the system may utilize an age residual threshold to determine whether there is a strong likelihood of cancer presence. In other embodiments, the system may provide predicted timing age values as features to a cancer classifier.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 392,980, filed on July 28, 2022, which is incorporated by reference in its entirety.

[0003] Background technology and technical field

[0004] Deoxyribonucleic acid (DNA) methylation plays an important role in regulating gene expression. Abnormal DNA methylation has been associated with many disease processes (including cancer). DNA methylation profiling using methylation sequencing (e.g., whole genome bisulfite sequencing (WGBS) or targeted methylation sequencing) is increasingly being considered an important diagnostic tool for detecting, diagnosing and / or monitoring cancer. For example, specific patterns in differentially methylated regions and / or allele-specific methylation patterns can be used as molecular markers for non-invasive diagnosis using circulating free (cf) DNA. As part of cancer classification, it is still necessary to understand the effects that covariates (or more generally, variables indicating cancer or non-cancer) may have on the human genome. In addition, it is still necessary to be able to distinguish variables that can indicate cancer and / or some other biological attributes (e.g., age or sex).

[0005] The present disclosure is directed to solving the above challenges. The background description provided herein is intended to generally introduce the background information of the present disclosure. Unless otherwise indicated herein, the materials described in this section are not prior art to the claims of this application and will not be considered as prior art or inspiration of prior art by inclusion in this section. Summary of the invention

[0006] In some aspects, the technology described herein relates to a method for detecting the presence or absence of cancer in a test sample. The method includes obtaining a plurality of training samples, each training sample: including a plurality of nucleic acid fragments, wherein each of these plurality of nucleic acid fragments has a genomic position overlapping at least one genomic region in a plurality of genomic regions, and is marked with the chronological age of the individual providing the training sample. The method includes, for each training sample, sequencing a plurality of nucleic acid fragments to identify the methylation pattern of each nucleic acid fragment. For each genomic region in a plurality of genomic regions, the method identifies a nucleic acid fragment having a genomic position overlapping the genomic region from a plurality of nucleic acid fragments, and for the genomic region, based on the chronological age of the individual providing the identified nucleic acid fragment and the methylation pattern of the identified nucleic acid fragment, calculates an indicative score representing the correlation between the chronological age and the methylation pattern. The method includes generating a feature set, the feature set including one or more genomic regions in a plurality of genomic regions, and one or more genomic regions in the feature set having an indicative score above a threshold. The method includes training a machine learning age prediction model to determine a predicted chronological age of a test individual who provides a test sample, wherein the training is performed based on methylation patterns of nucleic acid fragments in multiple training samples that overlap with one or more genomic regions in a feature set.

[0007] In some aspects, the method further includes training a linear regression for each genomic region in the feature set, the training being based on the methylation patterns of nucleic acid fragments overlapping each genomic region from a plurality of training samples labeled as non-cancer. The method obtains a plurality of additional training samples. Each additional training sample: comprises a plurality of additional nucleic acid fragments, the plurality of additional nucleic acid fragments having additional genomic locations overlapping at least one genomic region in the plurality of genomic regions, labeled with the chronological age of the individual providing the additional training sample, and labeled as non-cancer or cancer based on a previous determination of the presence of cancer in the additional training sample. The method includes sequencing a plurality of additional nucleic acid fragments to identify the methylation pattern of each additional nucleic acid fragment. For each genomic region of the plurality of genomic regions, the method applies linear regression to the methylation patterns of nucleic acid fragments of the plurality of additional training samples to determine a predicted chronological age of the individual providing the additional training samples, calculates an age residual for each additional training sample, the age residual being the difference between its predicted chronological age and its labeled chronological age, and compares the age residuals of the additional training samples labeled as cancer with the age residuals of the additional training samples labeled as non-cancer. The method includes generating a reduced feature set from the feature set based on the comparison of the age residuals, wherein the reduced feature set includes fewer genomic regions than the feature set, and the reduced feature set is used to train a machine learning age prediction model.

[0008] In some aspects, the method further includes obtaining a test sample, the test sample includes a plurality of additional nucleic acid fragments and is labeled with the chronological age of the test subject providing the test sample. The method includes sequencing a plurality of additional nucleic acid fragments for the test sample to identify the methylation patterns of these plurality of additional nucleic acid fragments. The method includes applying a trained age prediction model based on the methylation patterns of additional nucleic acid fragments overlapping with one or more genomic regions in the feature set to determine the predicted chronological age of the test subject providing the test sample, and calculating an age residual, the age residual being the difference between the labeled chronological age of the test subject and the predicted chronological age. The method includes determining that the test sample has a strong possibility of the presence of cancer when the age residual is determined to be above a residual threshold.

[0009] In some aspects, the method determines the residual threshold by applying the trained age prediction model to a second plurality of training samples identified as non-cancer to determine a predicted age for each of the second plurality of training samples, and calculating an age residual for each of the second plurality of training samples by comparing the predicted age with the labeled temporal age of the second plurality of training samples. The determining includes determining the residual threshold based on the age residuals calculated for the second plurality of training samples, wherein the age residuals calculated for the second plurality of training samples at least mostly satisfy the residual threshold.

[0010] In some aspects, the method further includes, when determining that the test sample has a strong possibility of the presence of cancer: using a p-value screening method to screen the methylation patterns of multiple additional nucleic acid fragments to identify a set of abnormal methylation patterns; generating a feature vector for the test sample based on age residuals and the set of abnormal methylation patterns; and determining a cancer prediction for the test sample by inputting the feature vector into a trained cancer classifier.

[0011] In some aspects, the cancer prediction used in the methods is a binary prediction between the presence and absence of cancer or another disease state.

[0012] In some aspects, the cancer prediction used in the methods is a multi-classification prediction between multiple cancer types.

[0013] In some aspects, the cancer prediction used in the methods is a multi-classification prediction between multiple disease states.

[0014] In some aspects, the method further includes determining the presence of cancer in the test sample using a secondary machine learning cancer classifier, wherein the secondary cancer classifier is configured to receive as input the predicted chronological age of the subject and the methylation patterns of multiple additional nucleic acid fragments, and output a prediction of the presence of cancer in the test sample.

[0015] In some aspects, the secondary machine learning cancer classifier is further configured to receive as input the clinical information and genetic background of the subject and output a prediction of the presence of cancer in the test sample.

[0016] In some aspects, the indicative score is a Pearson correlation or a covariance score.

[0017] In some aspects, the indicative score is determined by training a linear regression to regress chronological age on methylation density of non-cancer training samples, wherein the methylation density is calculated as the percentage of nucleic acid fragments having a genomic location that overlaps a particular genomic region and is methylated in the particular genomic region.

[0018] In some aspects, the machine learning age prediction model includes a multivariate regression. The multivariate regression can be penalized based on the number of one or more genomic regions in the feature set. The machine learning age prediction model can receive as input the methylation density corresponding to each genomic region in the feature set.

[0019] In some aspects, the number of one or more genomic regions in the feature set is selected from the range of 5-10,000. Further, sequencing the nucleic acid fragments comprises whole genome bisulfite sequencing (WGBS) and / or sequencing the nucleic acid fragments comprises targeted sequencing.

[0020] In some aspects, each training sample in the plurality of training samples was previously determined to not contain a presence of cancer or each training sample in the plurality of training samples was previously determined to contain a presence of cancer. Further, each training sample in the plurality of training samples was previously determined to contain a presence of cancer or an absence of cancer. Each training sample in the plurality of training samples is labeled as having a presence of cancer or not having a presence of cancer, the labeling being based on a previous determination of the cancer status of the training sample.

[0021] In some aspects, the technology described herein relates to a method for training a classifier. The method includes obtaining multiple training samples, each training sample: including multiple nucleic acid fragments, each of which has a genomic position overlapping at least one genomic region in multiple genomic regions, and is marked with the characteristics of the individual providing the training sample. The method includes, for each training sample, sequencing multiple nucleic acid fragments to identify the methylation pattern of each nucleic acid fragment. For each genomic region in multiple genomic regions, the method identifies a nucleic acid fragment with a genomic position overlapping the genomic region from multiple nucleic acid fragments, and for the genomic region, based on the characteristics of the individual providing the identified nucleic acid fragment and the methylation pattern of the identified nucleic acid fragment, calculates an indicative score representing the correlation between the feature and the methylation pattern. The method includes generating a feature set, the feature set including one or more genomic regions in multiple genomic regions, and one or more genomic regions in the feature set have an indicative score higher than a threshold. The method includes training a machine learning feature prediction model to determine the predicted features of the tested individual providing the test sample, and the training is based on the methylation pattern of the nucleic acid fragments overlapping one or more genomic regions in the feature set in multiple training samples.

[0022] In some aspects, the characteristic is the biological sex of the individual, and the characteristic is biological male or biological female.Alternatively or additionally, the characteristic is the smoking status of the individual, and the characteristic is smoking or non-smoking.

[0023] In some aspects, the machine learning feature prediction model includes logistic regression using a sigmoid function.

[0024] In some aspects, the method further includes obtaining a test sample, the test sample including a plurality of additional nucleic acid fragments and being labeled with a tag indicating a characteristic of the test sample. The method further includes sequencing a plurality of additional nucleic acid fragments of the test sample to identify a test methylation pattern of each additional nucleic acid fragment, and applying a trained machine learning feature prediction model based on the methylation patterns of these additional nucleic acid fragments overlapping with a feature set of a genomic region to predict the characteristics of the test sample. If the predicted label is different from the label of the test sample, the method includes marking the test sample as a contaminated sample and stopping further analysis of the test sample.

[0025] In some aspects, if the predicted signature matches the signature of the marker, the method further includes screening the methylation patterns of additional nucleic acid fragments of the test sample using a p-value screening method to identify a set of abnormal methylation patterns; generating a feature vector for the test sample based on the set of abnormal methylation patterns; and determining a cancer prediction for the test sample by inputting the feature vector into a trained cancer classifier.

[0026] In some aspects, the cancer prediction is a binary classification prediction between the presence and absence of cancer or another disease state. The cancer prediction can be a multi-classification prediction between multiple cancer types or multiple disease states.

[0027] In another aspect, the technology described herein relates to a system comprising a hardware processor and a non-transitory computer-readable storage medium storing instructions that, when executed by the hardware processor, cause the hardware processor to perform the methods disclosed herein. Similarly, when the instructions are executed by one or more processors, the non-transitory computer-readable storage medium storing the instructions causes the processors to perform the methods disclosed herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is an exemplary flow chart that describes an overall workflow for classifying cancer in a sample according to one or more embodiments.

[0029] Figure 2A is an exemplary flow chart describing a process for sequencing cell-free (cf) DNA fragments to obtain methylation state vectors according to one or more embodiments.

[0030] Figure 2B According to one or more embodiments Figure 2A An exemplary illustration of the process shown involves sequencing fragments of cell-free (cf) DNA to obtain methylation state vectors.

[0031] Figure 3A Methylation signatures according to one or more embodiments are presented that can be derived from a single CpG site as a genomic region.

[0032] Figure 3B Methylation signatures according to one or more embodiments are presented, which may be derived from a plurality of CpG sites as genomic regions.

[0033] Figure 4A is an exemplary flow chart describing a process of training a temporal age prediction model according to one or more embodiments.

[0034] Figure 4BDeployment of a temporal age prediction model according to one or more embodiments is demonstrated.

[0035] Figure 5A is an exemplary flow chart describing a process for generating a control group data structure for determining abnormally methylated fragments according to one or more embodiments.

[0036] Figure 5B is an exemplary flow chart describing a process for determining that a fragment is abnormally methylated based on a control group data structure according to one or more embodiments.

[0037] Fig. 6A is an exemplary flow chart describing a process for training a cancer classifier according to one or more embodiments.

[0038] Figure 6B Examples of generating feature vectors according to one or more embodiments are shown, and the feature vectors are used to train a cancer classifier.

[0039] Fig. 7A Exemplary flow diagrams of apparatuses for sequencing nucleic acid samples according to one or more embodiments are shown.

[0040] Figure 7B is an exemplary block diagram of an analysis system according to one or more embodiments.

[0041] Figure 8 Age-associated genomic regions are presented according to one or more example embodiments.

[0042] Fig.9A A process for identifying a set of features of genomic regions that can provide age covariate information according to one or more example embodiments is presented.

[0043] Fig. 9B A graph showing the results of age prediction for a non-cancer holdout cohort according to example embodiments.

[0044] Fig. 9C A graph showing results of age prediction for a cancer cohort according to example embodiments.

[0045] Fig. 10A Another process for identifying a set of features of genomic regions that can provide age covariate information according to one or more example embodiments is presented.

[0046] Fig. 10B A graph showing results of age prediction for a non-cancer holdout cohort according to an example embodiment.

[0047] Fig. 10C A graph showing results of age prediction for a cancer cohort according to example embodiments.

[0048] Fig.11 The distribution of the test cohort across cancer stages according to example embodiments is shown.

[0049] Fig. 12A The top series of graphs are shown representing test samples predicted to be negative by the cancer classifier according to some example embodiments.

[0050] Fig. 12B The bottom series of graphs are shown representing test samples predicted to be positive results by the cancer classifier according to some example embodiments.

[0051] Fig.13 A genomic region showing a slowed age increase in different cancer types is shown according to example embodiments.

[0052] Fig.14A Results are shown for a first genomic region that appears to consistently show accelerated age progression across multiple blood cancer types, but less consistently across multiple other non-blood cancer types, according to example embodiments.

[0053] Fig. 14B Results are shown for a second genomic region according to example embodiments, which appears to consistently show a slowed age progression in multiple blood cancer types, but has little to no apparent slowed age progression in multiple other non-blood cancer types.

[0054] Fig.15A A set of features for identifying genomic regions for predicting biological sex according to one or more example embodiments is presented.

[0055] Fig. 15B Results from a trained biological sex prediction model according to example embodiments are presented.

[0056] Fig.16A A set of features for identifying genomic regions for predicting smoking status according to one or more example embodiments is presented.

[0057] Fig. 16B Results from a trained smoking status prediction model according to example embodiments are presented.

[0058] These figures depict various embodiments for purposes of illustration only. One skilled in the art will readily recognize from the following discussion that alternative embodiments of the structures and methods described herein may be employed without departing from the principles described herein. DETAILED DESCRIPTION

[0059] I. Overview

[0060] Early detection and classification of cancer is an important technology. Being able to detect cancer before it becomes symptomatic is beneficial to all parties involved, including patients, physicians, and loved ones. For patients, early cancer detection allows for a greater chance of a beneficial outcome; for physicians, early cancer detection allows for more treatment pathways that may lead to a beneficial outcome; and for loved ones, early cancer detection increases the likelihood of not losing friends and family members to cancer.

[0061] Recently, early cancer detection technology has moved toward analyzing genetic fragments (e.g., DNA) from a person's (e.g., blood) to determine if any of those genetic fragments originate from cancer cells. These new technologies allow doctors to identify cancer in a patient that might not otherwise be detected. For example, consider the case of a person who is at high risk for breast cancer. Traditionally, this person would visit her doctor regularly for a mammogram, which would produce images (e.g., take X-ray images) of her breast tissue, which the doctor uses to identify cancerous tissue. Unfortunately, even with the highest-resolution mammograms, doctors can only identify a tumor when it is about one millimeter in size. This means that the cancer has been in the person's body for some time, undiagnosed and untreated. For most cancers, visual identification like this is common - that is, the tumor can only be identified when it has grown large enough to be identified by some imaging technology.

[0062] Cancer detection by analyzing genetic fragments from a patient (e.g., blood) alleviates this problem. For ease of explanation, assume that once cancer cells are formed, DNA fragments will be shed into the person's bloodstream. This happens when the number of cancer cells is still very small and before they can be observed by imaging technology. Therefore, with appropriate methods, a system that analyzes DNA fragments in the bloodstream can identify the presence of cancer in the human body based on the shed cancer DNA fragments, and more importantly, the system can identify the cancer before more traditional cancer detection technologies.

[0063] Cancer detection based on DNA fragment analysis can be achieved through next-generation sequencing ("NGS") technology. Broadly speaking, NGS is a group of technologies that can perform high-throughput sequencing of genetic material. As discussed in more detail herein, NGS mainly consists of (1) sample preparation, (2) DNA sequencing, and (3) data analysis. Sample preparation is the laboratory method required to prepare DNA fragments for sequencing, sequencing is the process of reading the ordered nucleotides in a sample, and data analysis is the processing and analysis of the genetic information in the sequencing data to identify the presence of cancer.

[0064] While these steps of NGS may help achieve early cancer detection, they themselves introduce complex and harmful issues to cancer detection, so any improvements in sample preparation, DNA sequencing, and / or data analysis (including preprocessing, algorithm processing, and summary or presentation of predictions or conclusions) will improve cancer detection technology and early cancer detection more broadly.

[0065] For illustration, for example, (1) problems with sample preparation include DNA sample quality, sample contamination, fragmentation bias, and accurate indexing. Solving these problems will provide better genetic data for cancer detection. Similarly, (2) problems with sequencing include, for example, errors in accurate transcription of fragments (e.g., reading "C" as "A", etc.), incorrect or difficult fragment assembly and overlap, inconsistent coverage uniformity, the trade-off between sequencing depth and cost and specificity, and insufficient sequencing length. Likewise, solving any of these problems will provide improved genetic data for cancer detection.

[0066] (3) The problems in data analysis are the most arduous and complex. The large amount of data generated by NGS sequencing technology brings many challenges. The genetic data sets generated usually reach the terabyte level. The effective and efficient analysis of such a large amount of data requires high requirements for both procedures and calculations. For example, analyzing NGS sequencing involves several baseline processing steps, such as aligning reads with each other, aligning and mapping reads to the reference genome, identifying and recognizing variant genes, identifying and recognizing abnormally methylated genes, generating functional annotations, etc. Performing any of these processing steps on terabyte-level genetic data is computationally expensive even for the most powerful computer architecture, and is completely impossible for a normal human brain. In addition, since the sample preparation and sequence reading processes are prone to errors, the genetic sequencing data derived from these processes, that is, a large portion of the genetic data obtained, may be of low quality or unusable for cancer identification. For example, a large amount of genetic data may include contaminated samples, transcription errors, mismatched regions, over-represented regions, etc., and may not be suitable for high-accuracy cancer detection. Identifying and processing low-quality genetic data from the large amount of genetic data obtained from NGS sequencing is also extremely difficult to achieve procedurally and computationally, and it is not practical for the human brain to perform. Overall, any process that helps to process large-scale sequencing data more efficiently will improve cancer detection using NGS sequencing.

[0067] Finally, and perhaps most importantly, it is also very difficult to accurately identify abnormal DNA from NGS data to identify the presence of cancer. In order to achieve effective detection, algorithms are being sought to compensate for errors such as those caused by sample preparation and sequencing, and to overcome the large-scale data analysis problems brought about by NGS technology. That is, when designing one or more machine learning models or other computational processing algorithms to achieve early cancer detection based on next-generation sequencing technology, they must be configured to take into account the problems brought about by those technologies. Some of those technologies and models are discussed below, and specific improvements to the state-of-the-art technologies and models are further discussed.

[0068] As mentioned above, one of the problems faced when establishing and properly applying cancer detection models is the large amount of sequencing data that the model may have to process. Among these sequencing data, there are a large number of "cancer signals" and "other signals", where cancer signals refer to sequencing data that indicate whether cancer exists in the test subject, and other signals refer to sequencing data that indicate other biological or clinical characteristics of the test subject (e.g., age, gender, smoker, etc.). Some other signals may appear to share similar features with cancer signals, or some cancer signals may appear to share similar features with other signals, making this problem more complicated. Further, some sequencing data may be both cancer signals and other signals. For example, a methylation pattern associated with age may look similar to a methylation pattern that occurs due to age, or, from a similar perspective, a specific methylation pattern may indicate both cancer and age.

[0069] This interaction of one or more signals in the sequencing data is troublesome because it can lead to erroneous cancer determinations. For example, an improperly trained cancer classifier might determine that a person has a relevant cancer signal based on an analysis of their cfDNA because the methylation of some of the cfDNA appears to be “both old and cancerous,” when in fact, the subject is simply old. Therefore, when the sequencing data can suggest several closely related biological processes or clinical features, great caution should be exercised in selecting sequencing data for training a cancer classifier. In these cases, any method that can improve cancer determination represents an improvement in the field of early cancer detection.

[0070] In addition, carefully curating a set of data that can accurately indicate cancer and / or other biological and / or clinical characteristics can reduce the complexity and corresponding computational cost of running a machine learning model to indicate or predict the presence of an identified cancer in a sample (i.e., "identify" cancer). In fact, by carefully selecting genomic sites that are "strongly" associated with cancer and / or a biological or clinical feature for the model to process, the computational load of the model will be reduced, resulting in higher speed and efficiency, which represents an improvement in the field of cancer determination. For example, consider the above-mentioned example machine learning model that is trained to identify cancer based on methylation. However, in this case, each suggestive feature is also related to the "strength" of cancer suggestion of the site, or to the "strength" of chronological age suggestion of the site. That is, abnormal methylation at the first site may strongly indicate cancer, while abnormal methylation at the second site may strongly or weakly indicate chronological age, etc. In this case, letting the machine learning model process genomic sites that only "weakly" indicate cancer will incur computational costs, but will not bring corresponding benefits to the accuracy or specificity of cancer determination.

[0071] To give a quantitative example, applying the machine learning model to 100 weakly suggestive sites may not bring as much benefit as applying the machine learning model to 1 strong suggestive site. Therefore, selecting appropriate sites to construct feature sets for machine learning models to process to identify the presence of cancer and / or biological or clinical features will greatly reduce processing costs without greatly reducing the accuracy or specificity of the model. In fact, this may reduce the analysis load from, for example, millions of genomic sites to tens of thousands of genomic sites without sacrificing the accuracy of the model. More simply, reducing the analysis load from millions of genomic sites to tens of thousands of sites can reduce the processing load and processing time by several orders of magnitude. This reduction can not only speed up cancer detection, but more importantly, it can also free up computer resources for other models and classification tasks (for example, processing more samples), improve the performance of computers running the models, reduce the capital cost of such systems, and improve public health, medicine, diagnosis, treatment, and other aspects by detecting cancer (even) before conventional methods (may detect).

[0072] The large amount of data generated by NGS sequencing also brings with it another problem: how to properly train a machine learning model to identify cancer in this large amount of data. For example, a machine learning model may be trained to identify cancer by comparing feature vectors to genomic data. As listed below, a "feature" in a feature vector can be any genomic site as long as the site has a sufficient number of abnormally methylated genomic locations corresponding to the presence of cancer. When constructing a feature vector for the entire genome, this typically results in tens of thousands of features, and as mentioned above, some of those features may be more indicative of the presence of cancer than others. In this context, it is difficult to select which features and corresponding genomic data to use for training a machine learning model. The machine learning model should be trained and configured so that it can accurately identify the presence of cancer, but the computational cost of the resulting model should not be too high. In other words, the appropriate selection of data and features used to train a machine learning model can improve early cancer detection.

[0073] Overview of the IA Cancer Classification Workflow

[0074] According to one or more embodiments, Figure 1is an exemplary flowchart that describes an overall workflow 100 for classifying cancer in a sample. The workflow 100 is performed by one or more entities (e.g., including medical personnel, sequencing devices, analysis systems, etc.). The purpose of the workflow includes detecting and / or monitoring cancer in an individual. From a healthcare perspective, the workflow 100 can be used to supplement other existing cancer diagnostic tools. The workflow 100 may be used to provide early cancer detection and / or routine cancer monitoring to better inform treatment plans for individuals diagnosed with cancer. The overall workflow 100 may include more than Figure 1 More / less steps than those shown.

[0075] The medical staff performs sample collection 110. An individual who needs cancer classification visits his or her medical staff. The medical staff collects samples for cancer classification. Examples of biological samples include, but are not limited to, tissue biopsies, blood, whole blood, plasma, serum, urine, cerebrospinal fluid, feces, saliva, sweat, tears, pleural fluid, pericardial effusion, or peritoneal fluid from the subject. The sample includes genetic material belonging to the individual, which may be extracted and sequenced for cancer classification. Once the sample is collected, the sample is provided to a sequencing device. In addition to the sample, the medical staff may also collect other information related to the individual, such as biological sex, age, race, smoking status, any previous diagnosis, etc.

[0076] The sequencing device performs sample sequencing 120. The laboratory clinician may perform one or more processing steps on the sample to prepare for sequencing. Once prepared, the clinician loads the sample into the sequencing device. Fig. 7A and Figure 7B , further describing examples of apparatus for sequencing. The sequencing apparatus typically extracts and separates nucleic acid fragments and sequences the fragments to determine the nucleobase sequences corresponding to the fragments. Sequencing may also include amplification of the nucleic acid material. Different sequencing processes include Sanger sequencing, fragment analysis, and next generation sequencing. Sequencing may be whole genome sequencing or targeted sequencing with a target panel. In the context of DNA methylation, bisulfite sequencing (e.g., in Figure 2A and Figure 2B (further described in) can determine the methylation state by bisulfite conversion of unmethylated cytosine at the CpG site. Sample sequencing 120 generates sequences of multiple nucleic acid fragments in the sample. In one or more embodiments, these sequences may include methylation state vectors, where each methylation state vector describes the methylation state of a CpG site on the fragment.

[0077] The analysis system performs pre-analysis processing 130. Figure 7B An example analysis system is described in . Pre-analysis processing 130 may include, but is not limited to, removing duplicate sequence reads, determining metrics related to coverage, determining whether a sample is contaminated, removing contaminated fragments, identifying sequencing errors, and the like.

[0078] The analysis system performs one or more analyses 140. These analyses are statistical analyses or the application of one or more trained models to at least predict the cancer status of the individual from whom the sample was derived. Different genetic features may be evaluated and considered, such as methylation of CpG sites, single nucleotide polymorphisms (SNPs), insertions or deletions (indels), other types of genetic mutations, etc. In the context of methylation, the analysis 140 may include identification of abnormal methylation 142 (e.g., in Figure 5A and Figure 5B ), feature extraction 144 (e.g., in Figure 3A , Figure 3B , Figure 4A , Figure 4B , Figure 5A and Figure 5B ) and applying a cancer classifier 146 to determine a cancer prediction (e.g., Fig. 6A and Figure 6B ). In one or more embodiments of feature extraction, the analysis system may utilize one or more age prediction models to generate one or more age covariate residuals as features for cancer classification. The cancer classifier 146 inputs the extracted features to determine a cancer prediction. The cancer prediction may be a label or a value. The label may indicate a specific cancer state, for example, a binary label may indicate the presence or absence of cancer, and a multi-classification label may indicate one or more cancer types among the multiple cancer types screened. The value may indicate the likelihood of a specific cancer state, for example, the likelihood of having cancer and / or the likelihood of a specific cancer type.

[0079] The analysis system returns the prediction 150 to the healthcare provider. The healthcare provider may develop or adjust a treatment plan based on the cancer prediction. Optimization of treatment is further described in Section IV.C. Treatment.

[0080] IB Methylation Overview

[0081] According to this specification, cfDNA fragments from individuals are processed, for example, unmethylated cytosine is converted into uracil, and then these fragments are sequenced, and these sequence reads are compared with the reference genome to identify the methylation status of specific CpG sites in these DNA fragments. Each CpG site may be methylated or unmethylated. Compared with healthy individuals, the identification of abnormally methylated fragments may provide clues to the cancer status of the subject. As is well known in the art, abnormal DNA methylation (compared with healthy controls) can have different effects, which may contribute to cancer. In the identification of abnormally methylated cfDNA fragments, various challenges have arisen. First, to determine that a DNA fragment is abnormally methylated, it is necessary to compare it with a group of control individuals, and the determination result is convincing. Therefore, if the number of individuals in the control group is small, the statistical variability caused by the small size of the control group will lose credibility. In addition, in a group of control individuals, there may be differences in methylation status, and it is difficult to consider this difference when determining whether the DNA fragment of the subject is abnormally methylated. On the other hand, methylation of cytosine at one CpG site can have a causal effect on the methylation of subsequent CpG sites, and taking this dependency into account is a challenge in itself.

[0082] Methylation usually occurs in deoxyribonucleic acid (DNA), when a hydrogen atom on the pyrimidine ring of the cytosine base is converted into a methyl group, forming 5-methylcytosine. In particular, methylation can occur at a dinucleotide consisting of cytosine and guanine, which is referred to herein as a "CpG site". In other cases, methylation may also occur on a cytosine that is not part of a CpG site or on another nucleotide that is not cytosine; however, these situations are relatively rare. In this disclosure, for the sake of clarity, methylation is discussed with reference to CpG sites. Abnormal DNA methylation can be identified as hypermethylation or hypomethylation, both of which may indicate a cancerous state. Throughout this disclosure, if a DNA fragment contains a number of CpG sites that exceeds a threshold, and more than a threshold percentage of those CpG sites are methylated or unmethylated, then the hypermethylation and hypomethylation of the DNA fragment can be characterized.

[0083] The principles described herein are equally applicable to methylation detection in non-CpG backgrounds, including methylation of non-cytosine. In such embodiments, the laboratory wet assays for detecting methylation may be different from those described herein. Further, the methylation state vectors discussed herein may contain some elements, which are usually sites where methylation has occurred or has not occurred (even if those sites are not specifically CpG sites). After this adjustment, the remaining processes described herein can remain unchanged, and therefore, the inventive concepts described herein are also applicable to those other forms of methylation.

[0084] IC Definition

[0085] The term "free nucleic acid" or "cfNA" refers to nucleic acid fragments circulating in an individual (e.g., in the blood), which are derived from one or more healthy cells and / or one or more unhealthy cells (e.g., cancer cells). The term "free DNA" or "cfDNA" refers to deoxyribonucleic acid fragments circulating in an individual (e.g., in the blood). In addition, cfNA or cfDNA in an individual may also come from other non-human sources.

[0086] The term "genomic nucleic acid", "genomic DNA" or "gDNA" refers to a nucleic acid molecule or a deoxyribonucleic acid molecule obtained from one or more cells. In various embodiments, gDNA can be extracted from healthy cells (e.g., non-tumor cells) or tumor cells (e.g., biopsy samples). In some embodiments, gDNA can be extracted from cells derived from a blood cell lineage (e.g., white blood cells).

[0087] The term "circulating tumor DNA" or "ctDNA" refers to nucleic acid fragments originating from tumor cells or other types of cancer cells that may be released into an individual's body fluids (e.g., blood, sweat, urine, or saliva) as a result of biological processes such as apoptosis or necrosis of dying cells or actively released by viable tumor cells.

[0088] The terms "DNA fragment," "fragment," or "DNA molecule" may generally refer to any deoxyribonucleic acid fragment, i.e., cfDNA, gDNA, ctDNA, etc.

[0089] The term "abnormal fragment", "abnormally methylated fragment" or "fragment with abnormal methylation pattern" refers to a fragment with abnormal methylation at a CpG site. Abnormal methylation of a fragment can be determined using a probabilistic model to identify the degree of unexpectedness of the methylation pattern of the fragment observed in the control group.

[0090] The term "abnormal fragments with extreme methylation" or "UFXM" refers to fragments that are hypomethylated or hypermethylated. Hypomethylated fragments and hypermethylated fragments refer to fragments that contain at least a certain number (e.g., 5) of CpG sites and have more than a certain threshold percentage (e.g., 90%) of these sites in a methylated or unmethylated state, respectively.

[0091] The term "abnormal score" refers to a score for a CpG site based on the number of abnormal fragments (or UFXMs, in some embodiments) overlapping the CpG site in a sample. The abnormal score is used in the context of characterizing a sample for classification.

[0092] As used herein, the term "about" or "approximately" may mean that a particular value determined by a person of ordinary skill in the art is within an acceptable error range, which depends in part on how the value is measured or determined, for example, by limitations of the measurement system. For example, according to the practice in the art, "about" may mean within 1 or more than 1 standard deviation. "About" may mean a range of ±20%, ±10%, ±5%, or ±1% of a given value. The term "about" or "approximately" may mean within the same order of magnitude, within 5 times, or within 2 times of a value. When describing a particular numerical value in the present application and claims, unless otherwise stated, it should be assumed that the term "about" means within an acceptable error range for the particular value. The term "about" may have the meaning commonly understood by a person of ordinary skill in the art. The term "about" may mean ±10%. The term "about" may mean ±5%.

[0093] As used herein, the term "biological sample", "patient sample" or "sample" refers to any sample taken from a subject, which can reflect a biological state associated with the subject and includes free DNA. Examples of biological samples include, but are not limited to, blood, whole blood, plasma, serum, urine, cerebrospinal fluid, feces, saliva, sweat, tears, pleural fluid, pericardial effusion or peritoneal fluid of the subject. A biological sample may include any tissue or material derived from a living or dead subject. A biological sample may be a free sample. A biological sample may contain nucleic acid (e.g., DNA or RNA) or a fragment thereof. The term "nucleic acid" may refer to deoxyribonucleic acid (DNA), ribonucleic acid (RNA) or any hybrid or fragment thereof. The nucleic acid in a sample may be a free nucleic acid. A sample may be a liquid sample or a solid sample (e.g., a cell or tissue sample). The biological sample can be a body fluid, such as blood, plasma, serum, urine, vaginal secretions, fluid in a hydrocele (e.g., a testicular hydrocele), vaginal washings, pleural fluid, ascites, cerebrospinal fluid, saliva, sweat, tears, sputum, bronchoalveolar lavage fluid, nipple discharge, aspirates from different parts of the body (e.g., thyroid, breast), etc. The biological sample can be a fecal sample. In multiple embodiments, most of the DNA in a biological sample in which free DNA has been enriched (e.g., a plasma sample obtained by a centrifugation scheme) can be free (e.g., greater than 50%, 60%, 70%, 80%, 90%, 95% or 99% of the DNA can be free). The biological sample can be processed to physically destroy the tissue or cell structure (e.g., centrifugation and / or cell lysis) to release the intracellular components into a solution, which can further contain enzymes, buffers, salts, detergents, etc., which can be used to prepare samples for analysis.

[0094] As used herein, the terms "control," "control sample," "reference," "reference sample," "normal," and "normal sample" describe a sample taken from a subject that does not suffer from a particular disease or is otherwise healthy. In one example, the methods disclosed herein can be applied to a subject suffering from a tumor, wherein the reference sample is a sample taken from healthy tissue of the subject. The reference sample can be obtained from the subject or from a database. The reference can be, for example, a reference genome, which is used to compare the sequences of nucleic acid fragments obtained from sequencing of the subject's sample. The reference genome can refer to a haploid or diploid genome, to which the sequences of nucleic acid fragments from biological samples and constitutional samples can be compared and compared. An example of a constitutional sample can be the DNA of white blood cells obtained from the subject. For a haploid genome, each locus has only one nucleotide. For a diploid genome, heterozygous loci can be identified; each heterozygous locus has two alleles, either of which can match the locus for comparison.

[0095] As used herein, the term "cancer" or "tumor" refers to an abnormal mass of tissue in which the growth of the mass exceeds and is out of coordination with the growth of normal tissue.

[0096] As used herein, the phrase "healthy" refers to a subject that is in good health. A healthy subject may be shown to be free of any malignant or non-malignant disease. A "healthy individual" may have other diseases or conditions unrelated to the condition being measured that are not generally considered "healthy".

[0097] As used herein, the term "methylation" refers to a modification of deoxyribonucleic acid (DNA) in which a hydrogen atom on the pyrimidine ring of a cytosine base is converted into a methyl group to form 5-methylcytosine. In particular, methylation tends to occur at dinucleotides consisting of cytosine and guanine, which are referred to herein as "CpG sites". In other cases, methylation may also occur on cytosine that is not part of a CpG site or on another nucleotide that is not cytosine; however, these situations are rare. Abnormal cfDNA methylation can be identified as hypermethylation or hypomethylation, both of which may indicate a cancer state. Abnormal DNA methylation (compared to healthy controls) can have different effects, which may contribute to cancer. The principles described herein also apply to methylation detection in CpG and non-CpG contexts, including methylation of non-cytosine. Further, the methylation state vector may contain some elements, which are usually vectors of sites where methylation has occurred or not occurred (even if those sites are not specifically CpG sites).

[0098] As used interchangeably herein, the term "methylation fragment" or "nucleic acid methylation fragment" refers to a string of sequences, and the methylation state of each CpG site in multiple CpG sites of the string of sequences is determined by methylation sequencing of nucleic acids (e.g., nucleic acid molecules and / or nucleic acid fragments). In the methylation fragment, the position and methylation state of each CpG site in the nucleic acid fragment are determined based on the comparison of these sequence reads (e.g., obtained by sequencing these nucleic acids) with the reference genome. The nucleic acid methylation fragment contains the methylation state of each CpG site in multiple CpG sites (e.g., a methylation state vector), which specifies the position of the nucleic acid fragment in the reference genome (e.g., as specified by the position of the first CpG site in the nucleic acid fragment or another similar index using a CpG index) and the number of CpG sites in the nucleic acid fragment. Based on the methylation sequencing of nucleic acid molecules, the CpG index can be used to compare the sequence reads with the reference genome. As used herein, the term "CpG index" refers to a list of each CpG site in a plurality of CpG sites of a reference genome (e.g., a human reference genome) (e.g., CpG 1, CpG 2, CpG 3, etc.), which list may be in electronic format. The CpG index further comprises the corresponding genomic position of each corresponding CpG site in the CpG index in the corresponding reference genome. Therefore, each CpG site in each corresponding nucleic acid methylation fragment can be indexed to a specific position in the corresponding reference genome determined by using the CpG index.

[0099] As used herein, the term "true positive" (TP) refers to a subject suffering from a certain disease. "True positive" may refer to a subject suffering from a tumor, cancer, precancerous lesions (e.g., precancerous lesions), localized or metastatic cancers, or non-malignant diseases. "True positive" may also refer to a subject suffering from a certain disease and identified as suffering from the disease by a determination or method disclosed herein. As used herein, the term "true negative" (TN) refers to a subject not suffering from a certain disease or a certain detectable disease. "True negative" may refer to a subject not suffering from a certain disease or a certain detectable disease (e.g., tumor, cancer, precancerous lesions (e.g., precancerous lesions), localized or metastatic cancers, non-malignant diseases) or a healthy subject in other respects. "True negative" may refer to a subject not suffering from a certain disease or a certain detectable disease, or a subject not suffering from the disease by a determination or method disclosed herein.

[0100] As used herein, the term "reference genome" refers to any known, sequenced or characterized specific genome, whether partial or complete, belonging to any organism or virus, which can be used to compare identified sequences from a subject. The online genome browser hosted by the National Center for Biotechnology Information ("NCBI") or the University of California, Santa Cruz (UCSC) provides exemplary reference genomes for human subjects and many other organisms. "Genome" refers to the complete genetic information of an organism or virus expressed in the form of a nucleic acid sequence. As used herein, a reference sequence or reference genome is typically an assembled or partially assembled genome sequence from one or more individuals. In some embodiments, a reference genome is an assembled or partially assembled genome sequence from one or more human individuals. A reference genome can be considered a representative example of a gene set of a species. In some embodiments, a reference genome comprises a sequence assigned to a chromosome. Exemplary human reference genomes include, but are not limited to, NCBI build 34 (UCSC equivalent: hg16), NCBI build 35 (UCSC equivalent: hg17), NCBI build 36.1 (UCSC equivalent: hg18), GRCh37 (UCSC equivalent: hg19), and GRCh38 (UCSC equivalent: hg38).

[0101] As used herein, the term "sequence read" or "read" refers to a nucleotide sequence generated by any sequencing process described herein or known in the art. Reads can be generated from one end of a nucleic acid fragment ("single-end read"), and sometimes from both ends of a nucleic acid (e.g., paired-end reads, double-end reads). In some embodiments, sequence reads (e.g., single-end or paired-end reads) can be generated from one or both strands of a targeted nucleic acid fragment. The length of the sequence read is generally associated with a specific sequencing technology. For example, the sequence reads provided by high-throughput methods vary in size from tens to hundreds of base pairs (bp). In some embodiments, the mean, median, or average length of the sequence reads is about 15 bp to 900 bp in length (e.g., about 20 bp, about 25 bp, about 30 bp, about 35 bp, about 40 bp, about 45 bp, about 50 bp, about 55 bp, about 60 bp, about 65 bp, about 70 bp, about 75 bp, about 80 bp, about 85 bp, about 90 bp, about 95 bp, about 100 bp, about 110 bp, about 120 bp, about 130 bp, about 140 bp, about 150 bp, about 200 bp, about 450 bp, about 300 bp, about 350 bp, about 400 bp, about 450 bp, or about 500 bp). In some embodiments, the mean length, median length or average length of these sequence reads is about 1000bp, 2000bp, 5000bp, 10,000bp or 50,000bp or longer. For example, nanopore sequencing can provide sequence reads ranging in size from tens to hundreds to thousands of base pairs. Illumina parallel sequencing can provide sequence reads with less variation in size, for example, most of these sequence reads are less than 200bp. A sequence read (or sequencing read) can refer to sequence information corresponding to a nucleic acid molecule (e.g., a string of nucleotides). For example, a sequence read can correspond to a string of nucleotides (e.g., about 20 to about 150) from a portion of a nucleic acid fragment, a string of nucleotides that can correspond to one or both ends of a nucleic acid fragment, or nucleotides that can correspond to the entire nucleic acid fragment. Sequence reads can be obtained in a variety of ways, for example, using sequencing technology or using probes, for example, in a hybridization array or capture probe, or amplification technology, for example, polymerase chain reaction (PCR) or linear amplification or isothermal amplification using a single primer.

[0102] As used herein, the term "sequencing" and the like generally refers to any and all biochemical processes that can be used to determine the order of biological macromolecules (such as nucleic acids or proteins). For example, sequencing data may include all or part of the nucleotide bases in a nucleic acid molecule (such as a DNA fragment).

[0103] As used herein, the term "sequencing depth" is used interchangeably with the term "coverage" and refers to the number of times that the consensus sequence reads corresponding to the unique nucleic acid target molecules aligned to a certain locus are aligned to the locus; for example, the sequencing depth is equal to the number of unique nucleic acid target molecules covering the locus. The locus can be as small as one nucleotide, or as large as a chromosome arm, or as large as the entire genome. Sequencing depth can be expressed as "Yx", such as 50x, 100x, etc., where "Y" refers to the number of times the locus is covered by the sequence corresponding to the target nucleic acid; for example, the number of times the information of independent sequence coverage of the specific locus is obtained. In some embodiments, the sequencing depth corresponds to the number of genomes that have been sequenced. Sequencing depth can also be applied to multiple loci or the entire genome, in which case Y can refer to the mean or average number of times a locus, a haploid genome, or the entire genome is sequenced, respectively. When referring to the average depth, the actual depth of different loci included in the data set can cover a range of values. Ultra-deep sequencing can refer to a sequencing depth of at least 100x at a certain locus.

[0104] As used herein, the term "sensitivity" or "true positive rate" (TPR) refers to the number of true positives divided by the sum of the number of true positives and false negatives. Sensitivity can characterize the ability of an assay or method to correctly identify the proportion of a population that actually has a disease. For example, sensitivity can characterize the ability of a method to correctly identify the number of subjects in a population that have cancer. In another example, sensitivity can characterize the ability of a method to correctly identify one or more markers that indicate cancer.

[0105] As used herein, the term "specificity" or "true negative rate" (TNR) refers to the number of true negatives divided by the sum of the number of true negatives and false positives. Specificity can characterize the ability of an assay or method to correctly identify the proportion of a population that is truly free of a disease. For example, specificity can characterize the ability of a method to correctly identify the number of subjects within a population that do not have cancer. In another example, specificity characterizes the ability of a method to correctly identify one or more markers that indicate cancer.

[0106] As used herein, the term "subject" refers to any living or inanimate organism, including but not limited to humans (e.g., male humans, female humans, fetuses, pregnant women, children, etc.), non-human animals, plants, bacteria, fungi, or protists. Any human or non-human animal can be used as a subject, including but not limited to mammals, reptiles, birds, amphibians, fish, ungulates, ruminants, bovines (e.g., cattle), equines (e.g., horses), sheep and sheep (e.g., sheep, goats), swine (e.g., pigs), camelids (e.g., camels, llamas, alpacas), monkeys, apes (e.g., gorillas, chimpanzees), ursids (e.g., bears), poultry, dogs, cats, mice, rats, fish, dolphins, whales, and sharks. In some embodiments, the subject is a male or female (e.g., an adult male, an adult female, or a child) at any stage. The subject from which the sample is collected or treated using any method or composition described herein can be of any age and can be an adult, an infant, or a child.

[0107] As used herein, the term "tissue" can correspond to a group of cells that are gathered together as a functional unit. More than one cell type can be found in a single tissue. Different types of tissues may be composed of different types of cells (e.g., hepatocytes, alveolar cells, or blood cells), but may also correspond to tissues from different organisms (maternal and fetal) or to healthy cells and tumor cells. The term "tissue" can generally refer to any group of cells found in the human body (e.g., heart tissue, lung tissue, kidney tissue, nasopharyngeal tissue, oropharyngeal tissue). In some respects, the term "tissue" or "tissue type" can be used to refer to the source tissue of free nucleic acids. In one example, viral nucleic acid fragments can be derived from blood tissue. In another example, viral nucleic acid fragments can be derived from tumor tissue.

[0108] As used herein, the term "genomic" refers to the characteristics of an organism's genome. Examples of genomic characteristics include, but are not limited to, those characteristics associated with the primary nucleic acid sequence of all or part of the genome (e.g., the presence or absence of nucleotide polymorphisms, insertions and deletions, sequence rearrangements, mutation frequencies, etc.), the number of copies of one or more specific nucleotide sequences within the genome (e.g., copy number, allele frequency fraction, ploidy of a single chromosome or the entire genome, etc.), the epigenetic state of all or part of the genome (e.g., covalent nucleic acid modifications such as methylation, histone modifications, nucleosome selection, etc.), the expression profile of the organism's genome (e.g., gene expression levels, isoform expression levels, gene expression ratios, etc.).

[0109] The terminology used herein is intended to describe a particular situation only and is not intended to be limiting. As used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. In addition, to the extent that the terms "including", "includes", "having", "has", "with" or variations thereof are used in the detailed description and / or claims, such terms are intended to mean inclusion in a manner similar to the term "comprising".

[0110] ID Sample Analysis System

[0111] According to one or more embodiments, Fig. 7A 7 is an exemplary flow chart of devices that are used to sequence nucleic acid samples. This illustrative flow chart includes devices such as a sequencer 720 and an analysis system 700. The sequencer 720 and the analysis system 700 can work together to perform Figure 3A Process 300, Figure 4A Process 400, Figure 4B The process 420 and one or more steps of other processes described herein.

[0112] In various embodiments, the sequencer 720 receives the enriched nucleic acid sample 710. Fig. 7A As shown, the sequencer 720 may include a graphical user interface 725 that enables a user to interact with specific tasks (e.g., start sequencing or terminate sequencing) and one or more loading stations 730 for loading sequencing boxes including the enriched fragment samples and / or for loading buffers necessary to perform sequencing assays. Thus, once the user of the sequencer 720 has provided the necessary reagents and sequencing boxes to the loading stations 730 of the sequencer 720, the user may start sequencing by interacting with the graphical user interface 725 of the sequencer 720. Once started, the sequencer 720 performs sequencing and outputs sequence reads of the enriched fragments in the nucleic acid sample 710.

[0113] In some embodiments, the sequencer 720 is communicatively coupled to the analysis system 700. The analysis system 700 includes a certain number of computing devices for processing sequence reads to meet various application requirements, such as evaluating the methylation status of one or more CpG sites, variant identification, or quality control. The sequencer 720 can provide the sequence reads to the analysis system 700 in a BAM file format. The analysis system 700 can be communicatively coupled to the sequencer 720 by wireless, wired, or a combination of wireless and wired communication technologies. In general, the analysis system 700 is configured with a processor and a non-transitory computer-readable storage medium storing computer instructions, which, when executed by the processor, causes the processor to process the sequence reads or perform one or more steps of any method or process disclosed herein.

[0114] In some embodiments, the sequence reads can be aligned with the reference genome using methods known in the art to determine alignment position information. The alignment position can generally describe the starting position and the ending position of a region in the reference genome, and the region corresponds to the starting nucleotide base and the ending nucleotide base of a given sequence read. For methylation sequencing, the concept of alignment position information can be generalized to indicate the first CpG site and the last CpG site included in the sequence read according to the alignment with the reference genome. The alignment position information can further indicate the methylation state and position of all CpG sites in a given sequence read. The region in the reference genome may be associated with a gene or a fragment of a gene; therefore, the analysis system 700 can mark the sequence read with one or more genes aligned with the sequence read. In one embodiment, the fragment length (or size) is determined by the starting position and the ending position.

[0115] In a number of embodiments, for example, when a paired-end sequencing process is used, the sequence reads consist of a pair of reads, represented as R_1 and R_2. For example, the first read R_1 can be sequenced from one end of a double-stranded DNA (dsDNA) molecule, while the second read R_2 is sequenced from the other end of the double-stranded DNA (dsDNA). Therefore, the nucleotide base pairs of the first read R_1 and the second read R_2 can be aligned in a consistent manner (e.g., in the opposite direction) with the nucleotide bases of the reference genome. The alignment position information derived from the reads R_1 and R_2 may include a starting position in the reference genome, the position corresponding to one end of the first read (e.g., R_1), and an end position in the reference genome, the position corresponding to one end of the second read (e.g., R_2). In other words, the starting position and the end position in the reference genome can represent the positions that the nucleic acid fragments may correspond to in the reference genome. Output files in SAM (sequence alignment map) format or BAM (binary) format can be generated and output for further analysis.

[0116] Reference now Figure 7B According to one embodiment, Figure 7B 7 is a block diagram of an analysis system 700 for processing DNA samples. The analysis system uses one or more computing devices for analyzing DNA samples. The analysis system 700 includes a sequence processor 740, a sequence database 745, a model database 755, a model 750, a parameter database 765, and a scoring engine 760. In some embodiments, the analysis system 700 performs part or all of the processes described throughout the present disclosure.

[0117] The sequence processor 740 generates a methylation state vector for each fragment from the sample. For each CpG site on the fragment, the sequence processor 740 generates a methylation state vector for each fragment, which indicates the position of the fragment in the reference genome, the number of CpG sites in the fragment, and the methylation state of each CpG site in the fragment (methylated, unmethylated, or uncertain). This process is performed by Figure 2A The sequence processor 740 may store the methylation state vector of each fragment in a sequence database 745. The data in the sequence database 745 may be organized such that the methylation state vectors from the same sample are associated.

[0118] Further, multiple different models 750 can be stored in the model database 755 or retrieved for use with a test sample. In one example, the model is a trained cancer classifier that uses feature vectors derived from abnormal segments to determine a cancer prediction for a test sample. The training and use of a cancer classifier will be further discussed in conjunction with Section III: Cancer Classifiers for Determining Cancer. The analysis system 700 can train one or more models 750 and store various trained parameters in a parameter database 765. The analysis system 700 stores the models 750 in the model database 755 along with the associated functions.

[0119] During inference, the scoring engine 760 uses one or more models 750 to return outputs. The scoring engine 760 accesses the models 750 in the model database 755 and the trained parameters from the parameter database 765. For each model, the scoring engine receives inputs suitable for the model and calculates outputs based on the received inputs, parameters, and functions in each model that relate the inputs and outputs. In some use cases, the scoring engine 760 further calculates metrics related to the confidence of the outputs calculated by the model. In other use cases, the scoring engine 760 calculates other intermediate values ​​for use in the model.

[0120] II. Sample Sequencing and Processing

[0121] II.A. Generating methylation state vectors for DNA fragments

[0122] According to one or more embodiments, Figure 2A is an exemplary flowchart that describes a process 200 for sequencing a fragment of cfDNA to obtain a methylation state vector. To analyze DNA methylation, the analysis system first obtains 210 a sample containing multiple cfDNA molecules from an individual. In other embodiments, the process 200 can be applied to sequence other types of DNA molecules. The process 200 is Figure 1 An embodiment of sample sequencing 120.

[0123] The analysis system can separate 210 each cfDNA molecule from the sample. The cfDNA molecules can be processed 220 to convert unmethylated cytosine to uracil. In one embodiment, the method uses bisulfite treatment of DNA, which can convert unmethylated cytosine to uracil without converting methylated cytosine. For example, commercially available kits can be used to perform bisulfite conversion, such as EZ DNA Methylation TM -Gold, EZ DNA Methylation TM -Direct or EZ DNA Methylation TM -Lightning kit (from Zymo Research, Inc. (Irvine, CA)). In another embodiment, the conversion of unmethylated cytosine to uracil is accomplished by an enzymatic reaction. For example, the conversion can be accomplished using a commercially available kit, such as APOBEC-Seq (NEBiolabs, Ipswich, MA).

[0124] From the converted cfDNA molecules, a sequencing library can be prepared 230 . During library preparation, a unique molecular identifier (UMI) can be added to a nucleic acid molecule (e.g., a DNA molecule) by adapter ligation. The UMI can be a short nucleic acid sequence (e.g., 4-10 base pairs) that is added to both ends of a DNA fragment (e.g., a DNA molecule cleaved by physical shearing, enzymatic digestion, and / or chemical cleavage) during adapter ligation. The UMI can be a degenerate base pair that serves as a unique tag for identifying sequence reads originating from a specific DNA fragment. During PCR amplification after adapter ligation, the UMI is replicated along with the ligated DNA fragment. This provides a method for identifying sequence reads from the same original fragment in downstream analysis.

[0125] Optionally, multiple hybridization probes can be used to enrich cfDNA molecules or genomic regions in the sequencing library 235, which can provide information about the cancer status. These hybridization probes are short oligonucleotides that can hybridize with specifically designated cfDNA molecules or targeted regions and enrich those fragments or regions for subsequent sequencing and analysis. Hybridization probes can be used for targeted, high-depth analysis of a set of designated CpG sites of interest to researchers. Hybridization probes can be tiled on one or more target sequences at a coverage of 1X, 2X, 3X, 4X, 5X, 6X, 7X, 8X, 9X, 10X, or greater than 10X. For example, a hybridization probe tiled at 2X coverage contains overlapping probes so that each portion of the target sequence can hybridize with 2 independent probes. Hybridization probes can be tiled on one or more target sequences at a coverage of less than 1X.

[0126] In one embodiment, hybridization probes are designed to enrich DNA molecules that have been treated (e.g., using bisulfite) to convert unmethylated cytosine into uracil. During enrichment, hybridization probes (also referred to herein as "probes") can be used to target and capture nucleic acid fragments that can provide information on the presence or absence of cancer (or disease), cancer status, or cancer classification (e.g., cancer category or tissue of origin). These probes can be designed to anneal (or hybridize) to a target (complementary) strand of DNA. The target strand can be a "positive" strand (e.g., a strand that is transcribed into mRNA and subsequently translated into a protein) or a complementary "negative" strand. The length range of the probe can be tens, hundreds, or thousands of base pairs. The probe can be designed based on a group of methylation site groups. These probes can be designed based on a group of targeted genes to analyze specific mutations or target regions suspected to correspond to certain cancers or other types of diseases in a genome (e.g., humans or other organisms). In addition, these probes can cover overlapping portions of the target region.

[0127] Once prepared, the sequencing library or a portion thereof may be sequenced 240 to obtain a plurality of sequence reads. These sequence reads may be in a computer-readable digital format for processing and interpretation by computer software. These sequence reads may be aligned with a reference genome to determine alignment position information. The alignment position information may indicate a starting position and an ending position of a region in the reference genome, the region corresponding to the starting nucleotide base and the ending nucleotide base of a given sequence read. The alignment position information may also include a sequence read length, which may be determined by the starting position and the ending position. A region in the reference genome may be associated with a gene or a fragment of a gene. A sequence read may consist of a pair of reads, represented as R1 and R2. For example, a first read R1 may be sequenced from one end of a nucleic acid fragment, while a second read R2 may be sequenced from the other end of the nucleic acid fragment. Therefore, the nucleotide base pairs of the first read R1 and the second read R2 may be aligned in a manner that is consistent with the nucleotide bases of the reference genome (e.g., in opposite directions). The alignment position information derived from the read pair R1 and R2 may include a starting position in the reference genome corresponding to one end of the first read (e.g., R1), and an ending position in the reference genome corresponding to one end of the second read (e.g., R2). In other words, the starting position and the ending position in the reference genome may represent the positions to which the nucleic acid fragments may correspond in the reference genome. An output file in SAM (sequence alignment map) format or BAM (binary) format may be generated and output for further analysis, such as methylation status determination.

[0128] From these sequence reads, the analysis system determines 250 the position and methylation state of each CpG site based on comparison with the reference genome. The analysis system generates 260 a methylation state vector for each fragment, which specifies the position of the fragment in the reference genome (e.g., as specified by the position of the first CpG site in each fragment or another similar indicator), the number of CpG sites in the fragment, and the methylation state of each CpG site in the fragment (whether methylated (e.g., represented as M), unmethylated (e.g., represented as U), or uncertain (e.g., represented as I)). The observed state can be a methylated and unmethylated state; while the unobserved state is uncertain. Uncertain methylation states may arise from sequencing errors and / or differences in methylation states between complementary strands of the DNA fragments. The methylation state vector may be stored in a transient or permanent computer memory for subsequent use and processing. Further, the analysis system may remove duplicate reads or duplicate methylation state vectors from a single sample. The analysis system may determine that a particular fragment with one or more CpG sites has a number or percentage of uncertain methylation states above a threshold and may exclude such fragments, or selectively include such fragments but construct a model to account for such uncertain methylation states.

[0129] According to one or more embodiments, Figure 2B is an exemplary diagram, which illustrates Figure 2A 200 of sequencing cfDNA molecules to obtain methylation state vectors. For example, the analysis system receives a cfDNA molecule 212, which, in this example, contains three CpG sites. As shown, the first and third CpG sites of the cfDNA molecule 212 are methylated 214. During the processing step 220, the cfDNA molecule 212 is converted to generate a converted cfDNA molecule 222. During the processing 220, the cytosine of the unmethylated second CpG site is converted to uracil. However, the first and third CpG sites are not converted.

[0130] After conversion, a sequencing library 230 is prepared and sequenced 240 to generate a sequence read 242. The analysis system compares the sequence read 242 with the reference genome 244 250. The reference genome 244 provides a context that can explain where the cfDNA fragment is derived from in the human genome. In this simplified example, the analysis system compares the sequence read 242 250 so that the three CpG sites correspond to CpG sites 23, 24, and 25 (these are arbitrary reference identifiers used for ease of description). Thus, the analysis system can generate information about the methylation status of all CpG sites on the cfDNA molecule 212 and the positions of these CpG sites mapped in the human genome. As shown in the figure, the methylated CpG sites on the sequence read 242 are read as cytosine. In this example, cytosine only appears in the first and third CpG sites of the sequence read 242, from which it can be inferred that the first and third CpG sites in the original cfDNA molecule are methylated. The second CpG site can be read as thymine (U is converted to T during the sequencing process), from which it can be inferred that the second CpG site in the original cfDNA molecule is unmethylated. With these two pieces of information, the methylation status and position, the analysis system generates 260 a methylation state vector 252 for the cfDNA fragment 212. In this example, the resulting methylation state vector 252 is <M 23 , U 24 , M 25 >, where M corresponds to methylated CpG sites, U corresponds to unmethylated CpG sites, and the subscript numbers correspond to the position of each CpG site in the reference genome.

[0131] One or more alternative sequencing methods can be used to obtain sequence reads from nucleic acids in biological samples. One or more sequencing methods can include any sequencing format that can be used to obtain a certain number of sequence reads measured from nucleic acids (e.g., free nucleic acids), including but not limited to high-throughput sequencing systems, such as Roche's 454 platform, Applied Biosystems' SOLID platform, Helicos' True Single Molecule DNA sequencing technology, Affymetrix's hybridization sequencing platform, Pacific Biosciences' single molecule real-time (SMRT) technology, 454 Life Sciences, Illumina / Solexa and Helicos Biosciences' synthetic sequencing platforms, Applied Biosystems' ligation sequencing platforms. Life technologies' ION TORRENT technology and nanopore sequencing can also be used to obtain sequence reads from nucleic acids (e.g., free nucleic acids) in biological samples. Synthesis sequencing and reversible terminator-based sequencing (e.g., Illumina Genome Analyzer; Genome Analyzer II; HISEQ 2000; HISEQ 4500 (Illumina, San Diego Calif.)) can be used to obtain sequence reads from cell-free nucleic acids obtained from biological samples of training subjects to form a genotype data set. Millions of cell-free nucleic acid (e.g., DNA) fragments can be sequenced in parallel. In one example of such sequencing technology, a flow cell is used, which includes an optically clear glass slide having eight independent lanes on the surface of the slide to which oligonucleotide anchors (e.g., adapter primers) are bound. The cell-free nucleic acid sample can include a signal or label for easy detection. Obtaining sequence reads from free nucleic acids obtained from biological samples can include obtaining quantitative information of the signal or tag through a variety of techniques, such as flow cytometry, quantitative polymerase chain reaction (qPCR), gel electrophoresis, gene chip analysis, microarray, mass spectrometry, cell fluorescence analysis, fluorescence microscopy, confocal laser scanning microscopy, laser scanning cytometry, affinity chromatography, manual batch separation mode, electric field suspension, sequencing, and combinations thereof.

[0132] One or more sequencing methods may include whole genome sequencing analysis. Whole genome sequencing analysis may include a physical analysis that generates sequence reads of the whole genome or most of the whole genome, which can be used to determine large-scale variations such as copy number variations or copy number aberrations. This physical analysis may use whole genome sequencing technology or full exome sequencing technology. Whole genome sequencing analysis may have an average sequencing depth of at least 1x, 2x, 3x, 4x, 5x, 6x, 7x, 8x, 9x, 10x, at least 20x, at least 30x or at least 40x on the genome of the test subject. In certain embodiments, the sequencing depth is about 30,000x. One or more sequencing methods may include targeted group sequencing analysis. For the targeted genome, the targeted group sequencing analysis may have an average sequencing depth of at least 50,000x, at least 55,000x, at least 60,000x or at least 70,000x. The targeted genome may include 450 to 500 genes. The targeted gene set can comprise a range of 500±5 genes, a range of 500±10 genes, or a range of 500±25 genes.

[0133] One or more sequencing methods may include paired end sequencing. One or more sequencing methods may generate multiple sequence reads. These multiple sequence reads may have an average length range of 10 to 700, 50 to 400, or 100 to 300. One or more sequencing methods may include methylation sequencing analysis. Methylation sequencing may be i) whole genome methylation sequencing or ii) targeted DNA methylation sequencing using multiple nucleic acid probes. For example, methylation sequencing is whole genome bisulfite sequencing (e.g., WGBS). Methylation sequencing may be targeted DNA methylation sequencing using multiple nucleic acid probes, and the sequencing targets the most informative region in the methylation group, may be a unique methylation database, may be a previous whole genome and targeted sequencing prototype analysis.

[0134] Methylation sequencing can detect one or more 5-methylcytosine (5mC) and / or 5-hydroxymethylcytosine (5hmC) in the corresponding nucleic acid methylation fragment. Methylation sequencing can include converting one or more unmethylated cytosines, or one or more methylated cytosines, in the corresponding nucleic acid methylation fragment into corresponding one or more uracils. During methylation sequencing, one or more uracils can be detected as one or more corresponding thymines. The conversion of one or more unmethylated cytosines or one or more methylated cytosines can include chemical conversion, enzymatic conversion, or a combination thereof.

[0135] For example, bisulfite conversion involves converting cytosine to uracil without converting methylated cytosine (e.g., 5-methylcytosine or 5-mC). In some DNA, approximately 95% of the cytosine in the DNA may be unmethylated, and the resulting DNA fragments may include many uracils represented by thymine. The enzymatic conversion process can be used to treat nucleic acids before sequencing, and the treatment can be performed in a variety of ways. An example of non-bisulfite conversion includes a bisulfite-free sequencing method with base-level resolution, TET-assisted pyridine borane sequencing (TAPS), which is used to non-destructively and directly detect 5-methylcytosine and 5-hydroxymethylcytosine without affecting unmodified cytosine. The methylation state of a CpG site in a plurality of corresponding CpG sites in a corresponding nucleic acid methylation fragment, if it is determined to be methylated by methylation sequencing, then the CpG site is in a methylated state; if it is determined to be unmethylated by methylation sequencing, then the CpG site is in an unmethylated state.

[0136] Methylation sequencing analysis (e.g., WGBS and / or targeted methylation sequencing) can have an average sequencing depth including, but not limited to, about 1,000x, 2,000x, 3,000x, 5,000x, 10,000x, 15,000x, 20,000x, or 30,000x. Methylation sequencing can have a sequencing depth greater than 30,000x, for example, at least 40,000x or 50,000x. Whole genome bisulfite sequencing methods can have an average sequencing depth of 20x to 50x, while targeted methylation sequencing methods have an average effective depth of 100x to 1000x, where the effective depth can be the equivalent coverage required when whole genome bisulfite sequencing is used to obtain the same number of sequence reads obtained by targeted methylation sequencing.

[0137] For more detailed information on methylation sequencing (e.g., WGBS and / or targeted methylation sequencing), see, for example, U.S. Patent Application No. 16 / 352,602, filed on March 13, 2019, entitled “Methylation Fragment Anomaly Detection,” and U.S. Patent Application No. 16 / 719,902, filed on December 18, 2019, entitled “Systems and Methods for Estimating Cell Source Fractions Using Methylation Information,” each of which is hereby incorporated by reference. Other methods for methylation sequencing, including those disclosed herein and / or any modifications, substitutions, or combinations thereof, can be used to obtain methylation patterns of fragments. Methylation sequencing can be used to identify one or more methylation state vectors, for example, as described in U.S. patent application Ser. No. 16 / 352,602, filed on March 13, 2019, entitled “Anomalous Fragment Detection and Classification,” or according to any of the techniques disclosed in U.S. patent application Ser. No. 15 / 931,022, filed on May 13, 2020, entitled “Model-Based Featurization and Classification,” each of which is hereby incorporated by reference.

[0138] The methylation sequencing of nucleic acid and one or more methylation state vectors obtained therefrom can be used to obtain multiple nucleic acid methylation fragments. Each group of corresponding multiple nucleic acid methylation fragments (for example, for each corresponding genotype data set) can include more than 100 nucleic acid methylation fragments. In each group of corresponding multiple nucleic acid methylation fragments, the average number of nucleic acid methylation fragments can include 1000 or more nucleic acid methylation fragments, 5000 or more nucleic acid methylation fragments, 10,000 or more nucleic acid methylation fragments, 20,000 or more nucleic acid methylation fragments or 30,000 or more nucleic acid methylation fragments. In each group of corresponding multiple nucleic acid methylation fragments, the average number of nucleic acid methylation fragments can be 10,000 nucleic acid methylation fragments-50,000 nucleic acid methylation fragments. The corresponding multiple nucleic acid methylation fragments may include 1000 or more, 10,000 or more, 100,000 or more, 1 million or more, 10 million or more, 100 million or more, 100 million or more, 500 million or more, 1 billion or more, 2 billion or more, 3 billion or more, 4 billion or more, 5 billion or more, 6 billion or more, 7 billion or more, 8 billion or more, 9 billion or more, or 10 billion or more nucleic acid methylation fragments. The average length of the corresponding multiple nucleic acid methylation fragments may be 140 to 480 nucleotides.

[0139] For more detailed information on nucleic acid sequencing methods and methylation sequencing data, please refer to U.S. patent application Ser. No. 17 / 191,914, filed on March 4, 2021, entitled “Systems and Methods for Cancer Condition Determination Using Autoencoders,” which is hereby incorporated by reference in its entirety.

[0140] III. Cancer Classifier for Identifying Cancer

[0141] Cancer classification may involve extracting genetic features and applying one or more models to these extracted features to determine a cancer prediction. The extracted features may include a feature vector generated for the test sample. Cancer classification for the test sample may involve determining a cancer prediction based on the feature vector. The cancer prediction may include a label and / or a value. The label may be binary, indicating the presence or absence of cancer in the test subject, and / or multi-classification, indicating one or more specific cancer types among multiple cancer types screened. In particular, a cancer classifier may be a machine learning model that includes multiple classification parameters and a function that represents the relationship between a feature vector as input and a cancer prediction as output. The feature vector is input into the function together with the classification parameters to generate a cancer prediction. In one or more embodiments, an age prediction model is used to predict the age of an individual associated with the test sample based on a methylation feature. The residual between the predicted age and the reported age of the test subject may be used as a feature in the cancer classifier. In one or more embodiments, the feature vector input to the cancer classifier is determined based on a set of abnormal fragments (also referred to as "abnormal methylation" or "extremely methylated abnormal fragments" (UFXM)) determined from the test sample. These abnormal fragments may be determined by Figure 5B The abnormal fragments are determined by the process 520 in the above method, or more specifically, the abnormal fragments are the hypermethylated and hypomethylated fragments determined by step 570 of the process 520, or the abnormal fragments are abnormal fragments determined according to some other process. Before deploying the cancer classifier, the analysis system can train the cancer classifier.

[0142] Cancer classification may involve extracting genetic features and applying one or more models to these extracted features to determine a cancer prediction. The extracted features may include a feature vector for a test sample, and the cancer prediction is determined based on the input feature vector. The cancer prediction may include a label and / or a value. The label may be binary, indicating the presence or absence of cancer in the test subject, and / or multi-classification, indicating one or more specific cancer types among multiple cancer types screened. In particular, a cancer classifier may be a machine learning model that includes multiple classification parameters and a function that represents the relationship between the feature vector as input and the cancer prediction as output. The feature vector is input into the function along with the classification parameters to generate a cancer prediction. In one or more embodiments, an age prediction model is used to predict the age of a test sample based on a methylation feature. The residual between the predicted age and the reported age of the test subject may be used as a feature in the cancer classifier. In one or more embodiments, the feature vector input to the cancer classifier is determined based on a set of abnormal fragments (also referred to as "abnormal methylation" or "extremely methylated abnormal fragments" (UFXM)) determined from the test sample. These abnormal fragments may be determined by Figure 5B The abnormal fragments are determined by the process 520 in the above method, or more specifically, the abnormal fragments are the hypermethylated and hypomethylated fragments determined by step 570 of the process 520, or the abnormal fragments are abnormal fragments determined according to some other process. Before deploying the cancer classifier, the analysis system can train the cancer classifier.

[0143] III.A. Age prediction model

[0144] The age prediction model can predict the age of the sample based on the methylation features extracted from the methylation pattern of the sample. The age prediction model may evaluate the methylation features on multiple age-informative or age-suggestive genomic regions. These genomic regions may be single CpG sites or regions covering multiple CpG sites. The methylation features can be derived from the methylation patterns of sequence reads in the sample. The number of age-suggestive regions and the number of age-suggestive features generated therefrom may be 1, 5, 10, 25, 50, 100, 1,000, 10,000, 100,000 or more genomic regions.

[0145] According to one or more embodiments, Figure 3AMethylation features are illustrated, and these methylation features can be derived from a single CpG site 305 as a genomic region. For example, there are six fragments 310 overlapping with a single CpG site 305. At each CpG site (represented as a diamond) on a fragment, the fragment has a methylation state. Methylation state may include methylation (shown as solid), unmethylation (shown as hollow) and variation (shown as diagonal slashes). Variant methylation may include an uncertain state caused by mutation or sequencing error. A methylation feature is the methylation density at CpG site 305. In this example, four (fragment 1 310A, fragment 2 310B, fragment 5 310E and fragment 6 310F) in six fragments have a methylation state at CpG site 305. Methylation density is exactly 4 / 6, 0.66 or 66%. Another methylation feature is to calculate the percentage of the highly methylated fragments overlapping with CpG site 305. Highly methylated fragments may have a methylation percentage above the threshold at the CpG sites they overlap. Example threshold percentages include 75%, 80%, 85%, 90%, 95%, etc. In this example, using a threshold percentage of 80%, three of the six fragments (fragment 1 310A, fragment 2 310B, and fragment 3 310C) are highly methylated, that is, at least four-fifths of the CpG sites are methylated. The percentage of overlapping highly methylated fragments is 3 / 6, 0.50, or 50%. Another methylation feature is to calculate the percentage of highly unmethylated fragments that overlap with CpG site 305. Example threshold percentages include 75%, 80%, 85%, 90%, 95%, etc. In this example, using 80% as the threshold percentage, one of the six fragments (fragment 310B) is highly unmethylated, that is, at least four-fifths of the CpG sites are unmethylated. The percentage of overlapping highly unmethylated fragments is 1 / 6, 0.16 or 16%. Fragment 5 310E and fragment 6 310F have mixed methylation, that is, neither highly methylated nor highly unmethylated. The purpose of using the methylation features of high methylation or high unmethylation is to identify more important fragments that overlap with CpG site 305. Taking fragment 3 310C as an example, fragment 3 310C is unmethylated at CpG site 305, but as a highly methylated fragment, fragment 3 310C will be counted in the number of highly methylated overlapping fragments. Other methylation features may be derived based on the above methylation features. For example, another methylation feature may involve a comparison between the number of highly methylated overlapping fragments and the number of highly unmethylated overlapping fragments, for example, the ratio of these two numbers.

[0146] According to one or more embodiments, Figure 3BMethylation signatures are illustrated that may be derived from a plurality of CpG sites as a genomic region 315. The CpG sites 317 include CpG sites 1, 2, 3, 4, and 5. Figure 3A In the figure, solid diamonds represent methylation, hollow diamonds represent unmethylation, and diagonal diamonds represent mutations. Figure 3A Similarly, methylation features include, but are not limited to, the methylation density of genomic region 315, the percentage of highly methylated fragments overlapping genomic region 315, and the percentage of highly unmethylated fragments overlapping genomic region 315. The total number of methylation states of CpG sites 317 on all fragments 420, divided by the total number of CpG sites on fragment 320, derives the methylation density. In this example, the methylation density is 0.63 or 63%. Fragment 1 320A, fragment 2 320B, and fragment 3 320C are highly methylated, with a methylation degree of more than 80% on the entire fragment (at least as shown in the figure). Fragment 4 320D is highly unmethylated, with a degree of unmethylation of more than 80% on the entire fragment (at least as shown in the figure). Fragment 5 320E and fragment 6 320F are mixed methylated, i.e., neither highly methylated nor highly unmethylated. Therefore, a methylation feature, i.e., the percentage of highly methylated fragments overlapping genomic region 315, is 0.50 or 50%. Another methylation feature, the percentage of highly unmethylated fragments overlapping genomic region 315, is 0.17 or 17%. The fragments overlapping genomic region 315 may be fragments overlapping at least one of CpG sites 317. In some embodiments, the fragments overlapping genomic region 315 overlap at least a certain percentage of CpG sites 317, for example, at least 20% of CpG sites 317 of genomic region 315.

[0147] According to one or more embodiments, Figure 4A The training 400 of the age prediction model is illustrated. The analysis system may perform part or all of the training 400. In other embodiments, Fig. 6A and Figure 6B Other components in may perform part or all of training 400. Training 400 may produce a trained age prediction model, which may input a set of methylation features of age-informative genomic regions and output a predicted age. The process of training 400 the age prediction model can be similarly applied to training other covariate prediction models. In embodiments involving other covariates, the analysis system trains the covariate prediction model using training samples with reported values.

[0148] The analysis system acquires 405 multiple training samples. The multiple training samples may include (1) only cancer training samples, (2) only non-cancer training samples, or (3) a combination of cancer and non-cancer training samples. Each cancer training sample is collected from an individual diagnosed with cancer. The diagnosis of cancer may occur before or after the sample is collected. In some instances, the type of cancer may be known. Each non-cancer training sample is collected from an individual without any cancer diagnosis and may generally be considered a healthy individual. Each training sample contains genetic material that can be sequenced and analyzed. In some embodiments, the sample is a blood sample containing nucleic acid fragments (e.g., cfDNA fragments). In addition, each training sample includes a chronological age reported by the individual. That is, each training sample is labeled with the chronological age of the subject from whom the sample was obtained. For example, if a cancer sample is obtained from a 39-year-old male, then the cancer training sample will be labeled as obtained from a 39-year-old male. In some cases, the label may include errors (e.g., due to sample exchange between subjects). For example, a non-cancer training sample may be labeled as obtained from a 62-year-old female, when in fact it was taken from a 76-year-old male.

[0149] The analysis system sequences 410 the nucleic acid fragments in each training sample to identify a methylation pattern for each nucleic acid fragment. Again, the methylation pattern represents the methylation status of CpG sites in the DNA fragments within the genomic region of the sample. The methylation pattern may be determined relative to other samples or the sample population. Sequencing may involve bisulfite sequencing to convert unmethylated CpG sites. In other embodiments, the sequencer sequences the nucleic acid fragments and the analysis system processes the sequence reads to determine the methylation pattern. The analysis system may further perform one or more processing steps on the sequence reads, such as deduplicating copies of the same original fragment, identifying 36 contaminated fragments, identifying sequencing errors, etc. The process of sequencing nucleic acid fragments and determining methylation patterns has been described above. Figure 2A and Figure 2B was discussed in .

[0150] For each genomic region, the analysis system calculates 415 an indicative score representing the correlation between the chronological age and the methylation pattern of the nucleic acid fragments overlapping the genomic region. More generally, the indicative score represents the correlation between the chronological age of the subject and the methylation pattern. The analysis system may determine one or more methylation signatures for each genomic region in an initial set of genomic regions. The initial set of genomic regions may be a broad set covering most CpG sites in the human genome. For example, for a single CpG site genomic region, the analysis system may determine Figure 3AA certain combination of the methylation features described in, for example, a certain combination of methylation density, overlapping highly methylated fragment percentages and overlapping highly unmethylated fragment percentages. The analysis system can calculate an indicative score for the genomic region by training a temporal age regression based on the methylation features of the genomic region. The analysis system uses the methylation features extracted for each training sample and the reported age of the training sample (for example, the temporal age of the sample marked) to train the regression. Regression types include linear regression, logarithmic regression, exponential regression, multivariate regression, logistic regression, polynomial regression, lasso regression, etc. From these trained regressions, the analysis system may measure various indicators for use as indicative scores. Example indicators may include, but are not limited to, covariance, Pearson correlation coefficient (or "Pearson correlation" for short), R2, residual sum of squares (RSS), total sum of squares (TSS), t-statistic of slope, two-tailed p-value of t-statistic (possibly adjusted, for example, for multiple hypothesis testing), other statistical indicators related to regression, etc.

[0151] For ease of illustration, the analysis system calculates a multivariate temporal age regression based on methylation density, the percentage of overlapping highly methylated fragments, and the percentage of overlapping highly unmethylated fragments, etc., which occurs in the genomic region. Based on the trained multivariate regression, the analysis system may determine a Pearson correlation, which can be used as an indicative score for the genomic region. In another example, the analysis system calculates a linear temporal age regression based only on methylation density, which occurs in the genomic region, and may determine a Pearson correlation as an indicative score. In some embodiments, the analysis system may train multiple regressions for each genomic region based on different sets of methylation features considered, for example, a first regression based only on methylation density, a second regression based only on the percentage of overlapping highly methylated fragments, a third regression based only on the percentage of overlapping highly unmethylated fragments, and a fourth regression based on a combination of multiple features of the above three methylation features.

[0152] The analysis system generates 420 a genomic region feature set based on the covariance score, which is used as a feature in the age prediction model. To this end, the analysis system may calculate one or more indicative scores for each region in the group of initial genomic regions, for example, an indicative score is calculated for each group of methylation features. In a genomic region, the group of methylation features that achieves the highest absolute indicative score is determined to be a group of methylation features that can provide the most information in the genomic region. In one embodiment, the analysis system uses a threshold absolute indicative score to identify these genomic regions to be used as a part of the genomic region feature set. For example, the threshold absolute indicative score is 0.5, so an indicative score higher than 0.5 or lower than -0.5 (for example, the absolute value of the indicative score is higher than 0.5) exceeds the threshold absolute indicative score. In turn, the analysis system generates feature vectors, which include genomic regions with identified indicative scores higher than the threshold. For example, the feature vector may include all genomic regions with absolute indicative scores higher than 0.5 or lower than -0.5. In another embodiment, the analysis system sorts the genomic regions based on its highest indicative score. Based on the ranking, the analysis system may select a sufficient number of genomic regions to exhaust the genomic region budget and generate feature vectors accordingly, for example, these regions may be the regions targeted by the detection panel.

[0153] In other embodiments, the analysis system further considers one or more other factors of the genomic region to determine the feature set of the genomic region, and these factors are therefore included in the feature vector generated. For example, the analysis system may further consider the balance of negatively correlated and positively correlated genomic regions. The negatively correlated genomic region reflects that the increase of the methylation characteristic value is relevant to the reduction of the age value. The positively correlated genomic region reflects that the increase of the methylation characteristic value is relevant to the increase of the age value. The analysis system may further consider the rate of change between age and the methylation characteristic. For example, some genomic regions may be gently related to age (methylation characteristic difference is less when individual age changes), some genomic regions may be stably related to age (methylation characteristic difference is medium when individual age changes), and some genomic regions may be sharply related to age (methylation characteristic difference is larger when individual age changes). The analysis system may further consider the position of genomic region in the human genome, for example, ensure that genomic region is appropriately distributed in the human genome. This can prevent the situation that all age information genomic regions are concentrated in a part of the human genome, and the part may be sparse in the test sample to be applied for age prediction for a variety of reasons, if all age information genomic regions are in that part, this will affect age prediction.

[0154] In some embodiments, the analysis system uses penalized regression to identify feature sets for genomic regions and generates feature vectors accordingly. The purpose of the penalty process is to optimize the feature set used to the minimum feature set that still provides the best predictive power. Other embodiments use relaxed lasso regression to achieve similar results.

[0155] In some embodiments, the analysis system prunes 425 the feature set of genomic regions to genomic regions that have a high correlation with the cancer signal. That is, the analysis system may remove or exclude genomic regions that are not highly correlated with the cancer signal from the feature vector or feature set. For example, the analysis system may separately identify genomic regions that are associated with a cancer signal (or another disease signal). The analysis system may then determine genomic regions where the correlation with age and the correlation with the cancer signal intersect. A method for identifying genomic regions associated with a cancer signal is disclosed below (e.g., in combination with Fig. 6A and Figure 6B ). Utilizing genomic regions that intersect correlation with age and correlation with cancer signatures can more efficiently utilize the budget for targeted regions on a test panel. That is, creating a panel that uses fewer probes that target regions that are more indicative of age and / or the presence of cancer. Essentially, probes that target regions that are strongly correlated with cancer and / or age are more "valuable" to include in a panel than probes that are weakly correlated with cancer and / or age. Additionally, utilizing intersecting genomic regions may prove advantageous when predicting age as a feature for cancer classification.

[0156] At this point, the analysis system has identified the genomic region feature set for training the age prediction model, and has generated a corresponding feature vector. Each genomic region in the feature set (included in the feature vector) may contain a single CpG site, may contain multiple CpG sites, or may contain multiple groups of CpG sites. Each genomic region has a specific set of methylation features in the feature set, and this set of features may be different from other genomic regions. For example, the first genomic region covers a single CpG site and considers the methylation density of the CpG site, and the second genomic region covers multiple CpG sites and considers both the overlapping highly methylated fragment percentage and the overlapping highly unmethylated fragment percentage. In other embodiments, the analysis system implements a penalty to minimize the number of genomic regions that maintain the accuracy of age prediction. Penalty is a factor that negatively affects age prediction based on the number of genomic regions. Penalty forces the analysis system to determine the minimum number of genomic regions required to maintain the optimal performance of the age prediction model.

[0157] The analysis system trains 430 age prediction models based on the methylation patterns of nucleic acid fragments overlapping with the genomic region feature set from the training sample. For each training sample, the analysis system determines the methylation feature value of the genomic region feature set. In certain embodiments, the analysis system trains the age prediction model as a machine learning model. Example machine learning models include linear regression, logarithmic regression, exponential regression, multivariate regression, logistic regression, polynomial regression, lasso regression, etc. The analysis system may train multiple age prediction models using different genomic region feature sets to evaluate the performance of each model. For example, a first model is trained on a small genomic region feature set, and a second model is trained on a large genomic region feature set, and the large feature set contains the small feature set. The analysis system uses the validation set of the training sample to evaluate the performance of the two age prediction models.

[0158] According to one or more embodiments, Figure 4B Deployment 440 of a temporal age prediction model is illustrated. The analysis system may perform part or all of deployment 440. In other embodiments, Fig. 6A and 6B Other components in may perform part or all of deployment 440. Deployment 440 of the age prediction model includes determining an age prediction for an individual associated with a test sample based on the methylation signature of the genomic region signature set of the test sample.

[0159] The analysis system obtains 445 a test sample with a plurality of nucleic acid fragments and a reported age (e.g., a label of the chronological age of the subject from which the sample was obtained). A doctor or other healthcare provider collects the test sample and may also obtain the reported age of the individual providing the test sample. In some embodiments, age may be a single value or an age range. For example, an individual may report an age of 47 or may report an age range of 40-50 years old. The sample may be any type of biological sample containing individual nucleic acid material. In embodiments using a blood sample, the blood sample contains at least cfDNA fragments sheared from cells.

[0160] The analysis system sequences 450 the nucleic acid fragments in the test sample to identify the methylation pattern for each nucleic acid fragment. The sequencing may involve bisulfite sequencing to convert unmethylated CpG sites. In other embodiments, the sequencer sequences the nucleic acid fragments and the analysis system processes the sequence reads to determine the methylation pattern. The analysis system may further perform one or more processing steps on the sequence reads, such as removing duplicate copies of the same original fragment, identifying contaminating fragments, identifying sequencing errors, etc. The process of sequencing nucleic acid fragments and determining methylation patterns has been described above. Figure 2A and Figure 2B was discussed in .

[0161] The analysis system applies 455 the trained age prediction model to predict the chronological age for the test sample based on the methylation pattern of the nucleic acid fragments of the test sample. The analysis system determines the methylation signature for the age prediction model, e.g. Figure 4A The process 400 trains a model. The age prediction model is configured to receive methylation features of a genomic region feature set as input and output a predicted chronological age based on these methylation features. Like the reported age, the predicted age may be a single value or an age range.

[0162] The analysis system compares the predicted age to the reported age 460. The comparison may be to determine whether the predicted age matches the reported age. For example, if the reported age is in the age range of 20-30 years, and the predicted age is 26 years or in the range of 20-30 years, then the predicted age matches the reported age range. In some embodiments, the comparison may be a residual, where the age residual is the difference between the predicted age and the reported age. For example, if the reported age is 63 years, and the predicted age is 72 years, then the residual is 9 years older than the reported age. The residual may also be an absolute value, for example, 9 years older than the reported age.

[0163] The analysis system continues the analysis using the predicted age. In some embodiments, the analysis system may perform 465 sample swap verification. In this context, sample swap verification can refer to identifying whether the test sample is correctly labeled. For example, if a sample obtained from a 54-year-old female is labeled as a sample obtained from a 24-year-old male, a "sample swap" may occur. However, more generally, identifying a sample swap is similar to identifying that the reported age of a test sample is incorrect.

[0164] In some configurations, the analysis system may utilize residuals to determine whether the sample is exchanged, so as to determine that the test sample is not really derived from the individual expected to be associated with the sample. If the predicted age is different from the reported age, the analysis system may identify or identify the sample exchange. In other embodiments, if the residual is higher than the threshold difference, the analysis system may identify the sample exchange. For example, the residual threshold can be set to 10 years old, so if the residual between the predicted age and the reported age is greater than the residual threshold of 10 years old, the analysis system can identify the sample exchange. In other embodiments, the analysis system may identify the sample exchange based on the comparison of the predicted age and the reported age and in combination with other analyses. For example, the analysis system may train a separate model for race determination to determine whether the predicted race matches the individual reported race, or train a separate model for gender determination to determine whether the predicted gender matches the individual reported gender. After identifying the sample exchange, the analysis system may prevent the sample from entering the downstream analysis. The sample that is not identified as a sample exchange may continue to enter the downstream analysis. For example, after identifying a training sample as a sample exchange, the analysis system may prevent the training sample from being used to train one or more models or build one or more distributions. As another example, upon identifying a test sample as a sample exchange, the analysis system may refrain from making a cancer prediction for the test sample.

[0165] The analysis system may use the comparison of the predicted age to the reported age as part of the cancer classification 470. In one or more embodiments, the analysis system uses the residual of the predicted age to the reported age as a feature for cancer classification, for example, in combination with other features extracted from the sequencing data. For example, the residual may indicate that the amount of methylation in the test sample is abnormally high or low (resulting in a high residual), which, as described above, indicates the presence or absence of cancer. Therefore, a high residual may be used as a suggestive feature for determining whether the test sample has cancer.

[0166] In some embodiments, the analysis system may compare the residual with a residual threshold to determine whether the sample has a strong possibility of cancer. The analysis system may use a set of training samples to set the residual threshold. The analysis system identifies the methylation features of the genomic region feature set for each training sample. The analysis system inputs the methylation features of each training sample into the age prediction model to determine the predicted age for each training sample. The analysis system may calculate the residual for each training sample by calculating the difference between the predicted age and the reported age. The analysis system may set a residual threshold that covers the vast majority of training samples. For example, the analysis system hopes to use a residual threshold that captures 75%, 80%, 85%, 90%, 95%, 97%, 98%, 99%, 99.5% or 99.9% of the training samples. In fact, setting the residual threshold is to train the model to recognize that the samples whose methylation patterns cause errors in the prediction of the temporal age are related to the presence or absence of cancer. Therefore, the residual threshold is used as a metric to determine whether the amount of difference observed in the methylation pattern is attributed to age or to cancer. In view of this, if the residual of the test sample is outside the residual threshold, the analysis system can preliminarily determine that the test sample has a strong possibility of cancer. Depending on the configuration, the strong possibility can indicate that the sample is more likely to indicate cancer, including a probability of cancer of, for example, at least 60%, 65%, 70%, 80%, 90%, etc., or may have an indicative score above a threshold. The analysis system can continue to perform cancer classification to confirm the preliminary determination.

[0167] III.B. Identification of aberrant fragments

[0168] The analysis system can use the methylation state vector of the sample to determine the abnormal fragment of the sample. For each fragment in the sample, the analysis system can use the methylation state vector corresponding to the fragment to determine whether the fragment is an abnormal fragment. In some embodiments, the analysis system calculates a p-value score for each methylation state vector, and the score describes the probability of observing the methylation state vector or other less likely methylation state vectors in a healthy control group. In some instances, the p-value score may be adjusted, for example, by controlling the false positive rate, family error rate, false discovery rate, etc., for multiple hypothesis testing. The process of calculating the p-value score is further discussed in Section II.Ci below. P-value screening method. The analysis system may determine the fragments whose p-value score of the methylation state vector is lower than the threshold as abnormal fragments. In some embodiments, the analysis system further marks the fragments having at least a certain number of CpG sites and the percentage of methylation or unmethylation exceeding a certain threshold as hypermethylated fragments and hypomethylated fragments. Hypermethylated fragments or hypomethylated fragments can also be referred to as extreme methylation abnormal fragments (UFXM). In other embodiments, the analysis system may use a variety of other probability models to determine abnormal fragments. Examples of other probabilistic models include hybrid models, deep probabilistic models, etc. In some embodiments, the analysis system may use any combination of multiple processes described below to identify abnormal fragments. Through the identified abnormal fragments, the analysis system may screen the methylation state vector set of the sample for use in other processes, such as for training and deploying cancer classifiers.

[0169] III. CIp value screening method

[0170] In some embodiments, the analysis system calculates a p-value score for each methylation state vector compared to the methylation state vector of the fragment in the healthy control group. The p-value score can describe the probability of observing a methylation state that matches the methylation state vector or other less likely methylation state vectors in the healthy control group. In order to determine whether a DNA fragment is abnormally methylated, the analysis system can use a healthy control group in which most fragments are normally methylated. When performing this probabilistic analysis for determining abnormal fragments, the determination is of great significance when compared with a group of control subjects that constitute the healthy control group. In order to ensure the robustness of the healthy control group, the analysis system may select a certain threshold number of healthy individuals to obtain samples including DNA fragments. The following Figure 5A Methods are described for generating a healthy control group data structure that an analysis system may use to calculate p-value scores. Figure 5B A method for calculating p-value scores using the generated data structure is described.

[0171] According to one embodiment, Figure 5A200 .

[0172] For the methylation state vector of each fragment, the analysis system may subdivide 510 the methylation state vector into strings of CpG sites. In some embodiments, the analysis system subdivides 510 the methylation state vector so that the resulting strings are all less than a given length. For example, a methylation state vector of length 11 may be subdivided into strings of length less than or equal to 3, resulting in 9 strings of length 3, 10 strings of length 2, and 11 strings of length 1. In another example, a methylation state vector of length 7 is subdivided into strings of length less than or equal to 4, resulting in 4 strings of length 4, 5 strings of length 3, 6 strings of length 2, and 7 strings of length 1. If a methylation state vector is shorter than or equal to a specified string length, then the methylation state vector may be converted into a single string containing all CpG sites of the vector.

[0173] The analysis system counts 515 these strings by calculating, for each possible CpG site and methylation state possibility in the vector, the number of strings in the control group that have the specified CpG site as the first CpG site in the string and have the methylation state possibility. For example, at a given CpG site and considering a string length of 3, there are 2^3 or 8 possible string configurations. At that given CpG site, for each of the 8 possible string configurations, the analysis system counts 510 the number of times each methylation state vector possibility occurs in the control group. Continuing with this example, this may involve counting the following quantities: For each starting CpG site x in the reference genome, <M x , M x+1 , M x+2 >、 <M x , M x+1 , U x+2 >、...、 x , U x+1 , U x+2 >. The analysis system creates 515 a data structure that stores statistical counts of each starting CpG site and string likelihood.

[0174] ​There are several benefits to setting an upper limit on string length. First, the size of the data structure created by the analysis system increases dramatically depending on the maximum length of the string. For example, a maximum string length of 4 means that there are at least 2^4 numbers to count for each CpG site for a string of length 4. Increasing the maximum string length to 5 means that there are an additional 2^4 or 16 numbers to count for each CpG site, doubling the numbers to count (and the computer memory required) compared to the previous string length. Reducing the string size can help keep the creation and performance of the data structure (e.g., for subsequent access, as described below) reasonable in terms of computation and storage. Second, a statistical consideration for limiting the maximum string length can be to avoid overfitting downstream models that use string counts. If long strings at CpG sites do not have a strong biological impact on the outcome (e.g., the prediction of an abnormality that is predictive of the presence of cancer), then calculating probabilities based on large strings of CpG sites may be problematic because it will use a large amount of data that may not be available and therefore may be too sparse for the model to perform well. For example, to calculate the probability of an abnormality / cancer condition based on the top 100 CpG sites, one could use the counts of strings of length 100 in a data structure, ideally some of which would exactly match the methylation states of the top 100. If only sparse counts of strings of length 100 are available, there may not be enough data to determine whether a given string of length 100 in a test sample is abnormal.

[0175] According to one embodiment, Figure 5B is a flow chart describing a process 530 for identifying aberrantly methylated fragments from an individual. In process 530, an analysis system generates 540 methylation state vectors from cfDNA fragments of a subject, for example, via process 200. The analysis system may process each methylation state vector as follows.

[0176] For a given methylation state vector, the analysis system enumerates 545 all possible methylation state vectors that have the same starting CpG site and the same length (i.e., CpG site set) as the given methylation state vector. Since each methylation state is typically either methylated or unmethylated, there can actually be two possible states at each CpG site, so the number of different possibilities for the methylation state vector can depend on a power of 2, such that a methylation state vector of length n will be associated with 2 n For methylation state vectors that include one or more CpG sites with uncertain states, the analysis system may enumerate 530 the possibilities of the methylation state vector by considering only those CpG sites with observable states.

[0177] The analysis system calculates 550 the probability of observing the possibility of each methylation state vector for the identified starting CpG site and the length of the methylation state vector by accessing the healthy control group data structure. In some embodiments, when calculating the probability of observing a given possibility, the Markov chain probability is used to model the joint probability calculation. The Markov model can be trained at least in part based on the evaluation of the methylation state of each CpG site in the corresponding multiple CpG sites of the corresponding fragments (e.g., nucleic acid methylation fragments) in the healthy non-cancer cohort data set with corresponding multiple CpG sites. For example, a Markov model (e.g., a hidden Markov model or HMM) is used to determine that for a nucleic acid methylation fragment in multiple nucleic acid methylation fragments, given a set of probabilities (for each state in the sequence, the set of probabilities determines the possibility of observing the next state in the sequence), the probability of observing a certain methylation state (including, for example, "M" or "U") sequence. The group probability can be obtained by training the HMM. Such training can involve calculating statistical parameters (e.g., the probability that a first state can be converted to a second state (conversion probability) and / or the probability that a given methylation state can be observed at a corresponding CpG site (emission probability)) given an initial training data set of observed methylation state sequences (e.g., methylation patterns). The HMM can be trained using supervised training (e.g., using samples in which both the base sequence and the observed states are known) and / or unsupervised training (e.g., Viterbi learning, maximum likelihood estimation, expectation maximization training, and / or Baum-Welch training). In other embodiments, a computational method other than Markov chain probability is used to determine the probability of observing the likelihood of each methylation state vector. For example, such a computational method can include a learning representation. The p-value threshold can be between 0.01 and 0.10, or between 0.03 and 0.06. The p-value threshold can be 0.05. The p-value threshold can be less than 0.01, less than 0.001, or less than 0.0001.

[0178] The analysis system calculates 555 a p-value score for the methylation state vector using the calculated probabilities for each possibility. In some embodiments, this includes identifying a calculated probability corresponding to a possibility that matches the methylation state vector in question. In particular, this can be a possibility that has the same set of CpG sites as the methylation state vector or similarly has the same starting CpG site and length as the methylation state vector. The analysis system can add the calculated probabilities for any possibilities that have a probability less than or equal to the identified probability to generate a p-value score.

[0179] This p-value can represent the probability of observing the methylation state vector of the fragment in the healthy control group or other less likely methylation state vectors. Therefore, a low p-value score can generally correspond to a methylation state vector that is rare in healthy individuals and causes the fragment to be marked as abnormally methylated relative to the healthy control group. A high p-value score can generally be associated with a methylation state vector that is expected to exist in a relatively healthy individual. For example, if the healthy control group is a non-cancer group, a low p-value can indicate that the fragment is abnormally methylated relative to the non-cancer group, and therefore may indicate the presence of cancer in the test subject.

[0180] As described above, the analysis system can calculate a p-value score for each of a plurality of methylation state vectors, each methylation state vector representing a cfDNA fragment in the test sample. In order to identify which fragments are abnormally methylated, the analysis system may filter 565 the set of methylation state vectors based on their p-value scores. In some embodiments, the screening is performed by comparing the p-value score to a threshold and retaining only those fragments below the threshold. This threshold p-value score can be on the order of 0.1, 0.01, 0.001, 0.0001, or a similar order of magnitude.

[0181] According to the example results of process 500, the analysis system can generate a median number (range) of 2,800 (1,500-12,000) of segments with abnormal methylation patterns in participants without cancer during training, and a median number (range) of 3,000 (1,200-420,000) of segments with abnormal methylation patterns in participants with cancer during training. These screening segments with abnormal methylation patterns may be used for downstream analysis as described in Part III below.

[0182] In some embodiments, the analysis system uses 560 a sliding window to determine the likelihood of a methylation state vector and calculate a p-value. The analysis system only needs to enumerate the likelihood and calculate the p-value for a window containing consecutive CpG sites, where the (CpG site) length of the window is shorter than at least some fragments (otherwise, the window would be meaningless), without enumerating the likelihood and calculating the p-value for all methylation state vectors. The window length may be static, user-determined, dynamic, or otherwise selected.

[0183] When calculating the p-value of a methylation state vector that is larger than a window, the window can start from the first CpG site in the vector to identify a set of continuous CpG sites in the vector within the window. The analysis system can calculate a p-value score for the window that includes the first CpG site. The analysis system can then "slide" the window to the second CpG site in the vector and calculate another p-value score for the second window. Therefore, for a window size of l and a methylation vector length of m, each methylation state vector can generate m–l+1 p-value scores. After completing the p-value calculation for each part of the vector, the lowest p-value score in all sliding windows can be used as the overall p-value score for the methylation state vector. In other embodiments, the analysis system aggregates the p-value scores of the methylation state vectors to generate an overall p-value score.

[0184] The use of sliding windows can help reduce the number of methylation state vector possibilities that need to be enumerated and the corresponding probability calculations that would otherwise have to be performed. As a practical example, a fragment may have up to 54 CpG sites. The analysis system can use (for example) a window size of 5 to perform 50 p-value calculations for each of the 50 windows of the methylation state vector of the fragment, without having to calculate the probabilities of 2^54 (approximately 1.8×10^16) possibilities in order to generate a single p-value. Each of these 50 calculations can enumerate 2^5 (32) methylation state vector possibilities, for a total of 50×2^5 (1.6×10^3) ​​probability calculations. This can greatly reduce the calculations that need to be performed without significantly affecting the accurate identification of abnormal fragments.

[0185] In the uncertain state embodiment, the analysis system may calculate a p-value score that sums the CpG sites with uncertain states in the methylation state vector of the fragment. The analysis system may identify all possibilities that are consistent with all methylation states in the methylation state vector except the uncertain state. The analysis system may assign a probability to the methylation state vector that is the sum of the probabilities of these identified possibilities. For example, the analysis system may calculate the methylation state vector<M1,I2,U3> The probability of methylation as the state vector<M1,M2,U3> and<M1,U2,U3> Because the methylation states of CpG sites 1 and 3 have been observed and are consistent with the methylation states of the fragment at CpG sites 1 and 3. This method of summing the CpG sites with uncertain states can use up to 2^i possibilities for probability calculation, where i represents the number of uncertain states in the methylation state vector. In other embodiments, a dynamic programming algorithm may be used to calculate the probability of a methylation state vector with one or more uncertain states. Advantageously, the dynamic programming algorithm runs in linear computation time.

[0186] In some embodiments, by caching at least some of the calculations, the computational load of calculating probabilities and / or p-value scores may be further reduced. For example, the analysis system may cache the probability calculations of the possibilities of the methylation state vector (or its window) in a transient or permanent memory. If other fragments have the same CpG site, caching the probabilities of these possibilities can allow the p-value scores to be calculated efficiently without recalculating the probabilities of the potential possibilities. Equivalently, the analysis system may calculate the p-value score for each possibility of the methylation state vector associated with a set of CpG sites in the vector (or its window). The analysis system may cache these p-value scores for determining the p-value scores of other fragments including the same CpG site. In general, the p-value scores calculated for multiple possibilities with the same CpG site in the methylation state vector can be used to determine the p-value score of another of these multiple possibilities under the same set of CpG sites.

[0187] In some instances, the p-value score may be adjusted for multiple hypothesis testing according to a variety of suitable techniques. For example, but not limited to, the technique may include controlling the false positive rate, family-wise error rate, experimental error rate, false discovery rate, etc. Known and applicable techniques for adjusting p-values ​​include, but are not limited to, the Bonferroni procedure, the Holm procedure, the Hochberg procedure, the harmonic mean p-value procedure, the Benjamini-Hochberg procedure, the Benjamini-Yekutieli procedure, the Storey-Tibshirani procedure, etc. Adjusting the p-value score for multiple hypothesis testing can be used to improve the accuracy of positive test determination based on the p-value and reduce the incidence of false positives.

[0188] Before training a regional model or a cancer classifier, one or more nucleic acid methylation fragments may be screened. Screening nucleic acid methylation fragments may include removing each corresponding nucleic acid methylation fragment that does not meet one or more selection criteria (e.g., lower than or higher than a selection criterion) from the corresponding multiple nucleic acid methylation fragments. One or more selection criteria may include a p-value threshold. The output p-value of the corresponding nucleic acid methylation fragment may be determined based at least in part on comparing the corresponding methylation pattern of the corresponding nucleic acid methylation fragment with the corresponding methylation pattern distribution of the nucleic acid methylation fragments of the corresponding multiple CpG sites of the corresponding nucleic acid methylation fragment in the healthy non-cancer cohort data set.

[0189] Screening multiple nucleic acid methylation fragments can include removing each corresponding nucleic acid methylation fragment that does not meet the p-value threshold.Use the methylation pattern observed in the first batch of multiple nucleic acid methylation fragments, the filter is applied to the methylation pattern of each corresponding nucleic acid methylation fragment.Each corresponding methylation pattern of each corresponding nucleic acid methylation fragment (e.g., fragment one, ..., fragment N) can include corresponding one or more methylation sites (e.g., CpG sites), which are identified by methylation site identifiers and corresponding methylation patterns, and are expressed as a string of sequences comprising multiple 1s and multiple 0s, wherein each "1" represents a methylated CpG site in these one or more CpG sites, and each "0" represents an unmethylated CpG site in these one or more CpG sites.The methylation pattern observed in the first batch of multiple nucleic acid methylation fragments can be used to construct a methylation state distribution for these CpG site states, and these states are represented jointly by the first batch of multiple nucleic acid methylation fragments (e.g., CpG site A, CpG site B, ..., CpG site ZZZ). For more detailed information on nucleic acid methylation fragment processing, please refer to U.S. Provisional Patent Application No. 17 / 191,914, filed on March 4, 2021, entitled "Systems and Methods for Cancer Condition Determination Using Autoencoders," which is hereby incorporated by reference in its entirety.

[0190] When the abnormal methylation score of the corresponding nucleic acid methylation fragment is lower than the abnormal methylation score threshold, the nucleic acid methylation fragment may not meet the selection criteria in one or more selection criteria. In this case, the abnormal methylation score can be determined by a hybrid model. For example, the hybrid model can determine the possibility of the methylation state vector (e.g., methylation pattern) of the corresponding nucleic acid methylation fragment based on the number of possible methylation state vectors of the same length and at the same corresponding genomic position, thereby detecting the abnormal methylation pattern in the nucleic acid methylation fragment. This process can be performed by generating multiple possible methylation states for vectors of a specified length at each genomic position of the reference genome. Using multiple possible methylation states, the total number of possible methylation states can be determined, and then the probability of each predicted methylation state at the genomic position is determined. Then, by matching the sample nucleic acid methylation fragment with the predicted (e.g., possible) methylation state, and retrieving the calculated probability of the predicted methylation state, the possibility of the sample nucleic acid methylation fragment corresponding to the genomic position in the reference genome can be determined. Then, based on the probability of the sample nucleic acid methylation fragment, the abnormal methylation score can be calculated.

[0191] When the number of residues of the corresponding nucleic acid methylation fragment is lower than the threshold number, the nucleic acid methylation fragment cannot meet the selection criteria in one or more selection criteria. The threshold number of residues can be between 10 and 50, between 50 and 100, between 100 and 150, or more than 150. The threshold number of residues can be a fixed value between 20 and 90. When the number of CpG sites of the corresponding nucleic acid methylation fragment is lower than the threshold number, the nucleic acid methylation fragment may not meet the selection criteria in one or more selection criteria. The threshold number of CpG sites can be 4, 5, 6, 7, 8, 9 or 10. When the genomic starting position and the genomic ending position of the corresponding nucleic acid methylation fragment indicate that the number of nucleotides represented by the nucleic acid methylation fragment in the human genome reference sequence is lower than the threshold number, the nucleic acid methylation fragment cannot meet the selection criteria in one or more selection criteria.

[0192] Screening can remove nucleic acid methylation fragments having the same corresponding methylation pattern, the same corresponding genome starting position and the same genome ending position as another nucleic acid methylation fragment in these corresponding multiple nucleic acid methylation fragments in the screening. This screening step can remove completely repeated redundant fragments, in some cases, including PCR repeat fragments. Screening can remove nucleic acid methylation fragments having the same corresponding genome starting position and the same genome ending position as another nucleic acid methylation fragment in the corresponding multiple nucleic acid methylation fragments, and the number of different methylation states is lower than the threshold number. The threshold number of different methylation states for retaining nucleic acid methylation fragments can be 1, 2, 3, 4, 5 or more than 5. For example, a first nucleic acid methylation fragment has the same corresponding genome starting and ending position as a second nucleic acid methylation fragment, but has at least 1, at least 2, at least 3, at least 4 or at least 5 different methylation states at the corresponding CpG site (e.g., aligned with a reference genome), then retain the first nucleic acid methylation fragment. For another example, if a first nucleic acid methylation fragment has the same methylation state vector (eg, methylation pattern) as a second nucleic acid methylation fragment, but has different corresponding genomic start and end positions, the first nucleic acid methylation fragment is also retained.

[0193] Screening can remove detection artifacts in multiple nucleic acid methylation fragments. Removing detection artifacts can include removing sequence reads obtained from sequencing hybridization probes and / or sequence reads obtained from sequences that were not converted during bisulfite conversion. Screening can remove contaminants (e.g., contaminants generated by sequencing, nucleic acid isolation, and / or sample preparation).

[0194] Screening can be based on the mutual information screening of the corresponding methylation fragments and the cancer state in multiple training subjects, and a methylation fragment subset is removed from multiple methylation fragments. For example, mutual information can measure the degree of interdependence between two conditions of interest sampled simultaneously. Mutual information can be determined by selecting an independent CpG site set (for example, in all or part of the nucleic acid methylation fragment) from one or more data sets, and comparing the probability of the methylation state of the CpG site set between two sample groups (for example, a subset and / or group of a genotype data set, a biological sample and / or a subject). The mutual information score can be represented at the corresponding region in the corresponding frame of the sliding window, the probability of the methylation pattern of the first condition and the second condition, thereby prompting the discrimination ability of the corresponding region. When the sliding window is moved forward on a selected CpG site set and / or a selected genomic region, a mutual information score can be similarly calculated for each region in each frame of the sliding window. For more details on mutual information screening, see U.S. patent application 17 / 119,606, filed on December 11, 2020, titled “Cancer Classification Using Patched Convolutional Neural Networks,” which is hereby incorporated by reference in its entirety.

[0195] III.C.II. Hypermethylated and hypomethylated fragments

[0196] In some embodiments, the analysis system identifies 570 and determines hypomethylated fragments or hypermethylated fragments from the screened set as abnormal fragments. The analysis system identifies hypermethylated fragments, which have more than a threshold number of CpG sites and more than a threshold percentage of these CpG sites are methylated. The analysis system identifies hypomethylated fragments, which have more than a threshold number of CpG sites and more than a threshold percentage of these CpG sites are unmethylated. Example thresholds for fragment (or CpG site) lengths include more than 3, 4, 5, 6, 7, 8, 9, 10, etc. Example thresholds for methylation or unmethylation percentages include more than 80%, 85%, 90% or 95%, or any other percentage in the range of 50%-100%.

[0197] III.C. Training of Cancer Classifier

[0198] According to one embodiment, Fig. 6A6 is a flowchart describing a process 600 for training a cancer classifier. The analysis system obtains 610 multiple training samples, each of which has an abnormal fragment set and a cancer type label. The multiple training samples may include any combination of samples, which are from healthy individuals with an overall label of "non-cancer"; from subjects with an overall label of "cancer" or with specific labels (e.g., "breast cancer", "lung cancer", etc.). The training samples from subjects for a cancer type may be referred to as the cancer type cohort or cancer type cohort.

[0199] The analysis system determines 620 feature vectors for each training sample based on the abnormal fragment set of the training sample. The analysis system can calculate an abnormal score for each CpG site in the initial CpG site set. The initial CpG site set may be all CpG sites in the human genome or a portion thereof - the order of magnitude may be 10 4 , 10 5 , 10 6 , 10 7 , 10 8 Etc. In one embodiment, the analysis system uses a binary score to define an abnormality score for the feature vector based on whether there are abnormal fragments in the set of abnormal fragments covering the CpG site. In another embodiment, the analysis system defines the abnormality score based on the count of abnormal fragments overlapping the CpG site. In one instance, the analysis system may use a three-category score to assign a first score to the absence of abnormal fragments, a second score to the presence of a small number of abnormal fragments, and a third score to the presence of multiple abnormal fragments. For example, the analysis system counts 5 abnormal fragments in a sample overlapping the CpG site, and calculates the abnormality score based on the count of 5. In one or more embodiments, the feature vector further includes a score based on Figure 4A and Figure 4B One or more features of the chronological age prediction (e.g., covariate prediction) described herein. For example, the feature vector may include age residuals as the difference between the predicted chronological age (e.g., obtained by applying the trained age prediction model through process 440) and the reported chronological age. In other examples, the feature vector may include other features based on the predicted covariates, such as one or more predicted covariates. In some embodiments, the feature vector further includes a set of features evaluated in the chronological age prediction (e.g., in Figure 4A One or more methylation features in the feature set of the temporal age prediction model determined in step 420).

[0200] Once all abnormal scores are determined for the training sample, the analysis system can determine the feature vector as a vector comprising a plurality of elements, each element in the vector being one of the abnormal scores associated with one of the CpG sites in the initial set. The analysis system can normalize the abnormal scores of the feature vector based on the coverage of the sample. Here, coverage can refer to the median or average sequencing depth on all CpG sites covered by the initial CpG site set used in the classifier (or based on the set of abnormal fragments for a given training sample).

[0201] For example, now refer to Figure 6B , which illustrates a matrix of training feature vectors 622. In this example, the analysis system has identified CpG sites [K] 626 for generating feature vectors for a cancer classifier. The analysis system selects training samples [N] 624. The analysis system determines a first anomaly score 628 for a first arbitrary CpG site [k1], which will be used in the feature vector for training sample [n1]. The analysis system examines each abnormal fragment in the abnormal fragment set. If the analysis system identifies at least one abnormal fragment that includes the first CpG site, then the analysis system determines the first anomaly score 628 for the first CpG site to be 1, such as Figure 6B Considering a second arbitrary CpG site [k2], the analysis system similarly checks the set of abnormal fragments to identify whether there is at least one abnormal fragment that includes the second CpG site [k2]. If the analysis system does not find any such abnormal fragments that include the second CpG site, then the analysis system determines the second abnormal score 629 for the second CpG site [k2] to be 0, as shown in FIG. Figure 6B Once the analysis system determines all anomaly scores for the initial set of CpG sites, the analysis system determines a feature vector for the first training sample [n1], the feature vector comprising anomaly scores, wherein a first anomaly score 628 for the first CpG site [k1] is 1, a second anomaly score 629 for the second CpG site [k2] is 0, and subsequent anomaly scores are added to form a feature vector [1, 0, ...].

[0202] Other methods for sample characterization can be found in the following: U.S. application No. 15 / 931,022, entitled "Model-Based Featurization and Classification"; U.S. application No. 16 / 579,805, entitled "Mixture Model for Targeted Sequencing"; U.S. application No. 16 / 352,602, entitled "Anomalous Fragment Detection and Classification"; and U.S. application No. 16 / 723,716, entitled "Source of Origin Deconvolution Based on Methylation Fragments in Cell-Free DNA Samples"; all of which are incorporated by reference in their entirety.

[0203] The analysis system may further restrict the CpG sites considered for use in the cancer classifier. The analysis system calculates 630 information gain for each CpG site in the initial CpG site set based on the feature vector of the training sample. According to step 620, each training sample has a feature vector, which may include abnormal scores for all CpG sites in the initial CpG site set, and the site set may include up to all CpG sites in the human genome. However, some CpG sites in the initial CpG site set may not provide as much information as other sites when distinguishing between multiple cancer types, or the information provided may be duplicated with other CpG sites.

[0204] In one embodiment, the analysis system calculates 630 information gain for each cancer type and each CpG site in the initial set to determine whether to include the CpG site into the classifier. Information gain is calculated for training samples, which are given cancer types compared to all other training samples. For example, two random variables "abnormal fragments" ("AF") and "cancer types" ("CT") are used. In one embodiment, AF is a binary variable that indicates whether there are abnormal fragments overlapping with a given CpG site in a given sample (such as the abnormal score / feature vector determined above). CT is a random variable that indicates whether the cancer belongs to a certain specific type. The analysis system calculates the mutual information relative to CT when AF is given. That is, if it is known whether there are abnormal fragments overlapping with a specific CpG site, how much information about the cancer type can be obtained. In practice, for the first cancer type, the analysis system calculates the pairwise mutual information gain with each other cancer type and adds the mutual information gain of all other cancer types.

[0205] For a given cancer type, the analysis system can use this information to rank the CpG sites based on their specificity for cancer. This process can be repeated for all cancer types under consideration. If a particular region is typically abnormally methylated in training samples for a given cancer, but not in training samples for other cancer types or healthy training samples, then for the given cancer type, the CpG sites overlapped by those abnormal fragments can have a high information gain. The ranked CpG sites for each cancer type can be added (selected) 640 by the greedy algorithm to the selected CpG site set based on their ranking for use in the cancer classifier.

[0206] In other embodiments, the analysis system may consider other selection criteria to select informative CpG sites for use in a cancer classifier. One selection criterion may be that the distance between a selected CpG site and other selected CpG sites is greater than a threshold. For example, a selected CpG site is more than a threshold number of base pairs (e.g., 100 base pairs) away from any other selected CpG site, such that CpG sites within the threshold distance are not also selected for consideration into the cancer classifier.

[0207] In one embodiment, based on the selected set of CpG sites from the initial set, the analysis system may modify 650 the feature vectors of the training samples as needed. For example, the analysis system may truncate the feature vectors to remove anomalous scores corresponding to CpG sites that are not in the selected set of CpG sites.

[0208] Using the feature vectors of the training samples, the analysis system may train a cancer classifier in any of many ways. The feature vectors may correspond to the initial set of CpG sites in step 620, or may correspond to the selected set of CpG sites in step 650. In one embodiment, the analysis system trains 660 a binary cancer classifier based on the feature vectors of the training samples to distinguish cancer from non-cancer. In this way, the training samples used by the analysis system include non-cancer samples from healthy individuals and cancer samples from subjects. Each training sample can have one of two labels, "cancer" or "non-cancer". In this embodiment, the classifier outputs a cancer prediction, indicating the possibility of the presence or absence of cancer.

[0209] In another embodiment, the analysis system trains 670 a multi-classification cancer classifier to distinguish between multiple cancer types (also referred to as tissue of origin (TOO) tags). Cancer types may include one or more cancers, and may also include non-cancer types (and may also include any additional other diseases or genetic disorders, etc.). To this end, the analysis system may use a cancer type cohort, and may or may not include a non-cancer type cohort. In this multi-cancer embodiment, a cancer classifier is trained to determine a cancer prediction (or, more particularly, a TOO prediction), which contains a predicted value for each cancer type being classified. The predicted value may correspond to the possibility that a given training sample (during inference, corresponding to a test sample) has each cancer type. In one embodiment, the predicted value score is between 0 and 100, where the sum of the predicted values ​​is equal to 100. For example, a cancer classifier returns a cancer prediction, which includes predicted values ​​for breast cancer, lung cancer, and non-cancer. For example, a classifier may return a cancer prediction, and the predicted content is that the test sample has a 65% probability of breast cancer, a 25% probability of lung cancer, and a 10% probability of non-cancer. The analysis system may further evaluate the predictive values ​​to generate a prediction of the presence of one or more cancers in the sample, also referred to as a TOO prediction, which suggests one or more TOO tags, for example, a first TOO tag with the highest predictive value, a second TOO tag with the second highest predictive value, etc. Continuing with the example above, given these percentages, in this example, the system may determine that the sample has breast cancer because breast cancer has the highest likelihood.

[0210] In both embodiments, the analysis system trains the cancer classifier by inputting a training sample set with its feature vector into the cancer classifier and adjusting the classification parameters so that the function of the classifier can accurately associate the training feature vector with its corresponding label. The analysis system may group the training samples into one or more sets of training samples for iterative batch training of the cancer classifier. After inputting all training sample sets including its training feature vector and adjusting the classification parameters, the cancer classifier can be fully trained to label it according to the feature vector of the test sample within a certain error limit. The analysis system may train the cancer classifier according to any of many methods. For example, a two-class cancer classifier may be an L2 regularized logistic regression classifier trained using a logarithmic loss function. For another example, a multi-cancer classifier may be a multinomial logistic regression classifier. In practice, other techniques may be used to train these two types of cancer classifiers. There are many such techniques, including possible use of functional kernel methods, random forest classifiers, hybrid models, autoencoder models, machine learning algorithms (e.g., multi-layer neural networks), etc.

[0211] The classifier may include a logistic regression algorithm, a neural network algorithm, a support vector machine algorithm, a naive Bayes algorithm, a nearest neighbor algorithm, a boosted tree algorithm, a random forest algorithm, a decision tree algorithm, a multinomial logistic regression algorithm, a linear model, or a linear regression algorithm.

[0212] III.D. Deployment of Cancer Classifier

[0213] During use of the cancer classifier, the analysis system may obtain a test sample from a subject whose cancer type is unknown. The analysis system may use any combination of processes 200 and 530 to process the test sample composed of DNA molecules to obtain a set of abnormal fragments. The analysis system can determine a test feature vector for use by the cancer classifier based on similar principles discussed in process 600. The analysis system can calculate an abnormality score for each CpG site in the multiple CpG sites used by the cancer classifier. For example, the cancer classifier receives as input a feature vector including abnormality scores for 1,000 selected CpG sites. Therefore, based on the set of abnormal fragments, the analysis system can determine a test feature vector including abnormality scores for the 1,000 selected CpG sites. The analysis system can calculate the abnormality score in the same manner as the training sample. In some embodiments, the analysis system defines the abnormality score as a binary score based on whether there are hypermethylated or hypomethylated fragments in the set of abnormal fragments covering the CpG site. In some embodiments, the analysis system performs covariate prediction (for example, Figure 4B The analysis system may generate a test feature vector including one or more features predicted based on the covariate.

[0214] The analysis system can then input the test feature vector into a cancer classifier. The function of the cancer classifier can then generate a cancer prediction based on the classification parameters trained in process 600 and the test feature vector. In a first approach, the cancer prediction can be binary, selected from a group consisting of "cancer" or "non-cancer"; in a second approach, the cancer prediction is selected from a group consisting of multiple cancer types and "non-cancer". In another embodiment, the cancer prediction includes a predicted value for each of the multiple cancer types. In addition, the analysis system may determine that the test sample is most likely to belong to one of these cancer types. Continuing with the above example, for a test sample with a cancer prediction of 65% probability of breast cancer, 25% probability of lung cancer, and 10% probability of non-cancer, the analysis system may determine that the test sample is most likely to have breast cancer. In another example, where the cancer prediction is binary, 60% probability of non-cancer and 40% probability of cancer, the analysis system determines that the test sample is most likely to have no cancer. In another embodiment, to identify which type of cancer the test subject suffers from, the most likely cancer prediction may still be compared with a threshold value (e.g., 40%, 50%, 60%, 70%). If the most likely cancer prediction does not exceed that threshold value, the analysis system may return an indeterminate result.

[0215] In another embodiment, the analysis system connects the cancer classifier trained in step 660 of process 600 in series with another cancer classifier trained in step 670 of process 600. The analysis system may input the test feature vector into the cancer classifier trained as a binary classifier in step 660 of process 600. The analysis system may receive the output of the cancer prediction. The cancer prediction may be binary for whether the test subject is likely to have cancer. In other embodiments, the cancer prediction includes a prediction value describing the possibility of cancer and the possibility of non-cancer. For example, the cancer prediction has an 85% cancer prediction value and a 15% non-cancer prediction value. The analysis system may determine that the test subject is likely to have cancer. Once the analysis system determines that the test subject is likely to have cancer, the analysis system may input the test feature vector into a multi-classification cancer classifier trained to distinguish different cancer types. The multi-classification cancer classifier may receive the test feature vector and return a cancer prediction for one of the multiple cancer types. For example, the multi-classification cancer classifier provides a cancer prediction indicating that the test subject is most likely to have ovarian cancer. In another embodiment, the multi-classification cancer classifier provides a prediction value for each of the multiple cancer types. For example, a cancer prediction might include a breast cancer type prediction value of 40%, a colorectal cancer type prediction value of 15%, and a liver cancer prediction value of 45%.

[0216] According to a general embodiment of binary cancer classification, the analysis system can determine a cancer score for the test sample based on the sequencing data of the test sample (e.g., methylation sequencing data, SNP sequencing data, other DNA sequencing data, RNA sequencing data, etc.). The analysis system can compare the cancer score of the test sample with a binary classification threshold critical value that predicts whether the test sample is likely to have cancer. The binary classification threshold critical value can be adjusted using a TOO threshold based on one or more TOO subtype categories. The analysis system may further generate a feature vector for the test sample for use in a multi-classification cancer classifier to determine a cancer prediction to indicate one or more likely cancer types.

[0217] The classifier can be used to determine the disease state of a test subject (e.g., a subject whose disease state is unknown). The method can include obtaining a test genomic data construct (e.g., single time point test data) in an electronic format, the construct including a value for each of a plurality of genomic features corresponding to a plurality of nucleic acid fragments in a biological sample obtained from the test subject. The method can then include applying the test genomic data construct to a test classifier to determine the state of the disease condition of the test subject. The test subject may not have been previously diagnosed with the disease condition.

[0218] The classifier can be a temporal classifier that uses at least (i) a first test genomic data construct generated from a first biological sample obtained from a test subject at a first time point and (ii) a second test genomic data construct generated from a second biological sample obtained from the test subject at a second time point.

[0219] The trained classifier can be used to determine the disease state of a test subject (e.g., a subject whose disease state is unknown). In this case, the method may include obtaining a test time series data set for the test subject in an electronic format, wherein: for each corresponding time point in a plurality of time points, the test time series data set includes a corresponding test genotype data construct, the construct including values ​​of a plurality of genotype features of a corresponding plurality of nucleic acid fragments in a corresponding biological sample obtained from the test subject at the corresponding time point; for each pair of corresponding consecutive time points in the plurality of time points, the test time series data set includes a prompt for the length of time between the pair of corresponding consecutive time points. The method may then include applying the test genotype data construct to the test classifier to determine the state of the disease condition of the test subject. The test subject may not have been previously diagnosed with the disease condition.

[0220] IV. Application

[0221] In some embodiments, the method of the present invention, analysis system and / or classifier can be used to detect the presence of cancer, monitor cancer progression or recurrence, monitor treatment response or effectiveness, determine the presence or monitor minimal residual disease (MRD) or any combination thereof. For example, as described herein, the classifier can be used to generate a probability score (e.g., from 0 to 100), and the score describes the possibility that the test feature vector is from a subject suffering from cancer. In some embodiments, the probability score is compared with the threshold probability to determine whether the subject suffers from cancer. In other embodiments, the possibility or probability score can be evaluated at multiple different time points (e.g., before or after treatment) to monitor disease progression or to monitor treatment effectiveness (e.g., therapeutic effect). In other embodiments, the possibility or probability score can be used to make or affect clinical decisions (e.g., cancer diagnosis, treatment selection, therapeutic effect evaluation, etc.). For example, in one embodiment, if the probability score exceeds a threshold, the doctor can prescribe suitable treatment.

[0222] IV.A. Early Detection of Cancer

[0223] In some embodiments, the methods and / or classifiers of the present invention are used to detect the presence or absence of cancer in a subject suspected of having cancer. For example, a classifier (e.g., as described in Section III above and exemplified in Section V) can be used to determine a cancer prediction that describes the likelihood that a test feature vector is from a subject suffering from cancer.

[0224] In one embodiment, cancer prediction is the possibility (e.g., score between 0 and 100) of whether a test sample has cancer (i.e., binary classification). Therefore, the analysis system may determine a threshold value for determining whether a test subject suffers from cancer. For example, a cancer prediction greater than or equal to 60 can prompt a subject to suffer from cancer. In yet other embodiments, a cancer prediction greater than or equal to 65, greater than or equal to 70, greater than or equal to 75, greater than or equal to 80, greater than or equal to 85, greater than or equal to 90, or greater than or equal to 95, prompts the subject to suffer from cancer. In other embodiments, cancer prediction can prompt the severity of the disease. For example, a cancer prediction of 80 is lower than 80 (e.g., a probability score of 70) compared to a cancer prediction, which may prompt a heavier form or a later stage of cancer. Similarly, over time, an increase in cancer prediction (e.g., by classifying a test feature vector of multiple samples collected from the same subject at two or more time points) can prompt disease progression, or over time, a decrease in cancer prediction can prompt treatment success.

[0225] In another embodiment, cancer prediction includes many prediction values, wherein each of the multiple cancer types being classified (i.e., multi-classification classification) has a prediction value (e.g., a score between 0 and 100). The prediction value may correspond to the possibility of a given training sample (during inference, corresponding to the training sample) having each type of cancer. The analysis system may identify the cancer type with the highest prediction value, and prompt the test subject to be likely to suffer from the cancer type. In other embodiments, the analysis system further compares the highest prediction value with a threshold value (e.g., 50, 55, 60, 65, 70, 75, 80, 85, etc.) to determine that the test subject is likely to suffer from that type of cancer. In other embodiments, the prediction value can also indicate the severity of the disease. For example, a prediction value greater than 80 may indicate a heavier form or a later stage of cancer compared to a prediction value of 60. Similarly, over time, an increase in the prediction value (e.g., by classifying the test feature vectors of multiple samples collected from the same subject at two or more time points) can indicate disease progression, or over time, a decrease in the prediction value can indicate treatment success.

[0226] According to various aspects of the invention, the methods and systems of the invention can be trained to detect or classify multiple signs of cancer. For example, the methods, systems and classifiers of the invention can be used to detect one or more, two or more, three or more, five or more, ten or more, fifteen or more, or twenty or more different cancer types.

[0227] Examples of cancers that can be detected using the methods, systems, and classifiers of the present invention include carcinomas, lymphomas, blastomas, sarcomas, and leukemias or lymphoid malignancies. More specific examples of such cancers include, but are not limited to, squamous cell carcinomas (e.g., epithelial squamous cell carcinomas), skin cancers, melanomas, lung cancers (including small cell lung cancers, non-small cell lung cancers ("NSCLC"), lung adenocarcinomas, and lung squamous cell carcinomas), peritoneal cancers, gastric (gastric or stomach) cancers (including gastrointestinal cancers), pancreatic cancers (e.g., pancreatic ductal adenocarcinomas), cervical cancers, ovarian cancers (e.g., high-grade serous ovarian cancers), liver cancers (e.g., hepatocellular carcinoma (HCC)), hepatomas, liver tumors, bladder cancers (e.g., urothelial bladder cancers), testicular cancers (e.g., ovarian cancers), ovarian ... and germ cell tumors), breast cancer (e.g., HER2-positive, HER2-negative, and triple-negative breast cancer), brain cancer (e.g., astrocytoma, glioma (e.g., glioblastoma)), colon cancer, rectal cancer, colorectal cancer, endometrial cancer or uterine cancer, salivary gland cancer, kidney (kidney or renal) cancer (e.g., renal cell carcinoma, nephroblastoma or Wilms' tumor)), prostate cancer, vulvar cancer, thyroid cancer, anal canal cancer, penile cancer, head and neck cancer, esophageal cancer, and nasopharyngeal carcinoma (NPC). Additional examples of cancer include, but are not limited to, retinoblastoma, thecoma, ovarian testicular blastoma, hematologic malignancies including, but not limited to, non-Hodgkin's lymphoma (NHL), multiple myeloma and acute hematologic malignancies, endometriosis, fibrosarcoma, choriocarcinoma, laryngeal cancer, Kaposi's sarcoma, schwannoma, oligodendroglioma, neuroblastoma, rhabdomyosarcoma, osteogenic sarcoma, leiomyosarcoma, and urinary tract cancer.

[0228] In some embodiments, the cancer is one or more of anorectal cancer, bladder cancer, breast cancer, cervical cancer, colorectal cancer, esophageal cancer, gastric cancer, head and neck cancer, hepatobiliary cancer, leukemia, lung cancer, lymphoma, melanoma, multiple myeloma, ovarian cancer, pancreatic cancer, prostate cancer, kidney cancer, thyroid cancer, uterine cancer, or any combination thereof.

[0229] In some embodiments, one or more cancers may be "high signal" cancers (defined as cancers with a 5-year cancer-specific mortality greater than 50%), such as anorectal cancer, colorectal cancer, esophageal cancer, head and neck cancer, hepatobiliary cancer, lung cancer, ovarian cancer, and pancreatic cancer, as well as lymphoma and multiple myeloma. High signal cancers tend to be more aggressive and typically have above-average free nucleic acid concentrations in test samples obtained from patients.

[0230] IV.B. Cancer and Treatment Monitoring

[0231] In some embodiments, cancer prediction can be evaluated at multiple different time points (e.g., before or after treatment) to monitor disease progression or to monitor treatment effectiveness (e.g., treatment efficacy). For example, the present invention includes methods involving obtaining a first sample (e.g., a first plasma cfDNA sample) from a cancer patient at a first time point, determining a first cancer prediction (as described herein) therefrom, obtaining a second test sample (e.g., a second plasma cfDNA sample) from the cancer patient at a second time point, determining a second cancer prediction (as described herein) therefrom.

[0232] In certain embodiments, the first time point is before cancer treatment (e.g., before resection surgery or therapeutic intervention), the second time point is after cancer treatment (e.g., after resection surgery or therapeutic intervention), and the classifier is used to monitor the effectiveness of treatment. For example, if the second cancer prediction decreases compared to the first cancer prediction, the treatment is considered to be successful. However, if the second cancer prediction increases compared to the first cancer prediction, the treatment is considered unsuccessful. In other embodiments, the first and second time points are both before cancer treatment (e.g., before resection surgery or therapeutic intervention). In yet other embodiments, the first and second time points are both after cancer treatment (e.g., after resection surgery or therapeutic intervention). In yet other embodiments, cfDNA samples may be obtained and analyzed from cancer patients at the first and second time points, for example, to monitor cancer progression, to determine whether the cancer is in remission (e.g., after treatment), to monitor or detect residual disease or disease recurrence, or to monitor treatment (e.g., therapeutic) effects.

[0233] It will be readily understood by those skilled in the art that a test sample can be obtained from a cancer patient at any desired set of time points and analyzed according to the methods of the present invention to monitor the cancer status of the patient. In some embodiments, the time interval between the first time point and the second time point ranges from about 15 minutes to about 30 years, such as about 30 minutes, such as about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 or about 24 hours, such as about 1, 2, 3, 4, 5, 10, 15, 20, 25 or about 50 days, or such as about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 or 12 months, or such as about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 or 12 months. 5, 4, 4.5, 5, 5.5, 6, 6.5, 7, 7.5, 8, 8.5, 9, 9.5, 10, 10.5, 11, 11.5, 12, 12.5, 13, 13.5, 14, 14.5, 15, 15.5, 16, 16.5, 17, 17.5, 18, 18.5, 19, 19.5, 20, 20.5, 21, 21.5, 22, 22.5, 23, 23.5, 24, 24.5, 25, 25.5, 26, 26.5, 27, 27.5, 28, 28.5, 29, 29.5 or about 30 years. In other embodiments, the frequency of obtaining test samples from patients may be: at least once every 5 months, at least once every 6 months, at least once a year, at least once every 2 years, at least once every 3 years, at least once every 4 years, or at least once every 5 years.

[0234] IV.C. Treatment

[0235] In yet another embodiment, cancer prediction can be used to make or influence clinical decisions (e.g., cancer diagnosis, treatment selection, treatment effect evaluation, etc.). For example, in one embodiment, if the cancer prediction (e.g., for cancer or for a specific cancer type) exceeds a threshold, a doctor can prescribe appropriate treatment (e.g., resection surgery, radiotherapy, chemotherapy, and / or immunotherapy).

[0236] Classifiers (as described herein) can be used to determine that the sample feature vector is from a cancer prediction of a subject suffering from cancer. In one embodiment, when the cancer prediction exceeds a threshold, a suitable treatment (e.g., resection surgery or therapeutic intervention) is prescribed. For example, in one embodiment, if the cancer prediction is greater than or equal to 60, one or more suitable treatments are prescribed. In another embodiment, if the cancer prediction is greater than or equal to 65, greater than or equal to 70, greater than or equal to 75, greater than or equal to 80, greater than or equal to 85, greater than or equal to 90 or greater than or equal to 95, one or more suitable treatments are prescribed. In other embodiments, the cancer prediction can indicate the severity of the disease. Then a suitable treatment that matches the severity of the disease may be prescribed.

[0237] In some embodiments, treatment is one or more cancer therapeutic agents selected from the group consisting of: chemotherapeutic agents, targeted cancer therapeutic agents, differentiation therapeutic agents, hormone therapeutic agents and immunotherapeutic agents. For example, treatment can be one or more chemotherapeutic agents selected from the group consisting of: alkylating agents, antimetabolites, anthracyclines, antitumor antibiotics, cell scaffold disruptors (taxanes), topoisomerase inhibitors, nuclear division blocking substances, corticosteroids, kinase inhibitors, nucleotide analogs, platinum-based agents and any combination thereof. In some embodiments, treatment is one or more targeted cancer therapeutic agents selected from the group consisting of: signal transduction inhibitors (e.g., tyrosine kinase and growth factor receptor inhibitors), histone deacetylase (HDAC) inhibitors, retinoic acid receptor agonists, proteosome inhibitors, angiogenesis inhibitors and monoclonal antibody conjugates. In some embodiments, treatment is one or more differentiation therapeutic agents including retinoids, such as retinoic acid, alitretinoin and bexarotene. In some embodiments, treatment is one or more hormonal therapeutic agents selected from the group consisting of antiestrogens, aromatase inhibitors, progestins, estrogens, antiandrogens, and GnRH agonists or analogs. In one embodiment, treatment is one or more immunotherapeutics selected from the group comprising: monoclonal antibody therapy (e.g., rituximab (RITUXAN (Rituxan)) and alemtuzumab (CAMPATH)), nonspecific immunotherapy and adjuvants (e.g., BCG, interleukin-2 (IL-2) and interferon-α), immunomodulatory drugs (e.g., thalidomide and lenalidomide (REVLIMID (RevliMID))). An experienced physician or oncologist can select an appropriate cancer therapeutic based on a variety of features, such as tumor type, cancer stage, previously exposed cancer treatments or therapeutic agents, and other features of the cancer.

[0238] V. Example Results

[0239] VA sample collection and processing

[0240] Study design and samples: CCGA (NCT02889978) is a prospective, multicenter, case-control, longitudinal follow-up observational study. De-identified biological samples were collected from approximately 15,000 participants at 342 research centers. The samples were divided into training sets (1,785) and test sets (1,015); samples were selected to ensure that the cancer types and non-cancers in each research center in each cohort had a predetermined distribution, and cancer and non-cancer samples were age-frequency matched by sex.

[0241] Whole-genome bisulfite sequencing: cfDNA was isolated from plasma and analyzed by whole-genome bisulfite sequencing (WGBS; depth 30×). cfDNA was extracted from two tubes of plasma (up to 10 ml) from each patient using a modified QIAamp circulating nucleic acid kit (Qiagen; Germantown, MD). Up to 75 ng of plasma cfDNA was bisulfite converted using the EZ-96 DNA methylation kit (Zymo Research, D5003). The converted cfDNA was used to prepare dual-index sequencing libraries using the Accel-NGS Methyl-Seq DNA Library Preparation Kit (Swift BioSciences; Ann Arbor, MI), and the constructed libraries were quantified using the KAPA Library Quantification Kit for the Illumina platform (Kapa Biosystems; Wilmington, MA). The four libraries were mixed with 10% PhiX v3 library (Illumina, FC-110-3001) and clustered on an Illumina NovaSeq7000S2 flow cell, followed by 150-bp paired-end sequencing (30x).

[0242] For each sample, the WGBS fragment set was reduced to a small number of fragments with abnormal methylation patterns. In addition, hypermethylated or hypomethylated cfDNA fragments were selected. cfDNA fragments with abnormal methylation patterns and hypermethylated or hypomethylated were selected, i.e., UFXM. Fragments that are frequently present in individuals without cancer or with unstable methylation are unlikely to generate highly discriminatory features for cancer status classification. Therefore, we generated a statistical model and data structure for typical fragments using an independent reference set (i.e., reference genome) of 108 non-smoking participants without cancer (age: 58±14 years, 79 [73%] women) from the CCGA study. These samples were used to train a Markov chain model (3rd order) that estimates the likelihood of a given sequence of CpG methylation states within a fragment as described in Section II.C. above. This model has been shown to be calibrated within the range of normal fragments (p-value>0.001) and is used to exclude fragments with p-values ​​≥0.001 from the Markov model because these fragments are not abnormal enough.

[0243] As described above, a further data reduction step selected only fragments that covered at least 5 CpGs and had an average methylation of either >0.9 (hypermethylated) or <0.1 (hypomethylated). This process resulted in a median (range) of 2,800 (1,500-12,000) UFXM fragments for participants without cancer in training and a median (range) of 3,000 (1,200-420,000) UFXM fragments for participants with cancer in training. Since this data reduction process only used the reference set data, this stage only had to be applied once per sample.

[0244] VB covariate prediction results

[0245] Figure 8 Age-related genomic regions according to one or more example embodiments are shown. The analysis system can train regression using methylation features determined for training samples. For the purpose of illustration, these examples use only non-cancer training samples. In this example, the analysis system calculates the two-tailed p-value of the regression slope from the t statistic for each genomic region based on trained linear regression. The lower the p-value, the lower the possibility of observing the slope, which means that the genomic region is more distinguishable in terms of the temporal age of the subject of the sample source. In the shown chart, the x-axis plots human chromosomes, and the y-axis plots the position on each chromosome. Each mark represents a region (CAR), which is included in a group of genomic regions within 500 CpG sites, and the suggestive scores of these genomic regions are higher than a certain threshold suggestive score. Each CAR indicates the lowest p-value of the genomic region clustered in the CAR. The legend on the right side of the chart is the negative logarithm of the p-value, that is, the higher the negative logarithm, the lower the p-value.

[0246] According to one or more example embodiments, Fig.9A A process is described, and the process identification can provide the feature set of the genomic region for the age information generated in the feature vector. The training sample utilized is from CCGA research. The analysis system performs glmnet loose lasso regression on the lambda grid to identify the optimal range of the genomic region, used as the feature set. The optimal genomic region set within the range of 58 to 83 genomic regions provides the lowest mean square error (plotted on the y-axis). The feature set of 83 genomic regions is used to train the age prediction model for the test set.

[0247] According to an example implementation, Fig. 9B and Fig. 9C Describes the use of Fig.9A The x-axis plots the reported chronological age (“actual”) and the y-axis plots the predicted chronological age (“predicted”) obtained from the age prediction model. Fig. 9B The result graph of age prediction for a non-cancer holdout cohort according to an example implementation is shown. The holdout cohort was not used to train regression. The holdout cohort had a total of 369 samples. The residual was calculated as the reported age minus the predicted age. It is noteworthy that the highest residuals in the non-cancer cohort are approximately -35 and +25, and the vast majority of the non-cancer cohorts have residuals between -10 and +10. Fig. 9C The result graph of age prediction for a cancer cohort according to an example implementation is shown. Since the age prediction model is fitted based on non-cancer samples, no cancer samples are used for training regression. The cancer cohort has a total of 1561 samples. Here, the highest residual is about -155. Fig. 9B The distribution of residuals is also more dispersed compared to the non-cancer cohort in the .

[0248] According to one or more example embodiments, Fig. 10A Another pipeline is described that identifies a set of features for genomic regions that can provide chronological age information. The training samples used are from a follow-up of the CCGA study, called CCGA2. Fig.9A Similarly, the analysis system performed glmnet relaxed lasso regression on the lambda grid to identify the optimal range of genomic regions to use as a feature set. The optimal set of genomic regions ranging from 31 to 57 provided the lowest mean squared error (plotted on the y-axis). The feature set of 57 genomic regions was used to train an age prediction model for the test set.

[0249] According to an example implementation, Fig. 10B and Fig. 10C Describes the use of Fig. 10AThe x-axis plots the reported age (“actual”) and the y-axis plots the predicted age (“predicted”) obtained by the age prediction model. Fig. 10B A graph showing the results of age prediction for a non-cancer holdout cohort according to an example implementation is shown. The holdout cohort was not used to train regression. The holdout cohort had a total of 466 samples. The residual was calculated as the reported age minus the predicted age. It is noteworthy that the highest residuals in the non-cancer cohort are approximately -10 and +10, and the vast majority of the non-cancer cohorts have residuals between -5 and +5. Fig. 10C The result graph of age prediction for a cancer cohort according to an example implementation is shown. Since the age prediction model is fitted based on non-cancer samples, no cancer samples are used for training regression. The cancer cohort has a total of 967 samples. Here, the highest residuals are -193 and +135 (corresponding to under-prediction and over-prediction, respectively). Fig. 10B The distribution of residuals is also more dispersed compared to the non-cancer cohort in the .

[0250] Fig.11 The distribution of the test cohort at each stage of cancer according to the example implementation is shown. The x-axis of the chart divides the known cancer status of the sample: non-cancer, cancer stage 1-4, and other miscellaneous status ("unexpected" and "missing"). The left chart covers samples predicted by the classifier to be free of cancer, i.e., negative results, including true negatives and false negatives. The right chart covers samples predicted by the classifier to have cancer, i.e., positive results, including true positives and false positives. The residual threshold (displayed as "z-score" above / below 4) is four standard deviations from the mean. Any sample with a chronological age residual higher than the threshold is marked in red, and the rest are marked in yellow. In the left chart, there are two important things to note. First, the chronological age residuals of all true non-cancer samples are not higher than the chronological age residual threshold (samples not marked in red). Second, there are many cancer samples that are false negatives, but some of those false negative samples have chronological age residuals that exceed the residual threshold. The chronological age residual threshold could have been used to identify these false negative samples as having cancer. In the right chart, there are also two important observations. First, compared with non-cancer samples in either graph, a considerable number of cancer samples are above the chronological age residual threshold. Second, in the later stages of cancer, the number of samples with chronological age residuals above the chronological age residual threshold increases. This suggests that as the cancer state accelerates, the accuracy of predicting chronological age from sample methylation features decreases.

[0251] According to an example implementation, Fig. 12A and Fig. 12B The distribution of these test cohorts across cancer types is illustrated. Fig. 12AA series of top graphs are shown representing test samples predicted as negative by the cancer classifier, ie predicted to be free of cancer. Fig. 12B A series of bottom graphs are shown, which represent test samples predicted as positive results by the cancer classifier, i.e., predicted to have cancer. Using similar chronological age residual thresholds, e.g., z-scores above or below 4. In the series of top graphs, there are at least nine false negative samples (samples known to have cancer but predicted by the cancer classifier as not having cancer) whose chronological age residuals (as calculated by the chronological age prediction model) are above the chronological age residual threshold. These samples could have been determined to have a strong likelihood of cancer.

[0252] According to an example implementation, Fig.13 A genomic region is illustrated that shows a slowdown in the chronological age of the chronological age in multiple cancer types. The samples shown are all cancer samples of different cancer types. The x-axis of each chart shows the true chronological age of the sample, and the y-axis shows the ratio of the predicted chronological age to the true age. In many cancer types, the predicted chronological age will decrease. This genomic region usually has no discrimination between different cancer types, but it will slow down the chronological age in most types of all cancer types (some types may be limited due to small sample sizes).

[0253] According to an example implementation, Fig.14A and Fig. 14B Two genomic regions are illustrated that can distinguish between blood cancer types and non-blood cancer types. The samples shown are all cancer samples of different cancer types. The x-axis of each graph shows the true chronological age of the sample, and the y-axis shows the ratio of the predicted chronological age to the true age. Fig.14A Results are presented for the first genomic region that appears to have consistently accelerated chronological age across multiple blood cancer types, whereas accelerated chronological age is less consistently observed across other cancer types, multiple non-blood cancer types. Fig. 14B Results are presented for a second genomic region that appears to consistently show slowed chronological age progression across multiple blood cancer types, whereas in other cancer types, multiple non-blood cancer types, there is little apparent slowing of chronological age progression.

[0254] Fig.15AThe feature set of the genomic region for predicting biological sex according to one or more example embodiments is shown. The training sample utilized is from CCGA2 research. The analysis system performs glmnet loose lasso regression on the lambda grid to identify the optimal range of the genomic region to be used as the feature set. The optimal genomic region set within 2 or more ranges provides the lowest mean square error (plotted on the y-axis). The feature set of 3 genomic regions is used to train the biological sex prediction model for the test set.

[0255] Fig. 15B Results from a trained biological sex prediction model according to an example embodiment are shown. Fig.15A The model was trained on three genomic regions identified in the 2019 study. The test cohort contained non-cancer samples. Overall, the biological sex prediction model had an accuracy of 99.8% and a specificity of 100%.

[0256] Fig.16A The feature set of the genomic region for predicting smoking status according to one or more example embodiments is shown. The training sample utilized is from CCGA2 research. The analysis system performs glmnet loose lasso regression on the lambda grid to identify the optimal range of the genomic region to be used as the feature set. The optimal genomic region set within 1 to 4 ranges provides the lowest mean square error (plotted on the y-axis). The feature set of 2 genomic regions is used to train the smoking status prediction model for the test set.

[0257] Fig. 16B Results from a trained smoking status prediction model according to an example embodiment are shown. Fig.16A The model was trained on 2 genomic regions identified in the 2019 study. The test cohort contained non-cancer samples. Overall, the model predicted smoking status with an accuracy of 96.2% and a specificity of 99.6%.

[0258] VI. Other Considerations

[0259] The above detailed description of the embodiments refers to the accompanying drawings, which illustrate specific embodiments of the present disclosure. Other embodiments with different structures and operations do not depart from the scope of the present disclosure. The term "present invention" and the like are used to refer to certain specific examples of the many alternative aspects or embodiments of the applicant's invention shown in this specification, and its use or non-use is not intended to limit the scope of the applicant's invention or the scope of the claims.

[0260] Embodiments of the present invention may also relate to an apparatus for performing the operations described herein. This apparatus may be specially constructed for the desired purpose and / or this apparatus may include a general-purpose computing device that is selectively activated or reconfigured by a computer program stored in a computer. Such a computer program may be stored in a non-transitory, tangible computer-readable storage medium, or in any type of medium suitable for storing electronic instructions, which may be coupled to a computer system bus. In addition, any computing system mentioned in this specification may include a single processor, or may be an architecture that uses a multi-processor design to increase computing power.

[0261] Any steps, operations or processes performed by the analysis system described herein may be performed or implemented by one or more hardware or software modules of the device alone or in combination with other computing devices. In one embodiment, the software module is implemented by a computer program product, the computer program product includes a computer readable medium, the computer readable medium includes a computer program code, and the computer program code can be executed by a computer processor to perform any or all of the steps, operations or processes described.

Claims

1. A method, characterized in that: The method comprises: Get multiple training samples, each training sample: comprising a plurality of nucleic acid fragments, each of the plurality of nucleic acid fragments having a genomic location that overlaps with at least one genomic region of the plurality of genomic regions, and Labeling with the chronological age of the individual providing the training sample; For each training sample, the multiple nucleic acid fragments are sequenced to identify the methylation pattern of each nucleic acid fragment; For each genomic region of the plurality of genomic regions, identifying, from the plurality of nucleic acid fragments, a nucleic acid fragment having a genomic location that overlaps with the genomic region, and For the genomic region, based on the chronological age of the individual providing the identified nucleic acid fragment and the methylation pattern of the identified nucleic acid fragment, calculating an indicative score representing the correlation between the chronological age and the methylation pattern; generating a feature set comprising one or more genomic regions among the plurality of genomic regions, the one or more genomic regions in the feature set having an indicative score above a threshold; and A machine learning age prediction model is trained to determine a predicted chronological age of a tested individual providing a test sample, wherein the training is performed based on methylation patterns of nucleic acid fragments in the plurality of training samples that overlap with one or more genomic regions in the feature set.

2. The method according to claim 1, characterized in that: The method further comprises: training a linear regression for each genomic region in the feature set based on methylation patterns of nucleic acid fragments overlapping each genomic region from a plurality of training samples labeled as non-cancer; Get multiple additional training samples, each additional training sample: comprising a plurality of additional nucleic acid fragments having additional genomic locations therein that overlap with at least one genomic region of the plurality of genomic regions, are labeled with the chronological age of the individuals providing the additional training samples, and labeling as non-cancer or cancer based on a previous determination of the presence of cancer in the additional training samples; sequencing the plurality of additional nucleic acid fragments to identify the methylation pattern of each additional nucleic acid fragment; For each of these multiple genomic regions: applying the linear regression to the methylation patterns of the nucleic acid fragments of the plurality of additional training samples to determine a predicted chronological age of the individuals providing the additional training samples, Calculate the age residual for each additional training sample, which is the difference between its predicted temporal age and its labeled temporal age, and comparing the age residuals of the additional training samples labeled as cancer to the age residuals of the additional training samples labeled as non-cancer; and Based on the comparison of the age residuals, a reduced feature set is generated from the feature set, wherein the reduced feature set comprises fewer genomic regions than the feature set, and the reduced feature set is used to train a machine learning age prediction model.

3. The method according to claim 1, characterized in that: The method further comprises: obtaining a test sample comprising a plurality of additional nucleic acid fragments and labeled with the chronological age of a test subject providing the test sample; For the test sample, sequencing the plurality of additional nucleic acid fragments to identify methylation patterns of the plurality of additional nucleic acid fragments; applying the trained age prediction model to determine a predicted chronological age of a test subject providing the test sample based on the methylation patterns of the additional nucleic acid fragments overlapping one or more genomic regions in the feature set; calculating an age residual, the age residual being the difference between the labeled chronological age and the predicted chronological age of the test subject; and When it is determined that the age residual is higher than the residual threshold, it is determined that the test sample has a strong possibility of the presence of cancer.

4. The method according to claim 3, characterized in that: The residual threshold is determined by: applying the trained age prediction model to a second plurality of training samples identified as non-cancer to determine a predicted age for each of the second plurality of training samples; calculating an age residual for each of the second plurality of training samples by comparing the predicted age to the labeled time-series ages of the second plurality of training samples; as well as The residual threshold is determined based on the age residuals calculated for the second plurality of training samples, wherein the age residuals calculated for the second plurality of training samples at least mostly satisfy the residual threshold.

5. The method according to claim 3, characterized in that: The method further comprises: When the test sample is determined to have a strong likelihood of the presence of cancer: screening the methylation patterns of the plurality of additional nucleic acid fragments using a p-value screening method to identify a set of abnormal methylation patterns; generating a feature vector for the test sample based on the age residual and the set of abnormal methylation patterns; and A cancer prediction for the test sample is determined by inputting the feature vector into a trained cancer classifier.

6. The method according to claim 5, characterized in that: Wherein the cancer prediction is a binary prediction between the presence and absence of cancer or another disease state.

7. The method according to claim 5, characterized in that: The cancer prediction is a multi-classification prediction between multiple cancer types.

8. The method according to claim 5, characterized in that: The cancer prediction is a multi-classification prediction between multiple disease states.

9. The method according to claim 3, characterized in that: The method further comprises: The presence of cancer in the test sample is determined using a secondary machine learning cancer classifier, which is configured to receive as input the predicted chronological age of the subject and the methylation patterns of the multiple additional nucleic acid fragments, and output a prediction of the presence of cancer in the test sample.

10. The method according to claim 9, characterized in that: Wherein the secondary machine learning cancer classifier is further configured to receive clinical information and genetic background of the subject as input and output a prediction of the presence of cancer in the test sample.

11. The method according to claim 1, characterized in that: Wherein the indicative score is the Pearson correlation.

12. The method according to claim 1, characterized in that: Wherein the indicative score is a covariance score.

13. The method of claim 1, wherein: The indicative score is determined by training a linear regression to regress the chronological age based on the methylation density of non-cancer training samples, wherein the methylation density is calculated as the percentage of nucleic acid fragments having the following genomic positions: the genomic position overlaps with a specific genomic region and is in a methylated state in the specific genomic region.

14. The method of claim 1, wherein: The machine learning age prediction model comprises multivariate regression.

15. The method according to claim 14, characterized in that: wherein the multivariate regression is penalized based on the number of one or more genomic regions in the feature set.

16. The method of claim 14, wherein: The machine learning age prediction model receives as input the methylation density corresponding to each genomic region in the feature set.

17. The method of claim 1, wherein: The number of one or more genomic regions in the feature set is selected from the range of 5-10,000.

18. The method of claim 1, wherein: Sequencing of these nucleic acid fragments includes whole genome bisulfite sequencing, i.e., WGBS.

19. The method of claim 1, wherein: The sequencing of these nucleic acid fragments comprises targeted sequencing.

20. The method of claim 1, wherein: Each of the plurality of training samples was previously determined to contain no presence of cancer.

21. The method of claim 1, wherein: Each of the plurality of training samples was previously determined to contain the presence of cancer.

22. The method of claim 1, wherein: Each of the plurality of training samples was previously determined to contain a cancer presence or a cancer absence.

23. The method of claim 1, wherein: Each of the plurality of training samples is labeled as having the presence of cancer or not having the presence of cancer, the labeling being based on a prior determination of the cancer status of the training sample.

24. A method, characterized in that: The method comprises: Get multiple training samples, each training sample: comprising a plurality of nucleic acid fragments, each of the plurality of nucleic acid fragments having a genomic location that overlaps with at least one genomic region of the plurality of genomic regions, and Labeling with characteristics of the individual providing the training sample; For each training sample, the multiple nucleic acid fragments are sequenced to identify the methylation pattern of each nucleic acid fragment; For each genomic region of the plurality of genomic regions, identifying, from the plurality of nucleic acid fragments, a nucleic acid fragment having a genomic location that overlaps with the genomic region, and For the genomic region, based on the characteristics of the individual providing the identified nucleic acid fragments and the methylation pattern of the identified nucleic acid fragments, calculating an indicative score representing the correlation between the characteristics and the methylation pattern; generating a feature set comprising one or more genomic regions among the plurality of genomic regions, the one or more genomic regions in the feature set having an indicative score above a threshold; and The machine learning feature prediction model is trained to determine a predictive feature for a tested individual providing a test sample, the training being performed based on methylation patterns of nucleic acid fragments in the plurality of training samples that overlap with one or more genomic regions in the feature set.

25. The method of claim 24, wherein: wherein the characteristic is the biological sex of the individual, and the characteristic is biological male or biological female.

26. The method of claim 24, wherein: Wherein the characteristic is the smoking status of the individual, and the characteristic is smoking or non-smoking.

27. The method of claim 24, wherein: The machine learning feature prediction model includes logistic regression using a S-type function.

28. The method of claim 24, wherein: The method further comprises: obtaining a test sample comprising a plurality of additional nucleic acid fragments and labeled with a tag indicative of a characteristic of the test sample; For the test sample, sequencing the plurality of additional nucleic acid fragments to identify a test methylation pattern of each additional nucleic acid fragment; applying the trained machine learning feature prediction model to predict a feature of the test sample based on the methylation patterns of the additional nucleic acid fragments that overlap with the feature set of genomic regions; and If the predicted signature is different from the signature of the test sample, the test sample is marked as a contaminated sample and further analysis of the test sample is stopped.

29. The method of claim 28, wherein: The method further comprises: If the predictor feature matches the marker feature: screening the methylation patterns of additional nucleic acid fragments of the test sample using a p-value screening method to identify a set of abnormal methylation patterns; generating a feature vector for the test sample based on the set of abnormal methylation patterns; and A cancer prediction for the test sample is determined by inputting the feature vector into a trained cancer classifier.

30. The method of claim 29, wherein: Wherein the cancer prediction is a binary prediction between the presence and absence of cancer or another disease state.

31. The method of claim 30, wherein: The cancer prediction is a multi-classification prediction between multiple cancer types or multiple disease states.

32. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer readable storage medium contains computer program instructions that, when executed by one or more processors, cause the one or more processors to: Get multiple training samples, each training sample: comprising a plurality of nucleic acid fragments, each of the plurality of nucleic acid fragments having a genomic location that overlaps with at least one genomic region of the plurality of genomic regions, and Labeling with the chronological age of the individual providing the training sample; For each training sample, the multiple nucleic acid fragments are sequenced to identify the methylation pattern of each nucleic acid fragment; For each genomic region of the plurality of genomic regions, identifying, from the plurality of nucleic acid fragments, a nucleic acid fragment having a genomic location that overlaps with the genomic region, and For the genomic region, based on the chronological age of the individual providing the identified nucleic acid fragments and the methylation pattern of the identified nucleic acid fragments, calculating an indicative score representing the correlation between the chronological age and the methylation pattern; generating a feature set, the feature set comprising one or more genomic regions among the plurality of genomic regions, the one or more genomic regions in the feature set having an indicative score above a threshold; as well as A machine learning age prediction model is trained to determine a predicted chronological age of a tested individual providing a test sample, wherein the training is performed based on methylation patterns of nucleic acid fragments in the plurality of training samples that overlap with one or more genomic regions in the feature set.

33. The non-transitory computer readable storage medium of claim 32, wherein: Execution of these computer program instructions may cause the one or more processors to: training a linear regression for each genomic region in the feature set based on methylation patterns of nucleic acid fragments overlapping each genomic region from a plurality of training samples labeled as non-cancer; Get multiple additional training samples, each additional training sample: comprising a plurality of additional nucleic acid fragments having additional genomic locations therein that overlap with at least one genomic region of the plurality of genomic regions, are labeled with the chronological age of the individuals providing the additional training samples, and labeling as non-cancer or cancer based on a previous determination of the presence of cancer in the additional training samples; sequencing the plurality of additional nucleic acid fragments to identify the methylation pattern of each additional nucleic acid fragment; For each of these multiple genomic regions: applying the linear regression to the methylation patterns of the nucleic acid fragments of the plurality of additional training samples to determine a predicted chronological age of the individuals providing the additional training samples, Calculate the age residual for each additional training sample, which is the difference between its predicted temporal age and its labeled temporal age, and comparing the age residuals of the additional training samples labeled as cancer to the age residuals of the additional training samples labeled as non-cancer; as well as Based on the comparison of the age residuals, a reduced feature set is generated from the feature set, wherein the reduced feature set comprises fewer genomic regions than the feature set, and the reduced feature set is used to train a machine learning age prediction model.

34. The non-transitory computer readable storage medium of claim 32, wherein: Execution of these computer program instructions may cause the one or more processors to: obtaining a test sample comprising a plurality of additional nucleic acid fragments and labeled with the chronological age of a test subject providing the test sample; For the test sample, sequencing the plurality of additional nucleic acid fragments to identify methylation patterns of the plurality of additional nucleic acid fragments; applying the trained age prediction model to determine a predicted chronological age of a test subject providing the test sample based on the methylation patterns of the additional nucleic acid fragments overlapping one or more genomic regions in the feature set; calculating an age residual, the age residual being the difference between the marked chronological age and the predicted chronological age of the test subject; as well as When it is determined that the age residual is higher than the residual threshold, it is determined that the test sample has a strong possibility of the presence of cancer.

35. The non-transitory computer readable storage medium of claim 34, wherein: The computer program instructions for determining a residual threshold, when executed, may further cause the one or more processors to: applying the trained age prediction model to a second plurality of training samples identified as non-cancer to determine a predicted age for each of the second plurality of training samples; calculating an age residual for each of the second plurality of training samples by comparing the predicted age to the labeled time-series ages of the second plurality of training samples; as well as The residual threshold is determined based on the age residuals calculated for the second plurality of training samples, wherein the age residuals calculated for the second plurality of training samples at least mostly satisfy the residual threshold.

36. The non-transitory computer readable storage medium of claim 34, wherein: Execution of these computer program instructions may cause the one or more processors to: When the test sample is determined to have a strong likelihood of the presence of cancer: screening the methylation patterns of the plurality of additional nucleic acid fragments using a p-value screening method to identify a set of abnormal methylation patterns; generating a feature vector for the test sample based on the age residual and the set of abnormal methylation patterns; as well as The cancer prediction of the test sample is determined by inputting the feature vector into a trained cancer classifier.

37. The non-transitory computer readable storage medium of claim 36, wherein: Wherein the cancer prediction is a binary prediction between the presence and absence of cancer or another disease state.

38. The non-transitory computer readable storage medium of claim 36, wherein: The cancer prediction is a multi-classification prediction between multiple cancer types.

39. The non-transitory computer readable storage medium of claim 36, wherein: The cancer prediction is a multi-classification prediction between multiple disease states.

40. The non-transitory computer readable storage medium of claim 34, wherein: Execution of these computer program instructions may cause the one or more processors to: The presence of cancer in the test sample is determined using a secondary machine learning cancer classifier, which is configured to receive as input the predicted chronological age of the subject and the methylation patterns of the multiple additional nucleic acid fragments, and output a prediction of the presence of cancer in the test sample.

41. The non-transitory computer readable storage medium of claim 34, wherein: Wherein the secondary machine learning cancer classifier is further configured to receive clinical information and genetic background of the subject as input and output a prediction of the presence of cancer in the test sample.

42. The non-transitory computer readable storage medium of claim 32, wherein: Wherein the indicative score is the Pearson correlation.

43. The non-transitory computer readable storage medium of claim 32, wherein: Wherein the indicative score is a covariance score.

44. The non-transitory computer readable storage medium of claim 32, wherein: The indicative score is determined by training a linear regression to regress the chronological age based on the methylation density of non-cancer training samples, wherein the methylation density is calculated as the percentage of nucleic acid fragments having the following genomic positions: the genomic position overlaps with a specific genomic region and is in a methylated state in the specific genomic region.

45. The non-transitory computer readable storage medium of claim 32, wherein: The machine learning age prediction model comprises multivariate regression.

46. ​​The non-transitory computer readable storage medium of claim 45, wherein: wherein the multivariate regression is penalized based on the number of one or more genomic regions in the feature set.

47. The non-transitory computer readable storage medium of claim 45, wherein: The machine learning age prediction model receives as input the methylation density corresponding to each genomic region in the feature set.

48. The non-transitory computer readable storage medium of claim 32, wherein: The number of one or more genomic regions in the feature set is selected from the range of 5-10,000.

49. The non-transitory computer readable storage medium of claim 32, wherein: Sequencing of these nucleic acid fragments includes whole genome bisulfite sequencing, i.e., WGBS.

50. The non-transitory computer readable storage medium of claim 32, wherein: The sequencing of these nucleic acid fragments comprises targeted sequencing.

51. The non-transitory computer readable storage medium of claim 32, wherein: Each of the plurality of training samples was previously determined to contain no presence of cancer.

52. The non-transitory computer readable storage medium of claim 32, wherein: Each of the plurality of training samples was previously determined to contain the presence of cancer.

53. The non-transitory computer readable storage medium of claim 32, wherein: Each of the plurality of training samples was previously determined to contain a cancer presence or a cancer absence.

54. The non-transitory computer readable storage medium of claim 32, wherein: Each of the plurality of training samples is labeled as having the presence of cancer or not having the presence of cancer, the labeling being based on a prior determination of the cancer status of the training sample.

55. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer readable storage medium contains computer program instructions that, when executed by one or more processors, cause the one or more processors to: Get multiple training samples, each training sample: comprising a plurality of nucleic acid fragments, each of the plurality of nucleic acid fragments having a genomic location that overlaps with at least one genomic region of the plurality of genomic regions, and Labeling with characteristics of the individual providing the training sample; For each training sample, the multiple nucleic acid fragments are sequenced to identify the methylation pattern of each nucleic acid fragment; For each genomic region of the plurality of genomic regions, identifying, from the plurality of nucleic acid fragments, a nucleic acid fragment having a genomic location that overlaps with the genomic region, and For the genomic region, based on the characteristics of the individual providing the identified nucleic acid fragments and the methylation pattern of the identified nucleic acid fragments, calculating an indicative score representing the correlation between the characteristics and the methylation pattern; generating a feature set, the feature set comprising one or more genomic regions among the plurality of genomic regions, the one or more genomic regions in the feature set having an indicative score above a threshold; as well as The machine learning feature prediction model is trained to determine a predictive feature for a tested individual providing a test sample, the training being performed based on methylation patterns of nucleic acid fragments in the plurality of training samples that overlap with one or more genomic regions in the feature set.

56. The non-transitory computer readable storage medium of claim 55, wherein: wherein the characteristic is the biological sex of the individual, and the characteristic is biological male or biological female.

57. The non-transitory computer readable storage medium of claim 55, wherein: Wherein the characteristic is the smoking status of the individual, and the characteristic is smoking or non-smoking.

58. The non-transitory computer readable storage medium of claim 55, wherein: The machine learning feature prediction model includes logistic regression using a S-type function.

59. The non-transitory computer readable storage medium of claim 55, wherein: The computer program instructions, when executed, may cause the one or more processors to: obtaining a test sample comprising a plurality of additional nucleic acid fragments and labeled with a tag indicative of a characteristic of the test sample; For the test sample, sequencing the plurality of additional nucleic acid fragments to identify a test methylation pattern of each additional nucleic acid fragment; applying the trained machine learning feature prediction model to predict a feature of the test sample based on the methylation patterns of the additional nucleic acid fragments that overlap with the feature set of genomic regions; as well as If the predicted signature is different from the signature of the test sample, the test sample is marked as a contaminated sample and further analysis of the test sample is stopped.

60. The non-transitory computer readable storage medium of claim 59, wherein: The computer program instructions, when executed, may cause the one or more processors to: If the predictor feature matches the marker feature: screening the methylation patterns of additional nucleic acid fragments of the test sample using a p-value screening method to identify a set of abnormal methylation patterns; generating a feature vector for the test sample based on the set of abnormal methylation patterns; as well as A cancer prediction for the test sample is determined by inputting the feature vector into a trained cancer classifier.

61. The non-transitory computer readable storage medium of claim 60, wherein: Wherein the cancer prediction is a binary prediction between the presence and absence of cancer or another disease state.

62. The non-transitory computer readable storage medium of claim 61, wherein: The cancer prediction is a multi-classification prediction between multiple cancer types or multiple disease states.

63. A system, characterized in that: It contains: one or more processors; A non-transitory computer-readable storage medium storing computer program instructions which, when executed by the one or more processors, cause the one or more processors to: Get multiple training samples, each training sample: comprising a plurality of nucleic acid fragments, each of the plurality of nucleic acid fragments having a genomic location that overlaps with at least one genomic region of the plurality of genomic regions, and Labeling with the chronological age of the individual providing the training sample; For each training sample, the multiple nucleic acid fragments are sequenced to identify the methylation pattern of each nucleic acid fragment; For each genomic region of the plurality of genomic regions, identifying, from the plurality of nucleic acid fragments, a nucleic acid fragment having a genomic location that overlaps with the genomic region, and For the genomic region, based on the chronological age of the individual providing the identified nucleic acid fragments and the methylation pattern of the identified nucleic acid fragments, calculating an indicative score representing the correlation between the chronological age and the methylation pattern; generating a feature set, the feature set comprising one or more genomic regions among the plurality of genomic regions, the one or more genomic regions in the feature set having an indicative score above a threshold; as well as A machine learning age prediction model is trained to determine a predicted chronological age of a tested individual providing a test sample, wherein the training is performed based on methylation patterns of nucleic acid fragments in the plurality of training samples that overlap with one or more genomic regions in the feature set.

64. The system of claim 63, wherein: Execution of these computer program instructions may cause the one or more processors to: training a linear regression for each genomic region in the feature set based on methylation patterns of nucleic acid fragments overlapping each genomic region from a plurality of training samples labeled as non-cancer; Get multiple additional training samples, each additional training sample: comprising a plurality of additional nucleic acid fragments having additional genomic locations therein that overlap with at least one genomic region of the plurality of genomic regions, are labeled with the chronological age of the individuals providing the additional training samples, and labeling as non-cancer or cancer based on a previous determination of the presence of cancer in the additional training samples; sequencing the plurality of additional nucleic acid fragments to identify the methylation pattern of each additional nucleic acid fragment; For each of these multiple genomic regions: applying the linear regression to the methylation patterns of the nucleic acid fragments of the plurality of additional training samples to determine a predicted chronological age of the individuals providing the additional training samples, Calculate the age residual for each additional training sample, which is the difference between its predicted temporal age and its labeled temporal age, and comparing the age residuals of the additional training samples labeled as cancer to the age residuals of the additional training samples labeled as non-cancer; as well as Based on the comparison of the age residuals, a reduced feature set is generated from the feature set, wherein the reduced feature set comprises fewer genomic regions than the feature set, and the reduced feature set is used to train a machine learning age prediction model.

65. The system of claim 63, wherein: Execution of these computer program instructions may cause the one or more processors to: obtaining a test sample comprising a plurality of additional nucleic acid fragments and labeled with the chronological age of a test subject providing the test sample; For the test sample, sequencing the plurality of additional nucleic acid fragments to identify methylation patterns of the plurality of additional nucleic acid fragments; applying the trained age prediction model to determine a predicted chronological age of a test subject providing the test sample based on the methylation patterns of the additional nucleic acid fragments overlapping one or more genomic regions in the feature set; calculating an age residual, the age residual being the difference between the marked chronological age and the predicted chronological age of the test subject; as well as When it is determined that the age residual is higher than the residual threshold, it is determined that the test sample has a strong possibility of the presence of cancer.

66. The system of claim 65, wherein: The computer program instructions for determining a residual threshold, when executed, may further cause the one or more processors to: applying the trained age prediction model to a second plurality of training samples identified as non-cancer to determine a predicted age for each of the second plurality of training samples; calculating an age residual for each of the second plurality of training samples by comparing the predicted age to the labeled time-series ages of the second plurality of training samples; as well as The residual threshold is determined based on the age residuals calculated for the second plurality of training samples, wherein the age residuals calculated for the second plurality of training samples at least mostly satisfy the residual threshold.

67. The system of claim 65, wherein: Execution of these computer program instructions may cause the one or more processors to: When the test sample is determined to have a strong likelihood of the presence of cancer: screening the methylation patterns of the plurality of additional nucleic acid fragments using a p-value screening method to identify a set of abnormal methylation patterns; generating a feature vector for the test sample based on the age residual and the set of abnormal methylation patterns; as well as The cancer prediction of the test sample is determined by inputting the feature vector into a trained cancer classifier.

68. The system of claim 67, wherein: Wherein the cancer prediction is a binary prediction between the presence and absence of cancer or another disease state.

69. The system of claim 67, wherein: The cancer prediction is a multi-classification prediction between multiple cancer types.

70. The system of claim 67, wherein: The cancer prediction is a multi-classification prediction between multiple disease states.

71. The system of claim 65, wherein: Execution of these computer program instructions may cause the one or more processors to: The presence of cancer in the test sample is determined using a secondary machine learning cancer classifier, which is configured to receive as input the predicted chronological age of the subject and the methylation patterns of the multiple additional nucleic acid fragments, and output a prediction of the presence of cancer in the test sample.

72. The system of claim 65, wherein: Wherein the secondary machine learning cancer classifier is further configured to receive clinical information and genetic background of the subject as input and output a prediction of the presence of cancer in the test sample.

73. The system of claim 63, wherein: Wherein the indicative score is the Pearson correlation.

74. The system of claim 63, wherein: Wherein the indicative score is a covariance score.

75. The system of claim 63, wherein: The indicative score is determined by training a linear regression to regress the chronological age based on the methylation density of non-cancer training samples, wherein the methylation density is calculated as the percentage of nucleic acid fragments having the following genomic positions: the genomic position overlaps with a specific genomic region and is in a methylated state in the specific genomic region.

76. The system of claim 63, wherein: The machine learning age prediction model comprises multivariate regression.

77. The system of claim 76, wherein: wherein the multivariate regression is penalized based on the number of one or more genomic regions in the feature set.

78. The system of claim 76, wherein: The machine learning age prediction model receives as input the methylation density corresponding to each genomic region in the feature set.

79. The system of claim 63, wherein: The number of one or more genomic regions in the feature set is selected from the range of 5-10,000.

80. The system of claim 63, wherein: Sequencing of these nucleic acid fragments includes whole genome bisulfite sequencing, i.e., WGBS.

81. The system of claim 63, wherein: The sequencing of these nucleic acid fragments comprises targeted sequencing.

82. The system of claim 63, wherein: Each of the plurality of training samples was previously determined to contain no presence of cancer.

83. The system of claim 63, wherein: Each of the plurality of training samples was previously determined to contain the presence of cancer.

84. The system of claim 63, wherein: Each of the plurality of training samples was previously determined to contain a cancer presence or a cancer absence.

85. The system of claim 63, wherein: Each of the plurality of training samples is labeled as having the presence of cancer or not having the presence of cancer, the labeling being based on a prior determination of the cancer status of the training sample.

86. A system, characterized in that: The system includes computer program instructions that, when executed by one or more processors, cause the one or more processors to: Get multiple training samples, each training sample: comprising a plurality of nucleic acid fragments, each of the plurality of nucleic acid fragments having a genomic location that overlaps with at least one genomic region of the plurality of genomic regions, and Labeling with characteristics of the individual providing the training sample; For each training sample, the multiple nucleic acid fragments are sequenced to identify the methylation pattern of each nucleic acid fragment; For each genomic region of the plurality of genomic regions, identifying, from the plurality of nucleic acid fragments, a nucleic acid fragment having a genomic location that overlaps with the genomic region, and For the genomic region, based on the characteristics of the individual providing the identified nucleic acid fragments and the methylation pattern of the identified nucleic acid fragments, calculating an indicative score representing the correlation between the characteristics and the methylation pattern; generating a feature set, the feature set comprising one or more genomic regions among the plurality of genomic regions, the one or more genomic regions in the feature set having an indicative score above a threshold; as well as The machine learning feature prediction model is trained to determine a predictive feature for a tested individual providing a test sample, the training being performed based on methylation patterns of nucleic acid fragments in the plurality of training samples that overlap with one or more genomic regions in the feature set.

87. The system of claim 86, wherein: wherein the characteristic is the biological sex of the individual, and the characteristic is biological male or biological female.

88. The system of claim 86, wherein: Wherein the characteristic is the smoking status of the individual, and the characteristic is smoking or non-smoking.

89. The system of claim 86, wherein: The machine learning feature prediction model includes logistic regression using a S-type function.

90. The system of claim 86, wherein: The computer program instructions, when executed, may cause the one or more processors to: obtaining a test sample comprising a plurality of additional nucleic acid fragments and labeled with a tag indicative of a characteristic of the test sample; For the test sample, sequencing the plurality of additional nucleic acid fragments to identify a test methylation pattern of each additional nucleic acid fragment; applying the trained machine learning feature prediction model to predict a feature of the test sample based on the methylation patterns of the additional nucleic acid fragments that overlap with the feature set of genomic regions; as well as If the predicted signature is different from the signature of the test sample, the test sample is marked as a contaminated sample and further analysis of the test sample is stopped.

91. The system of claim 90, wherein: The computer program instructions, when executed, may cause the one or more processors to: If the predictor feature matches the marker feature: screening the methylation patterns of additional nucleic acid fragments of the test sample using a p-value screening method to identify a set of abnormal methylation patterns; generating a feature vector for the test sample based on the set of abnormal methylation patterns; as well as A cancer prediction for the test sample is determined by inputting the feature vector into a trained cancer classifier.

92. The system of claim 91, wherein: Wherein the cancer prediction is a binary prediction between the presence and absence of cancer or another disease state.

93. The system of claim 92, wherein: The cancer prediction is a multi-classification prediction between multiple cancer types or multiple disease states.

Images