Augmented training of a cell-free DNA cancer classifier with white blood cell DNA

Abstract

In some aspects, the techniques described herein relate to a method for augmentation of a cell-free (cf) DNA cancer classifier with white blood cell (WBC) DNA, the method including: obtaining cfDNA samples and WBC DNA samples, with each sample including sequence reads and a label of an originating subject's disease state; identifying a first set of cfDNA derived features from the sequence reads of the cfDNA samples and a second set of WBC DNA derived features from the sequence reads of the WBC DNA samples; featurizing each cfDNA sample with the first set of cfDNA derived features and the second set of WBC DNA derived features yielding a concatenated feature vector; optionally featurizing each WBC DNA sample with the first set of cfDNA derived features and the second set of WBC DNA derived features yielding a concatenated feature vector; training a cancer classifier with the concatenated feature vectors for the cfDNA samples and optionally with the concatenated feature vectors for the WBC DNA samples, the cancer classifier trained to output a prediction on disease state for a test cfDNA sample.

Description

AUGMENTED TRAINING OF A CELL-FREE DNA CANCER CLASSIFIER WITH WHITE BLOOD CELL DNAInventors:Alvin ShiJing XiangFrank ChuOliver Claude VennCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The present application claims the benefit of and priority’ to U.S. Provisional Application No. 63 / 755,117 filed on February 6, 2025, which is incorporated by reference.BACKGROUND

[0002] Cancer is a leading cause of death worldwide. The fatality' of cancer is heightened by the fact that cancer is usually detected in later stages, limiting efficacy of treatment options for long-term survival. Current detection methods generally are specific to a particular type of cancer, i.e., each cancer is separately screened for. Accordingly, each subject screening process is tailored to the ty pe. For example, mammography scans are utilized in breast cancer detection, whereas colonoscopy or fecal tests have helped with colorectal cancer detection. Each varied screening method is generally not cross-applicable to other types. Furthermore, present screening methods are encumbered by low detection rates or high false positive rates. Low detection rates often fail to detect early-stage cancers as the cancers are just developing. A high false positive rate misdiagnoses cancer-free subjects as positive for cancer status. As a result, most screening tests are only practical when they are used to test subjects who have a high risk of developing the screened cancer or have symptoms indicative of the presence of suspected cancers. As such, most screening tests have limited ability' to detect cancers in the general population.

[0003] Novel research has implicated aberrant DNA methylation in many disease processes, including cancer. DNA methylation plays a role in regulating gene expression and defining tissue differentiation, cellular identity', and / or embryological lineage. Thus, aberrant DNA methylation can create issues in normal gene expression pathways or cellular identity', thereby leading to cancer or other diseases. For example, specific patterns of differentiallyAtty. Docket No. 32917-65573 / WO 1 P0252-WOmethylated regions may be useful as molecular markers for various disease states. Detection of these differentially methylated regions may be accomplished through sequencing analysis of cell-free DNA molecules. In general, cell-free DNA molecules are DNA molecules that arise in bodily fluids. These DNA molecules are typically released due to natural cell death, active release by healthy cells, or tumor-derived DNA molecules shed from tumor cells undergoing cell death. Nonetheless, tests to detect differentially methylated regions face a number of challenges. Early cancer detection is particularly challenging due to the miniscule ratio of tumor cells to non-cancer cells in the subject. The miniscule ratio may be on the order of 1:1000, 1:10,000, or even 1:100,000. This creates a challenge of detecting small amounts of cancer signal amidst healthy signal, especially when analyzing this signal with an easily accessible sample condition, for example a blood draw to assess presence of cancer signal in blood plasma, for example in cell-free DNA.

[0004] One particular challenge in cfDNA-based cancer classification arises in trying to detect solid cancer signal apart from confounding hematological cancer (or hematological disease) signal.

[0005] The present disclosure is directed to addressing the above-referenced challenges. The background description provided herein is for the purpose of generally presenting the context of the disclosure. Unless otherwise indicated herein, the materials described in this section are not prior art to the claims in this application and are not admitted to be prior art, or suggestions of the prior art, by inclusion in this section.SUMMARY

[0006] The invention(s) described herein this disclosure provide for improvements to cancer detection, diagnosis and treatment. In particular, white blood cell (WBC) DNA samples are used to augment a cancer classifier for predicting a cancer prediction for a cell-free (cf) DNA sample. Augmenting the cancer classifier with the WBC DNA samples can aid in excluding the confounding signals from hematological diseases and / or disorders, thereby yielding improvements to predictive sensitivity in detecting solid cancers and / or hematological cancers.

[0007] Clause 1. A method for augmentation of a cell-free (cf) DNA cancer classifier with white blood cell (WBC) DNA, the method comprising: accessing cfDNA samples and WBC DNA samples, with each sample comprising sequence reads and a label of an originating subject's disease state; identifying a first set of cfDNA derived features from theAtty. Docket No. 32917-65573 / WO 2 P0252-WOsequence reads of the cfDNA samples and a second set of WBC DNA derived features from the sequence reads of the WBC DNA samples; featurizing each cfDNA sample based on the first set of cfDNA derived features and the second set of WBC DNA derived features yielding a feature vector; training a cancer classifier with the feature vectors for the cfDNA samples, the cancer classifier trained to output a prediction on disease state for a test cfDNA sample.

[0008] Clause 2. The method of clause 1, wherein accessing the cfDNA samples and the WBC DNA samples comprises: accessing a blood sample drawn from a subject; isolating the cfDNA sample and the WBC DNA sample from the blood sample; and performing sequencing of nucleic acid fragments in the cfDNA sample to yield the sequence reads for the cfDNA sample and sequencing of nucleic acid fragments in the WBC DNA sample to yield the sequence reads for the WBC DNA sample.

[0009] Clause 3. The method of clause 1, wherein the sequence reads for each sample comprise methylation sequence reads obtained from targeted methylation sequencing of a plurality of CpG sites.

[0010] Clause 4. The method of clause 1, wherein identifying the first set of cfDNA derived features comprises: computing information gain for each feature of an initial set of features in discrimination of cfDNA samples having a positive label for the disease state and cfDNA samples having a negative label for the disease state; and selecting the cfDNA derived features based on the information gain computed with the cfDNA samples.

[0011] Clause 5. The method of clause 1, wherein identifying the second set of WBC DNA derived features comprises: computing information gain for each feature of the initial set of features in discrimination of WBC DNA samples having the positive label for the disease state and WBC DNA samples having the negative label for the disease state; and selecting the WBC DNA derived features based on the information gain computed with the WBC DNA samples.

[0012] Clause 6. The method of clause 1, further comprising: featurizing each WBC DNA sample with the first set of cfDNA derived features and the second set of WBC DNA derived features yielding a feature vector; wherein training the cancer classifier comprises training with the feature vectors for the WBC DNA samples.

[0013] Clause 7. The method of clause 1, further comprising: identify ing an aggregate set of features including one or more features from the first set of cfDNA derived features and one or more features from the second set of cfDNA derived features, wherein the aggregate set of features excludes one or more features from either the first set of cfDNAAtty. Docket No. 32917-65573 / WO 3 P0252-WOderived features or the second set of WBC DNA derived features, wherein featurizing each cfDNA sample yielding the feature vector comprises featurizing each cfDNA sample based on the aggregate set of features.

[0014] Clause 8. The method of clause 1, further comprising: identifying an aggregate set of features from the first set of cfDNA derived features and the second set of WBC DNA derived features by: identifying one or more features overlapping the first set and the second set, and identifying the aggregate set of features by removing the overlapping features from the first set of cfDNA derived features, wherein featurizing each cfDNA sample yielding the feature vector comprises featurizing each cfDNA sample based on the aggregate set of features.

[0015] Clause 9. The method of clause 1, wherein training the cancer classifier comprises performing supervised machine-learning training to predict the label of the disease state for each cfDNA sample based on the feature vector of the cfDNA sample.

[0016] Clause 10. The method of clause 1, wherein the samples comprise a label of one or more cancer types from a plurality of cancer types, wherein training the cancer classifier comprises training the cancer classifier to output a multiclass cancer prediction.

[0017] Clause 11. A method for augmentation of a cell-free (cf) DNA cancer classifier with white blood cell (WBC) DNA, the method comprising: accessing cell-free (cf) DNA samples and white blood cell (WBC) DNA samples, with each sample comprising sequence reads and a label of an originating subject's disease state; identifying a first set of cfDNA derived features from the sequence reads of the cfDNA samples and a second set of WBC DNA derived features from the sequence reads of the WBC DNA samples; featurizing each cfDNA sample with the first set of cfDNA derived features yielding a cfDNA featurespecific feature vector; training a cfDNA feature classifier with the cfDNA feature-specific feature vectors for the cfDNA samples; featurizing each cfDNA sample with the second set of WBC DNA derived features yielding a WBC DNA feature-specific feature vector; training a WBC DNA feature classifier with the WBC DNA feature-specific feature vectors for the cfDNA samples; and training a logistic layer configured to input predictions by the cfDNA feature classifier and the WBC DNA feature classifier to yield an aggregate prediction on disease state for a test cfDNA sample.

[0018] Clause 12. The method of clause 11, wherein accessing the cfDNA samples and the WBC DNA samples comprises: accessing a blood sample drawn from a subject; isolating the cfDNA sample and the WBC DNA sample from the blood sample; and performing sequencing of nucleic acid fragments in the cfDNA sample to yield the sequenceAtty. Docket No. 32917-65573 / WO 4 P0252-WOreads for the cfDNA sample and sequencing of nucleic acid fragments in the WBC DNA sample to yield the sequence reads for the WBC DNA sample.

[0019] Clause 13. The method of clause 11, wherein the sequence reads for each sample comprise methylation sequence reads obtained from targeted methylation sequencing of a plurality of CpG sites.

[0020] Clause 14. The method of clause 11, wherein identifying the first set of cfDNA derived features comprises: computing information gam for each feature of an initial set of features in discrimination of cfDNA samples having a positive label for the disease state and cfDNA samples having a negative label for the disease state; and selecting the cfDNA derived features based on the information gain computed with the cfDNA samples.

[0021] Clause 15. The method of clause 11, wherein identifying the second set of WBC DNA derived features comprises: computing information gain for each feature of the initial set of features in discrimination of WBC DNA samples having the positive label for the disease state and WBC DNA samples having the negative label for the disease state; and selecting the WBC DNA derived features based on the information gain computed with the WBC DNA samples.

[0022] Clause 16. The method of clause 11, further comprising: featurizing each WBC DNA sample w ith the first set of cfDNA derived features yielding a cfDNA featurespecific feature vector, wherein training the cfDNA feature classifier comprises training with the cfDNA feature-specific feature vectors for the WBC DNA samples.

[0023] Clause 17. The method of clause 1 1, further comprising: featurizing each WBC DNA sample with the second set of WBC DNA derived features yielding a WBC DNA feature-specific feature vector; wherein training the WBC DNA feature classifier comprises training with the WBC DNA feature-specific feature vectors for the WBC DNA samples.

[0024] Clause 18. The method of clause 11, wherein training the logistic layer comprises performing supervised machine-learning training to predict the label of the disease state for each sample based on the predictions by the cfDNA feature classifier and the WBC DNA feature classifier for the cfDNA sample.

[0025] Clause 19. The method of clause 11. wherein training the cfDNA feature classifier, the WBC DNA feature classifier, or the logistic layer comprises training on concatenated feature vectors from each paired cfDNA sample and WBC DNA sample.

[0026] Clause 20. The method of clause 11, wherein the samples each comprise a label of one or more cancer types from a plurality of cancer types, wherein training the logistic layer comprises training the cancer classifier to output a multiclass cancer prediction.Atty. Docket No. 32917-65573 / WO 5 P0252-WO

[0027] Clause 21. A non-transitory computer-readable storage medium storing instructions for execution of a cancer classifier trained by the method of any one of clauses 1-20.

[0028] Clause 22. A method for prediction of a disease state for a subject, the method comprising: accessing a test sample comprising sequence reads; featurizing the test sample based on the first set of cfDNA derived features and the second set of WBC DNA derived features yielding a feature vector for the test sample; applying the cancer classifier to the feature vector for the test sample to output a prediction on disease state for the test sample, wherein the cancer classifier is trained by the method of any one of clauses 1-20.

[0029] Clause 23. The method of clause 22, wherein the test sample comprises sequence reads derived from sequencing cell-free DNA (cfDNA) fragments and sequence reads derived from sequencing white blood cell (WBC) DNA fragments.

[0030] Clause 24. A method for prediction of a disease state for a subject, the method comprising: accessing a cfDNA sample comprising sequence reads derived from sequencing cell-free DNA (cfDNA) fragments; featurizing the cfDNA sample based on a first set of cfDNA derived features yielding a first feature vector; applying a first cancer classifier to the first feature vector for the test sample to output a prediction on disease state for the test sample; responsive to detection of positive signal in the disease state, accessing a white blood cell (WBC) DNA sample from a reflex assay, the WBC DNA sample comprising sequence reads derived from sequencing WBC DNA fragments; featurizing the WBC DNA sample based on a second set of WBC DNA derived features yielding a second feature vector; and applying a second cancer classifier to the second feature vector, and optionally the first feature vector, to output a second prediction on disease state for the test sample, wherein the second cancer classifier is trained by the method of any one of clauses 1-20.BRIEF DESCRIPTION OF DRAWINGS

[0031] FIG. 1 is an exemplary flowchart describing an overall workflow of cancer classification of a sample, according to one or more embodiments.

[0032] FIG. 2A is an exemplary flowchart describing a process of sequencing a fragment of cell-free (cf DNA to obtain a methylation state vector, according to one or more embodiments.Atty. Docket No. 32917-65573 / WO 6 P0252-WO

[0033] FIG. 2B is an exemplary illustration of the process of FIG. 2A of sequencing a fragment of cell-free (cf) DNA to obtain a methylation state vector, according to one or more embodiments.

[0034] FIG. 3A is an exemplary flowchart describing a process of generating a control group data structure for determining anomalously methylated fragments, according to one or more embodiments.

[0035] FIG. 3B is an exemplary flowchart describing a process of determining a fragment to be anomalously methylated based on the control group data structure, according to one or more embodiments.

[0036] FIG. 4A is an exemplary flowchart describing a process of training a cancer classifier, according to one or more embodiments.

[0037] FIG. 4B illustrates an example generation of feature vectors used for training the cancer classifier, according to one or more embodiments.

[0038] FIG. 5A illustrates an exemplary flowchart of devices for sequencing nucleic acid samples according to one or more embodiments.

[0039] FIG. 5B is an exemplary block diagram of an analytics system, according to one or more embodiments.

[0040] FIG. 6 illustrates a flowchart describing a process of augmenting a cancer classifier with features derived from WBC DNA samples, according to one or more embodiments.

[0041] FIG. 7 illustrates the augmentation of the cancer classifier with features derived from WBC DNA samples, according to one or more embodiments.

[0042] FIG. 8 illustrates a flow chart describing a process of architecting a late fusion cancer classifier, according to one or more embodiments.

[0043] FIG. 9 illustrates the late fusion cancer classifier with separately-trained feature-specific classifiers, according to one or more embodiments.

[0044] FIGs. 10A-10C illustrate receiver operating characteristic (ROC) curves for augmentation of an early fusion classifier, according to example implementations.

[0045] FIGs. 11 A- 11C illustrate receiver operating characteristic (ROC) curves for augmentation of a late fusion classifier, according to example implementations.

[0046] FIGs. 12A-12C illustrate receiver operating characteristic (ROC) curi es for augmentation of a late fusion classifier trained with paired cfDNA and WBC DNA samples, according to example implementations.Atty. Docket No. 32917-65573 / WO 7 P0252-WO

[0047] FIG. 13 illustrates a plot revealing the correlation between cfDNA derived features and WBC DNA derived features, according to example implementations.

[0048] FIG. 14 illustrates the spread of cancer predictions for early stage cancer samples, according to example implementations.

[0049] The 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 illustrated herein may be employed without departing from the principles described herein.DETAILED DESCRIPTIONI. OVERVIEW

[0050] 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, doctors, and loved ones. For patients, early cancer detection allows them a greater chance of a beneficial outcome; for doctors, early cancer detection allows more insight into the context and status of disease progression and availability of more pathways of treatment that may lead to a beneficial outcome; for loved ones, early cancer detection increases the likelihood of not losing friends and family to the disease.

[0051] In many instances today, early detection and classification of cancer is only achieved when a patient is being worked up for symptoms that can indicate the presence of a cancer and only after a wide variety of cost-intensive medical procedures. This process, sometimes referred to as a “diagnostic odyssey,’’ causes stress and anxiety for patients and loved ones over sometimes multiple months of uncertainty'.

[0052] Recently, early cancer detection technology has progressed towards analyzing genetic fragments (e.g., DNA) in a sample from a person to determine if any of those genetic fragments originate from cancer cells. To increase the benefit of early cancer detection, the sample is often a relatively easy to acquire liquid sample from the person, such as blood, saliva, or urine. These new techniques allow doctors to identify a cancer presence in a patient that may not be detectable otherwise, e.g.. in conventional screening processes. For instance, consider the example of a person at risk for breast cancer. Traditionally, this person will regularly visit their doctor for a mammogram, which creates an image of their breast tissue (e.g., taking x-ray images) that a doctor uses to identify cancerous tissue. Unfortunately, with even the highest resolution mammograms, doctors are only able to identify tumors once theyAtty. Docket No. 32917-65573 / WO 8 P0252-WOare approximately a millimeter in size. This means that the cancer has been present for some time in the person and has gone undiagnosed and untreated. Visual determinations like this are typical for most cancers — that is, only identifiable once it has grown to a sufficient size to be detected by some sort of imaging technology7. For many more types of cancer, there is no screening paradigm like mammography available, and even regular doctor visits cannot detect the presence of a growing and advancing cancer. For example, the person having a risk of breast cancer above might at the same time be at high risk for ovarian cancer, which is undetectable by mammography or any other currently available image-based cancer screening modality7.

[0053] Cancer detection using analysis of genetic fragments in a patient, e.g.. in their blood, provides for technical improvements to these issues in the conventional screening paradigm. To illustrate, cancer cells will start shedding DNA fragments into a person’s bloodstream as soon as they form. This occurs when there are very few- of the cancer cells, and long before they’ve collectively achieved a size large enough to be visible with conventional imaging techniques. With the appropriate methods, therefore, a system that analyzes DNA fragments in the bloodstream (so-called “cell-free DNA” or “cfDNA”) could identify cancer presence in a person based on shed cancer DNA fragments, and, more importantly, the system could do so before the cancer is identifiable using more traditional cancer detection techniques, and even more so for cancers where no traditional cancer screening method exists.

[0054] Cancer detection based on the analysis of DNA fragments is enabled by nextgeneration sequencing (“NGS”) techniques. NGS, broadly, is a group of technologies that allows for high throughput sequencing of genetic material. The high throughput sequencing promotes maximum capturability of nucleic acid fragments in a biological sample. Without high throughput sequencing, faint signals may go undetected. As discussed in greater detail herein, NGS largely consists of (1) sample preparation, (2) DNA sequencing, and (3) data analysis. Sample preparation includes the laboratory7methods necessary7to prepare DNA fragments for sequencing, sequencing is the process of reading the ordered nucleotides in the samples, and data analysis includes processing and analyzing the genetic information in the sequencing data to identify7cancer presence.

[0055] Such an analysis to identify7the presence of cancer and classify its type (including organ system or tissue type and / or histologic origin) is further relevant when a patient has been diagnosed with cancer and additional information is needed for prognosis, treatment decision, and for a patient on treatment to identify residual disease, recurrence, orAtty. Docket No. 32917-65573 / WO 9 P0252-WOrelapse for a patient on treatment when a treatment has not or is no longer able to control cancer growth.

[0056] While these steps of NGS may help enable early cancer detection, they also introduce their own complex, detrimental challenges to consistent early cancer detection and, therefore, any improvements to sample preparation, DNA sequencing, and / or data analysis, including the pre-processing, algorithmic processing, and summary or presentation of predications or conclusions, results in an improvement to NGS technology, cancer detection technologies broadly, and early cancer detection more generally.

[0057] To illustrate, as an example, problems introduced in (1) sample preparation include DNA sample quality, sample contamination, label swap events, fragmentation bias, and accurate indexing. Remedying these problems would yield better genetic data for cancer detection.

[0058] Similarly, problems introduced in (2) sequencing include, for example, errors in accurate transcribing of fragments (e.g., reading an “A"’ instead of a C . etc.), incorrect or difficult fragment assembly and overlap, disparate coverage uniformity, sequencing depth vs. cost vs. specificity, and insufficient sequencing length. Again, remedying any of these problems would yield improved genetic data for cancer detection.

[0059] The problems in (3) data analysis are the most daunting and complex. The introduced challenges stem from the vast amounts of data created by NGS sequencing techniques. Sequencing data for a single sample can be on the order of hundreds of thousands (up to millions) of sequence reads, amounting to terabytes of data. Training analytical models ty pically involves collecting and processing thousands (up to tens of thousands or more) of samples with ascertained clinical cancer status, affected organ or organ group, primary site of cancer, underlying cancer biology, and other factors.

[0060] Effectively and efficiently analyzing that amount of data is both procedurally and computationally demanding. For instance, analyzing NGS sequencing involves several baseline processing steps such as, e.g., aligning reads to one another, aligning and mapping reads to a reference genome, de-duplicating duplicative reads, detecting contamination of a sample, identifying and calling variant genes, identifying and calling abnormally methylated individual genomic sites or regions, generating functional annotations, making predictions or identifications based on the called genomic methylation sites, etc. Performing any of these processes on terabytes (or more) of genetic data is computationally expensive for even the most powerful of computer architectures, and completely impossible for a normal humanAtty. Docket No. 32917-65573 / WO 10 P0252-WOmind to achieve within a clinical relevant span of time (i. e. , while the cancer is still in early stages when it is most likely to be curable).

[0061] Additionally, with the genetic sequencing data derived from the error-prone processes of sample preparation and sequence reading, large portions of the resulting genetic data may be low-quality or unusable for cancer identification. For example, large amounts of the genetic data may include contaminated samples, transcription errors, mismatched regions, overrepresented regions, non-informative regions, etc. and may be unsuitable for high accuracy cancer detection.

[0062] Identifying and accounting for low qualify genetic data across the vast amount of genetic data obtained from NGS sequencing is also procedurally and computationally challenging to accomplish and is also not practically performable by a human mind.

[0063] Overall, any process created that leads to more efficient processing of the large and structurally complex collection of sequencing data typically relied on for the analytical models used in early cancer detection enabled by NGS would be an improvement to cancer detection using NGS sequencing. Moreover, such processes as described and elaborated on herein were crafted as a direct solution to the various hurdles native to NGS technologies, and as such are non-routine and unconventional activity' in this technical field.

[0064] As a further example, under (3) data analysis, accurate identification of informative DNA from NGS data to identify a cancer presence is another difficult task native to the field of early cancer detection. To be effective, algorithms are sought to compensate for, e.g., errors generated by sample preparation and sequencing, the large scale of genomic and methylation variety present in the population, and to overcome the large-scale data analysis problems accompanying NGS techniques. That is, designing a machine learning model or models, or other computational processing algorithms, that enable early cancer detection based on next generation sequencing techniques must account for the problems that those techniques create.

[0065] Some of those techniques and models are discussed hereinbelow and particular improvements to state-of-the-art techniques and models are further discussed. Furthermore, such techniques are non-routine and unconventional activity in the technical field of endeavor.

[0066] One particular set of challenges arises in screening against nucleic acid fragmentation originating from white blood cells. Cell-free nucleic acid is sourced of shed nucleic acid fragments from tissue into the blood stream as well as nucleic acid fragments shed from white blood cells. When analyzing the cell-free nucleic acids in the blood stream,Atty. Docket No. 32917-65573 / WO 11 P0252-WOthe fragmentation originating from white blood cells can confound the tissue-shed nucleic acids. This confounding signal can include noise (i.e., fragmentation in the natural course of white blood cell operation) and potential cancer signal (i.e., fragmentation from hematological cancer types).

[0067] One example confounding signal is clonal hematopoiesis. Bone marrow produces healthy hematopoietic stem cells (HSCs). These stems cells can be prone to somatic mutation, followed by clonal expansion that yields a progressively larger population of lineage cells derived from the mutated HSC. As this mutant clone expands, particularly with aging, a substantial fraction of circulating nucleic acids originate from hematopoietic precursors and white blood cells rather than from diseased tissues of interest, thereby skewing cfDNA sequencing and epigenetic readouts. Representative driver genes involved in such clones include DNMT3A, TET2, PPM1D, TP53, and JAK2. These driver genes that undergo somatic mutation perturb DNA methylation, generating sequence and epigenetic variants that can be misattributed in cfDNA assays.

[0068] On the other hand, WBC signal that can help with cancer detection could be keyed into. One source is hematologic malignancy, where neoplastic blood cells expand and amplify tumor-associated variants or epigenetic patterns in circulation. Another source is the host immune response to cancer invasion, which activates WBCs and generates characteristic transcriptional, epigenetic, and cytokine signatures. Either of these signals can implicate cancer in the subject.

[0069] Parsing the white blood cell signal is a particular challenge that is addressed through augmentation of the cancer classifier with pairings of a cfDNA sample and a WBC DNA sample. As used throughout, cfDNA sample may refer to the physical sample of cell-free DNA fragments, one or more sequence reads derived from sequencing the cfDNA fragments, or some combination thereof. Likewise, WBC DNA sample may refer to the physical sample of WBC DNA fragments collected from the subject, one or more sequence reads derived from sequencing the WBC DNA fragments, or some combination thereof.I. A. CANCER CLASSIFICATION WORKFLOW

[0070] FIG. 1 is an exemplary flowchart describing an overall workflow 100 of cancer classification of a sample, according to one or more embodiments. The workflow 100 is by one or more entities, e.g., including a healthcare provider, a sequencing device, an analytics system, etc. Objectives of the workflow include detecting and / or monitoring cancer in subjects. From a healthcare standpoint, the workflow 100 can serve to supplement other existing diagnostic workups, e.g., cancer screening and early detection tools. The workflowAtty. Docket No. 32917-65573 / WO 12 P0252-WO100 may serve to provide early cancer detection and / or routine cancer monitoring, minimal residual disease detection, prognosis, treatment prediction, or subtyping information to better inform treatment plans for subjects diagnosed with cancer. The overall workflow 100 may include additional / fewer steps than those shown in FIG. 1.

[0071] A healthcare provider performs sample collection 110. A subject to undergo screening, early detection or cancer classification, visits their healthcare provider. The healthcare provider collects the sample for performing cancer classification. Examples of biological samples include, but are not limited to: tissue biopsies, liquid biopsies (e.g., blood, whole blood, plasma, serum, urine, cerebrospinal fluid, fecal, saliva, sweat, tears, pleural fluid, pericardial fluid, or peritoneal fluid) of the subject. The sample includes genetic material belonging to the subject, which may be extracted and sequenced for screening, early detection, or cancer classification. Along with the sample, the healthcare provider may collect other information relating to the subject, e.g., biological sex, age, race, smoking status, other health metrics, any prior diagnoses, etc.

[0072] Once the sample is collected, the sample is provided to a laboratory process and sequencing device. A sequencing device performs sample sequencing 120. A lab clinician may perform one or more processing steps to the sample in preparation of sequencing. Once prepared, the lab technician loads the sample in the sequencing device. An example of devices utilized in sequencing is further described in conjunction with FIGs.9A & 9B. The sequencing device generally extracts and isolates fragments of nucleic acid that are sequenced to determine a sequence of nucleobases corresponding to the fragments. Sequencing may also include amplification of nucleic 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., further described in FIGs. 2A & 2B) can determine methylation status through bisulfite conversion of unmethylated cytosines at CpG sites. Sample sequencing 120 yields sequences for a plurality of nucleic acid fragments in the sample. In one or more embodiments, the sequences may include methylation state vectors, wherein each methylation state vector describes the methylation statuses for CpG sites on a fragment.

[0073] The sequencing data generated by the sequencing device is provided to an analytics system to perform analyses to generate predictions. An analytics system performs pre-analysis processing 130. An example analytics system is described in FIG. 9B. Preanalysis processing 130 may include, but not limited to, de-duplication of sequence reads,Atty. Docket No. 32917-65573 / WO 13 P0252-WOdetermining metrics relating to coverage, determining whether the sample is contaminated (including determining WBC contamination), removal of contaminated fragments, calling sequencing error, etc.

[0074] The analytics system performs one or more analyses 140 on the sequencing data. The analyses are statistical analyses or application of one or more trained models to predict at least a cancer status of the subject from whom the sample is 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 mutation, etc. The analyses 140 may include, anomalous methylation identification 142 (e.g., further described in FIGs. 3A & 3B), feature extraction 144 (e.g., further described FIGs. 4A & 4B), and cancer classification 146 to determine a cancer prediction (e.g.. further described in FIGs. 4A & 4B), and sample swap detection 148 to detect whether sample labels have been swapped. Cancer classification generally entails inputting the extracted features to determine a cancer prediction. The cancer prediction may be a label or a value. The label may indicate a particular cancer state. A binary label may indicate presence or absence of a cancer signal; whereas a multiclass label can indicate one or more cancer signal origins from a plurality' of potential cancer signal origins that are screened for. Cancer signal origins may be split based on various characteristics of the cancer. For example, cancer signal origins may be split according to organ type and / or tumor biology type. In other examples, cancer signal origins, or other predictive information generated with the cancer prediction, may be further split according to progression (e.g., Stage I, II, III, or IV), prognosis, prediction of response to a candidate treatment, or presence of residual disease after or on treatment. A value may indicate a likelihood of a particular cancer state, e g., a likelihood of cancer, and / or a likelihood of a particular cancer signal origin.

[0075] The analytics system returns the prediction 150 to the healthcare provider. The healthcare provider may establish or adjust a treatment plan based on the cancer prediction. In some embodiments, the analytics system may return to the healthcare provider and / or the subject a diagnostic workup recommendation 160 identifying one or more diagnostic steps that might be used to confirm the test result (cancer prediction) and result in a cancer diagnosis. In such embodiments, the analytics system may store a database associating diagnostic workup options generally accepted by healthcare professionals to be useful diagnostic workup steps in the presence of a predicted cancer signal origin.Optimization of treatment is further described in Section V.D. Treatment. In some embodiments, the analytics system may leverage the cancer classification workflow forAtty. Docket No. 32917-65573 / WO 14 P0252-WOprognosis determination, treatment personalization, evaluation of treatment, monitoring cancer status, etc.LB. METHYLATION OVERVIEW

[0076] In accordance with the present description, cfDNA fragments from a subject are treated, for example by converting unmethylated cytosines to uracils, sequenced and the sequence reads compared to a reference genome to identify the methylation states at specific CpG sites within the DNA fragments. Each CpG site may be methylated or unmethylated. Identification of anomalously methylated fragments, in comparison to healthy subjects, may provide insight into a subject’s cancer status. As is well known in the art, DNA methylation changes (compared to healthy controls) can cause different effects, which may contribute to cancer. Various challenges arise in the identification of informatively methylated cfDNA fragments. First off, determining a DNA fragment to be informatively methylated can hold weight in comparison with a group of control subjects, such that if the control group is small in number, the determination loses confidence due to statistical variability' within the smaller size of the control group. Additionally, among a group of control subjects, methylation status can vary which can be difficult to account for when determining a subject’s DNA fragments to be informatively methylated. On another note, methylation of a cytosine at a CpG site can causally influence methylation at a subsequent CpG site. To encapsulate this dependency can be another challenge in itself.

[0077] Methylation can typically occur in deoxyribonucleic acid (DNA) when a hydrogen atom on the pyrimidine ring of a cytosine base is converted to a methyl group, forming 5-methylcytosine. In particular, methylation can occur at dinucleotides of cytosine and guanine referred to herein as CpG sites”. In other instances, methylation may occur at a cytosine not part of a CpG site or at another nucleotide that is not cytosine; however, these are rarer occurrences. In this present disclosure, methylation is discussed in reference to CpG sites for the sake of clarity. Anomalous DNA methylation can be identified as hypermethylation or hypomethy lation, both of which may be indicative of cancer status. Throughout this disclosure, hypermethylation and hypomethylation can be characterized for a DNA fragment, if the DNA fragment comprises more than a threshold number of CpG sites with more than a threshold percentage of those CpG sites being methylated or unmethylated. In addition to hypermethylation or hypomethylation, an informative methylation status of a DNA fragment can further be characterized of a sequence of methylated and unmethylated CpGs that are not frequently observed in a healthy population and that is indicative for the presence of cancer or the presence of one or few cancer signal origins.Atty. Docket No. 32917-65573 / WO 15 P0252-WO

[0078] The principles described herein can be equally applicable for the detection of methylation in a non-CpG context, including non-cytosine methylation. In such embodiments, the wet laboratory' assay used to detect methylation may vary from those described herein. Further, the methylation state vectors discussed herein may contain elements that are generally sites where methylation has or has not occurred (even if those sites are not CpG sites specifically). With that substitution, the remainder of the processes described herein can be the same, and consequently the inventive concepts described herein can be applicable to those other forms of methylation.I.C. DEFINITIONS

[0079] The term “cell free nucleic acid’' or “cfNA” refers to nucleic acid fragments that circulate in a subject's body (e.g.. blood) or are present in other bodily liquids like urine, cerebrovascular fluid, and originate from one or more healthy cells and / or from one or more unhealthy cells (e.g., cancer cells). The term “cell free DNA,” or “cfDNA” refers to deoxyribonucleic acid fragments present in a subject’s body outside of the cell (e.g., in blood plasma, urine, sputum, cerebrospinal fluid, etc.).

[0080] The term "genomic nucleic acid,” “genomic DNA,” or “gDNA” refers to nucleic acid molecules or deoxyribonucleic acid molecules obtained from one or more cells. In various embodiments, gDNA can be extracted from healthy cells (e.g., non-tumor cells) or from tumor cells (e.g., a biopsy sample). In some embodiments, gDNA can be extracted from a cell derived from a blood cell lineage, such as a white blood cell.

[0081] The term “circulating tumor DNA” or “ctDNA” refers to nucleic acid fragments that originate from tumor cells or other ty pes of cancer cells, and which may be released into a bodily fluid of a subject (e.g., blood, sweat, urine, or saliva) as result of biological processes such as apoptosis or necrosis of dying cells or actively released by viable tumor cells.

[0082] The term “DNA fragment,” “fragment,” or “DNA molecule” may7generally refer to any deoxyribonucleic acid fragments, i.e., cfDNA, gDNA, ctDNA, etc.

[0083] The term “informative fragment,” may generally refer to any fragment having an informative genetic signature. The term “anomalously methylated fragment,” or “fragment with an anomalous methylation pattern” refers to an informative fragment that has an informative methylation pattern. Anomalous methylation of a fragment may be determined using probabilistic models to identify unexpectedness of observing a fragment’s methylation pattern in a control group.Atty. Docket No. 32917-65573 / WO 16 P0252-WO

[0084] As used herein, the term ‘‘about’' or “approximately"’ can mean within an acceptable error range for the particular value as determined by one of ordinary’ skill in the art, which can depend in part on how the value is measured or determined, e.g., the limitations of the measurement system. For example, “about” can mean within 1 or more than 1 standard deviation, per the practice in the art. “About” can mean a range of ±20%, ±10%, ±5%, or ±1% of a given value. The term “about” or “approximately” can mean within an order of magnitude, within 5-fold, or within 2-fold, of a value. Where particular values are described in the application and claims, unless otherw ise stated the term “about” meaning within an acceptable error range for the particular value should be assumed. The term “about” can have the meaning as commonly understood by one of ordinary skill in the art. The term “about” can refer to ±10%. The term “about” can refer to ±5%.

[0085] 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 that includes gDNA or cell-free DNA. A sample can be a liquid sample or a solid sample (e.g., a cell or tissue sample). A biological sample can be a bodily fluid, such as blood, plasma, serum, urine, vaginal fluid, fluid from a hydrocele (e.g., of the testis), vaginal flushing fluids, pleural fluid, pericardial fluid, peritoneal fluid, ascitic fluid, cerebrospinal fluid, saliva, sw eat, tears, sputum, bronchoalveolar lavage fluid, discharge fluid from the nipple, aspiration fluid from different parts of the body (e g., thyroid, breast), etc. A biological sample can be a stool sample. A biological sample can include any tissue or material derived from a living or dead subject. A biological sample can be a cell-free sample. A biological sample can comprise a nucleic acid (e.g., DNA or RNA) or a fragment thereof.

[0086] As used herein, the terms “control,” “control sample,” “reference,” “reference sample,” “normal,” and “normal sample” describe a sample from a subject that does not have a particular condition, or is otherwise healthy. In an example, a method as disclosed herein can be performed on a subject having a tumor, where the reference sample is a sample taken from a healthy tissue of the subject. A reference sample can be obtained from the subject, or from a database. The reference can be, e.g., a reference genome that is used to map nucleic acid fragment sequences obtained from sequencing a sample from the subject. A reference genome can refer to a haploid or diploid genome to which nucleic acid fragment sequences from the biological sample and a constitutional sample can be aligned and compared. An example of a constitutional sample can be DNA of white blood cells obtained from the subject. For a haploid genome, there can be only one nucleotide at each locus. For a diploidAtty. Docket No. 32917-65573 / WO 17 P0252-WOgenome, heterozy gous loci can be identified; each heterozygous locus can have two alleles, where either allele can allow a match for alignment to the locus.

[0087] As used herein, the term “cancer” or “tumor” refers to an abnormal mass of tissue in which the growth of the mass surpasses and is not coordinated with the growth of normal tissue.

[0088] As used herein, the term “genomic” refers to a characteristic of the genome of an organism. Examples of genomic characteristics include, but are not limited to, those relating to the primary nucleic acid sequence of all or a portion of the genome (e.g., the presence or absence of a nucleotide polymorphism, indel, sequence rearrangement, mutational frequency, etc.), the copy number of one or more particular nucleotide sequences within the genome (e.g., copy number, allele frequency fractions, single chromosome or entire genome ploidy, etc.), the epigenetic status of all or a portion of the genome (e.g., covalent nucleic acid modifications such as methylation, histone modifications, nucleosome positioning, etc.), the expression profile of the organism’s genome (e.g., gene expression levels, isotype expression levels, gene expression ratios, etc.).

[0089] As used herein, the phrase “healthy,” refers to a subject possessing good health. A healthy subj ect can demonstrate an absence of any malignant or non-malignant disease. A “healthy subject” can have other diseases or conditions, unrelated to the condition being assayed, which can normally not be considered “healthy.”

[0090] As used herein, the term “methylation” refers to a modification of deoxyribonucleic acid (DNA) where a hydrogen atom on the pyrimidine ring of a cytosine base is converted to a methyl group, forming 5-methylcytosine. In particular, methylation tends to occur at dinucleotides of cytosine and guanine referred to herein as “CpG sites.” In other instances, methylation may occur at a cytosine not part of a CpG site or at another nucleotide that’s not cytosine; however, these are rarer occurrences. Anomalous cfDNA methylation can be identified as hypermethylation or hypomethylation, both of which may be indicative of cancer status. DNA methylation anomalies (compared to healthy controls) can cause different effects, which may contribute to cancer. The principles described herein are equally applicable for the detection of methylation in a CpG context and non-CpG context, including non-cytosine methylation. Further, the methylation state vectors may contain elements that are generally vectors of sites where methylation has or has not occurred (even if those sites are not CpG sites specifically).

[0091] As used interchangeably herein, the term “methylation fragment” or “nucleic acid methylation fragment” refers to a sequence of methylation states for each CpG site in aAtty. Docket No. 32917-65573 / WO 18 P0252-WOplurality of CpG sites, determined by a methylation sequencing of nucleic acids (e.g., a nucleic acid molecule and / or a nucleic acid fragment). In a methylation fragment, a location and methylation state for each CpG site in the nucleic acid fragment is determined based on the alignment of the sequence reads (e.g., obtained from sequencing of the nucleic acids) to a reference genome. A nucleic acid methylation fragment comprises a methylation state of each CpG site in a plurality’ of CpG sites (e.g.. a methylation state vector), which specifies the location of the nucleic acid fragment in a reference genome (e.g., as specified by the position of the first CpG site in the nucleic acid fragment using a CpG index, or another similar metric) and the number of CpG sites in the nucleic acid fragment. Alignment of a sequence read to a reference genome, based on a methylation sequencing of a nucleic acid molecule, can be performed using a CpG index. As used herein, the term “CpG index” refers to a list of each CpG site in the plurality of CpG sites (e.g., CpG 1, CpG 2, CpG 3, etc.) in a reference genome, such as a human reference genome, which can be in electronic format. The CpG index further comprises a corresponding genomic location, in the corresponding reference genome, for each respective CpG site in the CpG index. Each CpG site in each respective nucleic acid methylation fragment is thus indexed to a specific location in the respective reference genome, which can be determined using the CpG index.

[0092] The term “nucleic acid” can refer to deoxyribonucleic acid (DNA), ribonucleic acid (RNA) or any hybrid or fragment thereof. The nucleic acid in the sample can be a cell-free nucleic acid. In various embodiments, the majority of DNA in a biological sample that has been enriched for cell-free DNA (e g., a plasma sample obtained via a centrifugation protocol) can be cell-free (e.g., greater than 50%, 60%, 70%, 80%, 90%, 95%, or 99% of the DNA can be cell-free). A biological sample can be treated to physically disrupt tissue or cell structure (e.g., centrifugation and / or cell lysis), thus releasing intracellular components into a solution which can further contain enzymes, buffers, salts, detergents, and the like which can be used to prepare the sample for analysis.

[0093] As used herein, the term “true positive” (TP) refers to a subject having a condition. “True positive” can refer to a subject that has a tumor, a cancer, a pre-cancerous condition (e.g., a pre-cancerous lesion), a localized or a metastasized cancer, or a non-malignant disease. “True positive” can refer to a subject having a condition and is identified as having the condition by an assay or method of the present disclosure, like having a cancer that will respond to a treatment, having residual disease after or during a treatment, or being likely to encounter an event like disease progression, relapse, or even cancer-related death within a pre-specified time frame. As used herein, the term “true negative” (TN) refers to aAtty. Docket No. 32917-65573 / WO 19 P0252-WOsubject that does not have a condition or does not have a detectable condition. True negative can refer to a subject that does not have a disease or a detectable disease, such as a tumor, a cancer, a pre-cancerous condition (e.g., a pre-cancerous lesion), a localized or a metastasized cancer, anon-malignant disease, or a subject that is otherwise healthy. True negative can refer to a subject that does not have a condition or does not have a detectable condition, or is identified as not having the condition by an assay or method of the present disclosure.

[0094] As used herein, the term ‘"reference genome’" refers to any particular known, sequenced or characterized genome, whether partial or complete, of any organism or virus that may be used to reference identified sequences from a subject. Exemplary reference genomes used for human subjects as well as many other organisms are provided in the online genome browser hosted by the National Center for Biotechnology Information (“NCBI”) or the University of California, Santa Cruz (UCSC). A ‘‘genome” refers to the complete genetic information of an organism or virus, expressed in nucleic acid sequences. As used herein, a reference sequence or reference genome often is an assembled or partially assembled genomic sequence from a subject or multiple subjects. In some embodiments, a reference genome is an assembled or partially assembled genomic sequence from one or more huma subjects. The reference genome can be viewed as a representative example of a species’ set of genes. In some embodiments, a reference genome comprises sequences assigned to chromosomes. Exemplary human reference genomes include but are not limited to NCBI build 34 (UCSC equivalent: hgl6). NCBI build 35 (UCSC equivalent: hg!7). NCBI build 36.1 (UCSC equivalent: hgl 8), GRCh37 (UCSC equivalent: hg 19), and GRCh38 (UCSC equivalent: hg38).

[0095] As used herein, the term “sequence reads” or “reads” refers to nucleotide sequences produced by any sequencing process described herein or known in the art. Reads can be generated from one end of nucleic acid fragments (“single-end reads”), and sometimes are generated from both ends of nucleic acids (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 often associated with the particular sequencing technology. High-throughput methods, for example, provide sequence reads that can vary in size from tens to hundreds of base pairs (bp). In some embodiments, the sequence reads are of a mean, median or average length of about 15 bp to 900 bp long (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, aboutAtty. Docket No. 32917-65573 / WO 20 P0252-WO120 bp. about 130, 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 sequence reads are of a mean, median or average length of about 1000 bp, 2000 bp, 5000 bp, 10,000 bp, or 50,000 bp or more. Nanopore sequencing, for example, can provide sequence reads that can vary in size from tens to hundreds to thousands of base pairs. Illumina parallel sequencing can provide sequence reads that do not vary as much, for example, most of the sequence reads can be smaller than 200 bp. 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 part of a nucleic acid fragment, can correspond to a string of nucleotides at one or both ends of a nucleic acid fragment, or can correspond to nucleotides of the entire nucleic acid fragment. A sequence read can be obtained in a variety of ways, e.g., using sequencing techniques or using probes, e.g., in hybridization arrays or capture probes, or amplification techniques, such as the polymerase chain reaction (PCR) or linear amplification using a single primer or isothermal amplification.

[0096] As used herein, the terms “sequencing” and the like as used herein refers generally to any and all biochemical processes that may be used to determine the order of biological macromolecules such as nucleic acids or proteins. For example, sequencing data can include all or a portion of the nucleotide bases in a nucleic acid molecule such as a DNA fragment.

[0097] As used herein, the term “sequencing depth,” is interchangeably used with the term “coverage” and refers to the number of times a locus is covered by a consensus sequence read corresponding to a unique nucleic acid target molecule aligned to the locus; e.g., the sequencing depth is equal to the number of unique nucleic acid target molecules covering the locus. The locus can be as small as a nucleotide, or as large as a chromosome arm, or as large as an entire genome. Sequencing depth can be expressed as “Yx”, e.g., 50x, lOOx, etc., where “Y” refers to the number of times a locus is covered with a sequence corresponding to a nucleic acid target, e g., the number of times independent sequence information is obtained covering the particular locus. 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 whole genome, in which case Y can refer to the mean or average number of times a locus or a haploid genome, or a whole genome, respectively, is sequenced. When a mean depth is quoted, the actual depth for different lociAtty. Docket No. 32917-65573 / WO 21 P0252-WOincluded in the dataset can span over a range of values. Ultra-deep sequencing can refer to at least lOOx in sequencing depth at a locus.

[0098] 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. Sensitivity7can characterize the ability of an assay or method to correctly identify a proportion of the population that truly has a condition. For example, sensitivity7can characterize the ability of a method to correctly identify the number of subjects within a population having cancer. In another example, sensitivity7can characterize the ability of a method to correctly identify the one or more markers indicative of cancer.

[0099] 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 a proportion of the population that truly does not have a condition. For example, specificity can characterize the ability of a method to correctly identify the number of subjects within a population not having cancer. In another example, specificity characterizes the ability of a method to correctly identify one or more markers indicative of cancer.

[0100] As used herein, the term “subject” refers to any living or non-living organism, including but not limited to a human (e.g., a male human, female human, fetus, pregnant female, child, or the like), a non-human animal, a plant, a bacterium, a fungus or a protist. Any human or non-human animal can serve as a subject, including but not limited to mammal, reptile, avian, amphibian, fish, ungulate, ruminant, bovine (e.g., cattle), equine (e.g., horse), caprine and ovine (e.g., sheep, goat), swine (e.g., pig), camelid (e.g., camel, llama, alpaca), monkey, ape (e.g., gorilla, chimpanzee), ursid (e.g., bear), poultry, dog, cat, mouse, rat. fish, dolphin, whale, and shark. In some embodiments, a subject is a male or female of any stage (e.g., a man, a woman or a child). A subject from whom a sample is taken, or is treated by any of the methods or compositions described herein can be of any age and can be an adult, infant or child.

[0101] As used herein, the term “tissue” can correspond to a group of cells that group together as a functional unit. More than one type of cell can be found in a single tissue. Different types of tissue may consist of different types of cells (e.g., hepatocytes, alveolar cells or blood cells), but also can correspond to tissue from different organisms (mother vs. fetus) or to healthy cells vs. 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 aspects, the term “tissue” or “tissue type” can be usedAtty. Docket No. 32917-65573 / WO 22 P0252-WOto refer to a tissue from which a cell-free nucleic acid originates. 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.

[0102] The terminology used herein is for the purpose of describing particular cases 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. Furthermore, to the extent that the terms “including,” “includes,” “having,” “has,” “with,” or variants thereof are used in either the detailed description and / or the claims, such terms are intended to be inclusive in a manner similar to the term “comprising.”I D. EXAMPLE ANALYTICS SYSTEM

[0103] FIG. 5A is an exemplary flowchart of devices for sequencing nucleic acid samples according to one or more embodiments. This illustrative flowchart includes devices such as a sequencer 520 and an analytics system 500. The sequencer 520 and the analytics system 500 may work in tandem to perform one or more steps in the processes.

[0104] In various embodiments, the sequencer 520 receives an enriched nucleic acid sample 510. As shown in FIG. 5 A, the sequencer 520 can include a graphical user interface 525 that enables user interactions with particular tasks (e g., initiate sequencing or terminate sequencing) as well as one more loading stations 530 for loading a sequencing cartridge including the enriched fragment samples and / or for loading necessary buffers for performing the sequencing assays. Therefore, once a user of the sequencer 520 has provided the necessary' reagents and sequencing cartridge to the loading station 530 of the sequencer 520, the user can initiate sequencing by interacting with the graphical user interface 525 of the sequencer 520. Once initiated, the sequencer 520 performs the sequencing and outputs the sequence reads of the enriched fragments from the nucleic acid sample 510.

[0105] In some embodiments, the sequencer 520 is communicatively coupled with the analytics system 500. The analytics system 500 includes some number of computing devices used for processing the sequence reads for various applications such as assessing methylation status at one or more CpG sites, variant calling or quality control. The sequencer 520 may provide the sequence reads in a BAM file format to the analytics system 500. The analytics system 500 can be communicatively coupled to the sequencer 520 through a wireless, wired, or a combination of wireless and wired communication technologies. Generally, the analytics system 500 is configured with a processor and non-transitory computer-readable storage medium storing computer instructions that, when executed by the processor, cause theAtty. Docket No. 32917-65573 / WO 23 P0252-WOprocessor to process the sequence reads or to perform one or more steps of any of the methods or processes disclosed herein.

[0106] In some embodiments, the sequence reads may be aligned to a reference genome using known methods in the art to determine alignment position information.Alignment position may generally describe a beginning position and an end position of a region in the reference genome that corresponds to a beginning nucleotide base and an end nucleotide base of a given sequence read. Corresponding to methylation sequencing, the alignment position information may be generalized to indicate a first CpG site and a last CpG site included in the sequence read according to the alignment to the reference genome. The alignment position information may further indicate methylation statuses and locations of all CpG sites in a given sequence read. A region in the reference genome may be associated with a gene or a segment of a gene; as such, the analytics system 500 may label a sequence read with one or more genes that align to the sequence read. In one embodiment, fragment length (or size) is determined from the beginning and end positions.

[0107] In various embodiments, for example when a paired-end sequencing process is used, a sequence read is comprised of a read pair denoted as R_1 and R_2. For example, the first read R_1 may be sequenced from the first end of a double-stranded DNA (dsDNA) molecule w hereas the second read R_2 may be sequenced from the second end of the doublestranded DNA (dsDNA). Therefore, nucleotide base pairs of the first read R_1 and second read R 2 may be aligned consistently (e.g., in opposite orientations) with nucleotide bases of the reference genome. Alignment position information derived from the read pair R_1 and R_2 may include a beginning position in the reference genome that corresponds to an end of a first read (e.g., R_l) and an end position in the reference genome that corresponds to an end of a second read (e.g., R_2). In other words, the beginning position and end position in the reference genome can represent the likely location within the reference genome to which the nucleic acid fragment corresponds. An output file having SAM (sequence alignment map) format or BAM (binary ) format may be generated and output for further analysis.

[0108] Referring now to FIG. 5B, FIG. 5B is a block diagram of an analytics system 500 for processing DNA samples according to one embodiment. The analytics system implements one or more computing devices for use in analyzing DNA samples. The analytics system 500 includes a sequence processor 540, sequence database 545, model database 555, models 550, parameter database 565, and score engine 560. In some embodiments, the analytics system 500 performs some or all of the processes described throughout this disclosure.Atty. Docket No. 32917-65573 / WO 24 P0252-WO

[0109] The sequence processor 540 generates methylation state vectors for fragments from a sample. At each CpG site on a fragment, the sequence processor 540 generates a methylation state vector for each fragment specifying a location of the fragment in the reference genome, a number of CpG sites in the fragment, and the methylation state of each CpG site in the fragment whether methylated, unmethylated, or indeterminate via the process 200 of FIG. 2A. The sequence processor 540 may store methylation state vectors for fragments in the sequence database 545. Data in the sequence database 545 may be organized such that the methylation state vectors from a sample are associated to one another.

[0110] Further, multiple different models 550 may be stored in the model database 555 or retrieved for use with test samples. In one example, a model is a trained cancer classifier for determining a cancer prediction for a test sample using a feature vector derived from informative fragments. The training and use of the cancer classifier will be further discussed in conjunction with Section III. Cancer Classifier for Determining Cancer. The analytics system 500 may train the one or more models 550 and store various trained parameters in the parameter database 565. The analytics system 500 stores the models 550 along with functions in the model database 555.

[0111] During inference, the score engine 560 uses the one or more models 550 to return outputs. The score engine 560 accesses the models 550 in the model database 555 along with trained parameters from the parameter database 565. According to each model, the score engine receives an appropriate input for the model and calculates an output based on the received input, the parameters, and a function of each model relating the input and the output. In some use cases, the score engine 560 further calculates metrics correlating to a confidence in the calculated outputs from the model. In other use cases, the score engine 560 calculates other intermediary values for use in the model.II. SAMPLE SEQUENCING & PROCESSINGII. A. GENERATING METHYLATION STATE VECTORS FOR DNA FRAGMENTS

[0112] FIG. 2 A is an exemplary flowchart describing a process 200 of sequencing a fragment of cfDNA to obtain a methylation state vector, according to one or more embodiments. In order to analyze DNA methylation, an analytics system first obtains 210 a sample from a subject comprising a plurality of cfDNA molecules. In additional embodiments, the process 200 may be applied to sequence other types of DNA molecules. The process 200 is an embodiment of sample sequencing 120 of FIG. 1.

[0113] From the sample, the analytics system can isolate 210 each cfDNA molecule. The cfDNA molecules can be treated 220 to convert unmethylated cytosines to uracils. InAtty. Docket No. 32917-65573 / WO 25 P0252-WOone embodiment, the method uses a bisulfite treatment of the DNA which converts the unmethylated cytosines to uracils without converting the methylated cytosines. For example, a commercial kit such as the EZ DNA Methylation™ - Gold, EZ DNA Methylation™ -Direct or an EZ DNA Methylation™ - Lightning kit (available from Zymo Research Corp (Irvine, CA)) is used for the bisulfite conversion. In another embodiment, the conversion of unmethylated cytosines to uracils is accomplished using an enzymatic reaction. For example, the conversion can use a commercially available kit for conversion of unmethylated cytosines to uracils, such as APOBEC-Seq (NEBiolabs, Ipswich, MA).

[0114] From the converted cfDNA molecules, a sequencing library can be prepared 230. During library' preparation, molecular identifiers, including but not limited to unique molecular identifiers (UMI) or sample barcodes, can be added to the nucleic acid molecules (e.g., DNA molecules) through adapter ligation. The molecular identifiers can be short nucleic acid sequences (e.g., 4-10 base pairs) that are added to ends of DNA fragments (e.g., DNA molecules fragmented by physical shearing, enzymatic digestion, and / or chemical fragmentation) during adapter ligation. Molecular identifiers can serve as a tag to identify sequence reads originating from a single sample or subject. UMIs represent a special case as degenerate base pairs that serve as a unique tag that can be used to identify sequence reads originating from a specific DNA fragment. During PCR amplification following adapter ligation, the molecular identifiers can be replicated along with the attached DNA fragment. This can provide a way to identify sequence reads that came from the same original fragment in downstream analysis.

[0115] Optionally, the sequencing library may be enriched 235 for cfDNA molecules, or genomic regions, that are informative for cancer status using a plurality' of hybridization probes. The hybridization probes are short oligonucleotides capable of hybridizing to particularly specified cfDNA molecules, or targeted regions, and enriching for those fragments or regions for subsequent sequencing and analysis. Hybridization probes may be used to perform a targeted, high-depth analysis of a set of specified CpG sites of interest to the researcher. Hybridization probes can be tiled across one or more target sequences at a coverage of IX, 2X, 3X, 4X, 5X, 6X, 7X, 8X, 9X, 10X, or more than 10X. For example, hybridization probes tiled at a coverage of 2X comprises overlapping probes such that each portion of the target sequence is hybridized to 2 independent probes. Hybridization probes can be tiled across one or more target sequences at a coverage of less than IX.

[0116] In one embodiment, the hybridization probes are designed to enrich for DNA molecules that have been treated (e.g., using bisulfite) for conversion of unmethylatedAtty. Docket No. 32917-65573 / WO 26 P0252-WOcytosines to uracils. During enrichment, hybridization probes (also referred to herein as “probes”) can be used to target and pull-down nucleic acid fragments informative for the presence or absence of cancer (or disease), cancer status, or a cancer classification (e.g., cancer class or tissue of origin). The probes may be designed to anneal (or hybridize) to a target (complementary) strand of DNA. The target strand may be the “positive” strand (e.g., the strand transcribed into mRNA. and subsequently translated into a protein) or the complementary “negative” strand. The probes may range in length from 10s, 100s, or 1000s of base pairs. The probes can be designed based on a methylation site panel. The probes can be designed based on a panel of targeted genes to analyze particular mutations or target regions of the genome (e.g., of the human or another organism) that are suspected to correspond to certain cancers or other types of diseases. Moreover, the probes may cover overlapping portions of a target region.

[0117] Once prepared, the sequencing library or a portion thereof can be sequenced 240 to obtain a plurality7of sequence reads. The sequence reads may be in a computer-readable, digital format for processing and interpretation by computer software. The sequence reads may be aligned to a reference genome to determine alignment position information. The alignment position information may indicate a beginning position and an end position of a region in the reference genome that corresponds to a beginning nucleotide base and end nucleotide base of a given sequence read. Alignment position information may also include sequence read length, which can be determined from the beginning position and end position. A region in the reference genome may be associated with a gene or a segment of a gene. A sequence read can be comprised of a read pair denoted as R and R2. For example, the first read R1may be sequenced from a first end of a nucleic acid fragment whereas the second read Rzmay be sequenced from the second end of the nucleic acid fragment. Therefore, nucleotide base pairs of the first read R±and second read R2may be aligned consistently (e.g., in opposite orientations) with nucleotide bases of the reference genome. Alignment position information derived from the read pair R and R2may include a beginning position in the reference genome that corresponds to an end of a first read (e.g., R-,. ) and an end position in the reference genome that corresponds to an end of a second read (e.g., R2). In other words, the beginning position and end position in the reference genome can represent the likely location within the reference genome to which the nucleic acid fragment corresponds. An output file having SAM (sequence alignment map) format or BAM (binary)Atty. Docket No. 32917-65573 / WO 27 P0252-WOformat may be generated and output for further analysis such as methylation state determination.

[0118] From the sequence reads, the analytics system determines 250 a location and methylation state for each CpG site based on alignment to a reference genome. The analytics system generates 260 a methylation state vector for each fragment specifying a location 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 metric), a number of CpG sites in the fragment, and the methylation state of each CpG site in the fragment whether methylated (e.g., denoted as M), unmethylated (e.g., denoted as U), or indeterminate (e.g., denoted as I). Observed states can be states of methylated and unmethylated, whereas an unobserved state is indeterminate. Indeterminate methylation states may originate from sequencing errors and / or disagreements between methylation states of a DNA fragment's complementary strands. The methylation state vectors may be stored in temporary or persistent computer memory for later use and processing. Further, the analytics system may remove duplicate reads or duplicate methylation state vectors from a single sample. The analytics system may determine that a certain fragment with one or more CpG sites has an indeterminate methylation status over a threshold number or percentage, and may exclude such fragments or selectively include such fragments but build a model accounting for such indeterminate methylation statuses.

[0119] FIG. 2B is an exemplary illustration of the process 200 of FIG. 2A of sequencing a cfDNA molecule to obtain a methylation state vector, according to one or more embodiments. As an example, the analytics system receives a cfDNA molecule 212 that, 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 treatment step 220, the cfDNA molecule 212 is converted to generate a converted cfDNA molecule 222. During the treatment 220, the second CpG site which was unmethylated has its cytosine converted to uracil. However, the first and third CpG sites w ere not converted.

[0120] After conversion, a sequencing library 230 is prepared and sequenced 240 to generate a sequence read 242. The analytics system aligns 250 the sequence read 242 to a reference genome 244. The reference genome 244 provides the context as to what position in a human genome the fragment cfDNA originates from. In this simplified example, the analytics system aligns 250 the sequence read 242 such that the three CpG sites correlate to CpG sites 23, 24, and 25 (arbitrary reference identifiers used for convenience of description). The analytics system can thus generate information both on methylation status of all CpG sites on the cfDNA molecule 212 and the position in the human genome that the CpG sitesAtty. Docket No. 32917-65573 / WO 28 P0252-WOmap to. As shown, the CpG sites on sequence read 242 which are methylated are read as cytosines. In this example, the cytosines appear in the sequence read 242 only in the first and third CpG site which allows one to infer that the first and third CpG sites in the original cfDNA molecule are methylated. Whereas, the second CpG site can be read as a thymine (U is converted to T during the sequencing process), and thus, one can infer that the second CpG site is unmethylated in the original cfDNA molecule. With these two pieces of information, the methylation status and location, the analytics system generates 260 a methylation state vector 252 for the fragment cfDNA 212. In this example, the resulting methylation state vector 252 is < M23, U24, M25 >, wherein M corresponds to a methylated CpG site, U corresponds to an unmethylated CpG site, and the subscript number corresponds to a position of each CpG site in the reference genome.

[0121] One or more alternative sequencing methods can be used for obtaining sequence reads from nucleic acids in a biological sample. The one or more sequencing methods can comprise any form of sequencing that can be used to obtain a number of sequence reads measured from nucleic acids (e.g.. cell-free nucleic acids), including, but not limited to, high-throughput sequencing systems such as the Roche 454 platform, the Applied Biosystems SOLID platform, the Helicos True Single Molecule DNA sequencing technology, the sequencing-by -hybridization platform from Affymetrix Inc., the single-molecule, realtime (SMRT) technology of Pacific Biosciences, the sequencing-by-synthesis platforms from 454 Life Sciences, Illumina / Solexa and Helicos Biosciences, and the sequencing-by-ligation platform from Applied Biosystems. The ION TORRENT technology from Life technologies and Nanopore sequencing can also be used to obtain sequence reads from the nucleic acids (e.g, cell-free nucleic acids) in the biological sample. Sequencing-by-synthesis and reversible terminator-based sequencing (e g., Illumina's Genome Analyzer; Genome Analyzer II; HISEQ 2000; HISEQ 4500 (Illumina, San Diego Calif.)) can be used to obtain sequence reads from the cell-free nucleic acid obtained from a biological sample of a training subject in order to form the genoty pic dataset. Millions of cell-free nucleic acid (e.g., DNA) fragments can be sequenced in parallel. In one example of this type of sequencing technology, a flow cell is used that contains an optically transparent slide with eight subject lanes on the surfaces of which are bound oligonucleotide anchors (e g., adapter primers). A cell-free nucleic acid sample can include a signal or tag that facilitates detection. The acquisition of sequence reads from the cell-free nucleic acid obtained from the biological sample can include obtaining quantification information of the signal or tag via a variety of techniques such as, for example, flow cytometry', quantitative polymerase chain reactionAtty. Docket No. 32917-65573 / WO 29 P0252-WO(qPCR), gel electrophoresis, gene-chip analysis, microarray, mass spectrometry', cytofluorimetric analysis, fluorescence microscopy, confocal laser scanning microscopy, laser scanning cytometry, affinity chromatography, manual batch mode separation, electric field suspension, sequencing, and combination thereof.

[0122] The one or more sequencing methods can comprise a whole-genome sequencing assay. A whole-genome sequencing assay can comprise a physical assay that generates sequence reads for a whole genome or a substantial portion of the whole genome which can be used to determine large variations such as copy number variations or copy number aberrations. Such a physical assay may employ whole-genome sequencing techniques or whole-exome sequencing techniques. A whole-genome sequencing assay can have an average sequencing depth of at least lx, 2x. 3x. 4x. 5x. 6x, 7x, 8x, 9x, lOx, at least 20x, at least 30x, or at least 40x across the genome of the test subject. In some embodiments, the sequencing depth is about 30,000x. The one or more sequencing methods can comprise a targeted panel sequencing assay. A targeted panel sequencing assay can have an average sequencing depth of at least 50,000x, at least 55,000x, at least 60.000x. or at least 70,000x sequencing depth for the targeted panel of genes. The targeted panel of genes can comprise between 450 and 500 genes. The targeted panel of genes can comprise a range of 500±5 genes, a range of 500±10 genes, or a range of 500±25 genes.

[0123] The one or more sequencing methods can comprise paired-end sequencing. The one or more sequencing methods can generate a plurality of sequence reads. The plurality of sequence reads can have an average length ranging between 10 and 700, between 50 and 400, or between 100 and 300. The one or more sequencing methods can comprise a methylation sequencing assay. The methylation sequencing can be i) whole-genome methylation sequencing or ii) targeted DNA methylation sequencing using a plurality’ of nucleic acid probes. For example, the methylation sequencing is whole-genome bisulfite sequencing (e.g. , WGBS). The methylation sequencing can be a targeted DNA methylation sequencing using a plurality of nucleic acid probes targeting the most informative regions of the methylome, a unique methylation database and prior prototype whole-genome and targeted sequencing assays.

[0124] The methylation sequencing can detect one or more 5-methylcytosine (5mC) and / or 5-hydroxymethylcytosine (5hmC) in respective nucleic acid methylation fragments. The methylation sequencing can comprise conversion of one or more unmethylated cytosines or one or more methylated cytosines, in respective nucleic acid methylation fragments, to a corresponding one or more uracils. The one or more uracils can be detected during theAtty. Docket No. 32917-65573 / WO 30 P0252-WOmethylation sequencing as one or more corresponding thymines. The conversion of one or more unmethylated cytosines or one or more methylated cytosines can comprise a chemical conversion, an enzymatic conversion, or combinations thereof.

[0125] For example, bisulfite conversion involves converting cytosine to uracil while leaving methylated cytosines (e.g., 5-methylcytosine or 5-mC) intact. In some DNA, about 95% of cytosines may not be methylated in the DNA, and the resulting DNA fragments may include many uracils which are represented by thymines. Enzymatic conversion processes may be used to treat the nucleic acids prior to sequencing, which can be performed in various ways. One example of a bisulfite-free conversion comprises a bisulfite-free and baseresolution sequencing method, TET-assisted pyridine borane sequencing (TAPS), for nondestructive and direct detection of 5-methylcytosine and 5-hydroxymethylcytosine without affecting unmodified cytosines. The methylation state of a CpG site in the corresponding plurality of CpG sites in the respective nucleic acid methylation fragment can be methylated when the CpG site is determined by the methylation sequencing to be methylated, and unmethylated when the CpG site is determined by the methylation sequencing to not be methylated.

[0126] A methylation sequencing assay (e.g. , WGBS and / or targeted methylation sequencing) can have an average sequencing depth including but not limited to up to about l,000x. 2,000x, 3,000x, 5,000x, 10,000x, 15,000x, 20,000x, or 30,000x. The methylation sequencing can have a sequencing depth that is greater than 30.000x. e.g., at least 40,000x or 50,000x. A whole-genome bisulfite sequencing method can have an average sequencing depth of between 20x and 50x, and a targeted methylation sequencing method has an average effective depth of between lOOx and lOOOx, where effective depth can be the equivalent whole-genome bisulfite sequencing coverage for obtaining the same number of sequence reads obtained by targeted methylation sequencing. The methylation sequencing assay may comprise probes targeting 500 or more CpG sites, 1,000 or more CpG sites, 1,500 or more CpG sites, 2,000 or more CpG sites, 2,500 or more CpG sites, 3,000 or more CpG sites, 3,500 or more CpG sites, 4,000 or more CpG sites, 4,500 or more CpG sites, 5,000 or more CpG ,000 or more CpG sites, sites, 7,000 or more CpG sites, 8.000 or more CpG sites. 9,000 or more CpG sites, 10,000 or more CpG sites, 20,000 or more CpG sites, 30,000 or more CpG sites, 40,000 or more CpG sites, 50,000 or more CpG sites, 60,000 or more CpG sites, 70,000 or more CpG sites, 80,000 or more CpG sites, 90,000 or more CpG sites, 100,000 or more CpG sites, 150,000 or more CpG sites, etc.Atty. Docket No. 32917-65573 / WO 31 P0252-WO

[0127] For further details regarding methylation sequencing (e.g., WGBS and / or targeted methylation sequencing), see. e.g., United States Patent Application No. 16 / 719,902, entitled “Systems and Methods for Estimating Cell Source Fractions Using Methylation Information,” filed December 18, 2019, 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 fragment methylation patterns. A methylation sequencing can be used to identify one or more methylation state vectors, as described, for example, in United States Patent Application No.16 / 352,602, entitled “Anomalous Fragment Detection and Classification,” filed March 13, 2019, or in accordance with any of the techniques disclosed in United States Patent Application No. 15 / 931,022. entitled “Model-Based Featurization and Classification,” filed May 13, 2020, each of which is hereby incorporated by reference.

[0128] The methylation sequencing of nucleic acids and the resulting one or more methylation state vectors can be used to obtain a plurality of nucleic acid methylation fragments. Each corresponding plurality of nucleic acid methylation fragments (e.g., for each respective genotypic dataset) can comprise more than 100 nucleic acid methylation fragments. An average number of nucleic acid methylation fragments can comprise 1,000 or more nucleic acid methylation fragments, 5,000 or more nucleic acid methylation fragments, 10,000 or more nucleic acid methylation fragments, 20,000 or more nucleic acid methylation fragments, 30,000 or more nucleic acid methylation fragments, 40,000 or more nucleic acid methylation fragments, or 50,000 or more nucleic acid methylation fragments. An average number of nucleic acid methylation fragments across each corresponding plurality of nucleic acid methylation fragments can be between 10,000 nucleic acid methylation fragments and 50,000 nucleic acid methylation fragments. The corresponding plurality of nucleic acid methylation fragments can comprise one thousand or more, ten thousand or more, 100 thousand or more, one million or more, ten million or more, 100 million or more, 500 million or more, one billion or more, two billion or more, three billion or more, four billion or more, five billion or more, six billion or more, seven billion or more, eight billion or more, nine billion or more, or 10 billion or more nucleic acid methylation fragments. An average length of a corresponding plurality of nucleic acid methylation fragments can be between 140 and 480 nucleotides. An average length of a corresponding plurality of nucleic acid methylation fragments can be between 100 and 600 nucleotides. An average length of a corresponding plurality of nucleic acid methylation fragments can be between 50 and 750 nucleotides. AnAtty. Docket No. 32917-65573 / WO 32 P0252-WOaverage length of a corresponding plurality of nucleic acid methylation fragments can be between 30 and 1,000 nucleotides.

[0129] Further details regarding methods for sequencing nucleic acids and methylation sequencing data are disclosed in U.S. Patent Application No. 17 / 191,914, titled “Systems and Methods for Cancer Condition Determination Using Autoencoders,’' filed March 4, 2021. which is hereby incorporated herein by reference in its entirety.III. CANCER CLASSIFICATION FOR DETERMINING CANCER

[0130] Cancer classification involves extracting genetic features and applying one or more models to the extracted features to determine a cancer prediction. The analytics system aggregates extracted features into a feature vector which can then be input into a trained cancer prediction model to determine a cancer prediction based on the input feature vector. The cancer prediction may comprise one or more labels and / or one or more values. One label may be binary7, indicating a presence or absence of cancer in the test subject. Another label may be multiclass, indicating one or more particular cancer signal origins from a plurality of screened cancer signal origins. One value may indicate a likelihood of presence of cancer. Another value may indicate a likelihood of absence of cancer. Yet another value may otherwise indicate another prognosis of the cancer. For example, the value may quantify a progression and / or potential response to treatment of the cancer.

[0131] In one or more embodiments, the feature vectors input into the cancer classifier are based on a set of informative fragments (also referred to as “anomalously methylated” or “unusual fragments of extreme methylation” (UFXM)) determined from the test sample.

[0132] In some embodiments, a cancer classifier may be a machine-learned model which is a computation model comprising a plurality of classification parameters and a function representing a relation between the feature vector as input and the cancer prediction as output. Inputting the feature vector into the function with the classification parameters yields the cancer prediction. The machine-learned model may be trained using training samples derived from subjects with known cancer diagnoses. The training samples may be divided into cohorts of varying labels. For example, there may be a cohort of training samples for each cancer signal origin.

[0133] A machine-learned model may be trained through any combination of machine-learning techniques including, but not limited to, Linear Regression, Logistic Regression, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Decision TreesAtty. Docket No. 32917-65573 / WO 33 P0252-WO(e.g., ID3, C4.5, CART), Random Forest, Neural Networks (e.g., Multi-Layer Perceptron, Convolutional Neural Networks. Recurrent Neural Networks), Naive Bayes Classifier.Gradient Boosting Machines (e.g., XGBoost, LightGBM, CatBoost), AdaBoost, K-Means Clustering, Hierarchical Clustering, DBSCAN (Density-Based Spatial Clustering of Applications with Noise), Gaussian Mixture Models (GMM), PCA (Principal Component Analysis), t-SNE (t-Distributed Stochastic Neighbor Embedding), UMAP (Uniform Manifold Approximation and Projection), ICA (Independent Component Analysis), Autoencoders (e.g.. Variational Autoencoders, Denoising Autoencoders), Self-Organizing Maps (SOM), Q-Leaming, Deep Q Network (DQN), SARSA (State-Action-Reward-State-Action), Monte Carlo Methods, Temporal Difference Learning (TD Learning), Policy Gradients (e g., REINFORCE, Actor-Critic). Proximal Policy Optimization (PPO), Advantage Actor-Critic (A2C), Soft Actor-Critic (SAC), Deep Deterministic Policy Gradient (DDPG), Hidden Markov Models (HMM), Conditional Random Fields (CRF), Latent Dirichlet Allocation (LDA), Restricted Boltzmann Machines (RBM), Genetic Algorithms, and Swarm Intelligence Algorithms (e.g.. Particle Swarm Optimization, Ant Colony Optimization). Expectation Maximization (EM).III. A. IDENTIFYING INFORMATIVE FRAGMENTS

[0134] The analytics system can determine informative fragments for a sample using the sample’s methylation state vectors. For each fragment in a sample, the analytics system can determine whether the fragment is an informative fragment using the methylation state vector corresponding to the fragment. In some embodiments, the analytics system calculates a p-value score for each methylation state vector describing a probability7of observing that methylation state vector or other methylation state vectors even less probable in the healthy control group. The process for calculating a p-value score is further discussed below in Section III. A.i. P-Value Filtering. The analytics system may determine fragments with a methylation state vector having below a threshold p-value score as informative fragments. In some embodiments, the analytics system further labels fragments with at least some number of CpG sites that have over some threshold percentage of methylation or unmethylation as hypermethylated and hypomethylated fragments, respectively. A hypermethylated fragment or a hypomethylated fragment may also be referred to as an unusual fragment with extreme methylation (UFXM). In other embodiments, the analytics system may implement various other probabilistic models for determining informative fragments. Examples of other probabilistic models include a mixture model, a deep probabilistic model, etc. In some embodiments, the analytics system may use any combination of the processes describedAtty. Docket No. 32917-65573 / WO 34 P0252-WObelow for identifying informative fragments. With the identified informative fragments, the analytics system may filter the set of methylation state vectors for a sample for use in other processes, e.g., for use in training and deploying a cancer classifier.III.A.i. P-VALUE FILTERING

[0135] In some embodiments, the analytics system calculates a p-value score for each methylation state vector compared to methylation state vectors from fragments in a healthy control group. The p-value score can describe a probability of observing the methylation status matching that methylation state vector or other methylation state vectors even less probable in the healthy control group. In order to determine a DNA fragment to be anomalously methylated, the analytics system can use a healthy control group with a majority of fragments that are normally methylated. When conducting this probabilistic analysis for determining informative fragments, the determination can hold weight in comparison with the group of control subjects that make up the healthy control group. To bolster robustness in the healthy control group, the analytics system may select some threshold number of healthy subjects to source samples including DNA fragments. FIG. 3 A below describes the method of generating a data structure for a healthy control group with which the analytics system may calculate p-value scores. FIG. 3B describes the method of calculating a p-value score with the generated data structure.

[0136] FIG. 3A is a flowchart describing a process 300 of generating a data structure for a healthy control group, according to an embodiment. To create a healthy control group data structure, the analytics system can receive a plurality of DNA fragments (e g., cfDNA) from a plurality of healthy subjects. The analytics system can generate 305 a methylation state vector for each fragment, for example via the process 200 of FIG. 2A.

[0137] With each fragment's methylation state vector, the analytics system can subdivide 310 the methylation state vector into strings of CpG sites. In some embodiments, the analytics system subdivides 310 the methylation state vector such 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 would result 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 3 being subdivided into strings of length less than or equal to 4 can result 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 the same length as the specified string length, then the methylation state vector may be converted into a single string containing all of the CpG sites of the vector.Atty. Docket No. 32917-65573 / WO 35 P0252-WO

[0138] The analytics system tallies 315 the strings by counting, for each possible CpG site and possibility of methylation states in the vector, the number of strings present in the control group having the specified CpG site as the first CpG site in the string and having that possibility of methylation states. For example, at a given CpG site and considering string lengths of 3, there are 2A3 or 8 possible string configurations. At that given CpG site, for each of the 8 possible string configurations, the analytics system tallies 310 how many occurrences of each methylation state vector possibility come up in the control group.Continuing this example, this may involve tallying the following quantities: < Mx, Mx+i, Mx+2 >, < Mx, Mx+i, Ux+2 >, . . ., < Ux, Ux+i, Ux+2 > for each starting CpG site in the reference genome. The analytics system creates 315 the data structure storing the tallied counts for each starting CpG site and string possibility.

[0139] There are several benefits to setting an upper limit on string length. First, depending on the maximum length for a string, the size of the data structure created by the analytics system can dramatically increase in size. For instance, a maximum string length of 4 means that every CpG site has at the very least 2A4 numbers to tally for strings of length 4. Increasing the maximum string length to 5 means that every' CpG site has an additional 2A4 or 16 numbers to tally, doubling the numbers to tally (and computer memory required) compared to the prior string length. Reducing string size can help keep the data structure creation and performance (e.g., use for later accessing as described below), in terms of computational and storage, reasonable. Second, a statistical consideration to limiting the maximum string length can be to avoid overfitting downstream models that use the string counts. If long strings of CpG sites do not, biologically, have a strong effect on the outcome (e.g., predictions of anomalousness that predictive of the presence of cancer), calculating probabilities based on large strings of CpG sites can be problematic as it uses a significant amount of data that may not be available, and thus can be too sparse for a model to perform appropriately. For example, calculating a probability of anomalousness / cancer conditioned on the prior 100 CpG sites can use counts of strings in the data structure of length 100, ideally some matching exactly the prior 100 methylation states. If only sparse counts of strings of length 100 are available, there can be insufficient data to determine whether a given string of length of 100 in a test sample is anomalous or not.

[0140] FIG. 3B is a flowchart describing a process 330 for identifying anomalously methylated fragments from a subject, according to an embodiment. In process 330, the analytics system generates 340 methylation state vectors from cfDNA fragments of theAtty. Docket No. 32917-65573 / WO 36 P0252-WOsubject, e.g., via the process 200 of FIG. 2A. The analytics system can handle each methylation state vector as follows.

[0141] For a given methylation state vector, the analytics system enumerates 345 all possibilities of methylation state vectors having the same starting CpG site and same length (i.e., set of CpG sites) in the methylation state vector. As each methylation state is generally either methylated or unmethylated there can be effectively two possible states at each CpG site, and thus the count of distinct possibilities of methylation state vectors can depend on a power of 2, such that a methylation state vector of length n would be associated with 2npossibilities of methylation state vectors. With methylation state vectors inclusive of indeterminate states for one or more CpG sites, the analytics system may enumerate 330 possibilities of methylation state vectors considering only CpG sites that have observed states.

[0142] The analytics system calculates 350 the probability of observing each possibility7of methylation state vector for the identified starting CpG site and methylation state vector length by accessing the healthy control group data structure. In some embodiments, calculating the probability of observing a given possibility uses a Markov chain probability to model the joint probability calculation. The Markov model can be trained, at least in part, based upon evaluation of a methylation state of each CpG site in the corresponding plurality of CpG sites of the respective fragment (e.g., nucleic acid methylation fragment) across those nucleic acid methylation fragments in a healthy7noncancer cohort dataset that have the corresponding plurality of CpG sites. For example, a Markov model (e.g., a Hidden Markov Model or HMM) is used to determine the probability that a sequence of methylation states (comprising, e.g., “M"’ or “U’") can be observed for a nucleic acid methylation fragment in a plurality of nucleic acid methylation fragments, given a set of probabilities that determine, for each state in the sequence, the likelihood of observing the next state in the sequence. The set of probabilities can be obtained by training the HMM. Such training can involve computing statistical parameters (e.g., the probability that a first state can transition to a second state (the transition probability) and / or the probability that a given methylation state can be observed for a respective CpG site (the emission probability)), given an initial training dataset of observed methylation state sequences (e.g., methylation patterns). HMMs can be trained using supervised training (e.g., using samples where the underlying sequence as well as 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.Atty. Docket No. 32917-65573 / WO 37 P0252-WOcalculation methods other than Markov chain probabilities are used to determine the probability of observing each possibility of methylation state vector. For example, such calculation method can include a learned 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.

[0143] The analytics system calculates 355 a p-value score for the methylation state vector using the calculated probabilities for each possibility. In some embodiments, this includes identifying the calculated probability corresponding to the possibility that matches the methylation state vector in question. Specifically, this can be the possibility having the same set of CpG sites, or similarly the same starting CpG site and length as the methylation state vector. The analytics system can sum the calculated probabilities of any possibilities having probabilities less than or equal to the identified probability to generate the p-value score.

[0144] This p-value can represent the probability of observing the methylation state vector of the fragment or other methylation state vectors even less probable in the healthy control group. A low p-value score can, thereby, generally correspond to a methylation state vector which is rare in a healthy subject, and which causes the fragment to be labeled anomalously methylated, relative to the healthy control group. A high p-value score can generally relate to a methylation state vector that is expected to be present, in a relative sense, in a healthy subject. If the healthy control group is a non-cancerous group, for example, a low p-value can indicate that the fragment is anomalous methylated relative to the non-cancer group, and therefore possibly indicative of the presence of cancer in the test subject.

[0145] As above, the analytics system can calculate p-value scores for each of a plurality of methylation state vectors, each representing a cfDNA fragment in the test sample. To identify which of the fragments are anomalously methylated, the analytics system may filter 365 the set of methylation state vectors based on their p-value scores. In some embodiments, filtering is performed by comparing the p-values scores against a threshold and keeping 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 similar.

[0146] According to example results from the process 300, the analytics system can yield a median (range) of 2,800 (1,500-12,000) fragments with anomalous methylation patterns for participants without cancer in training, and a median (range) of 3,000 (1,200-420,000) fragments with anomalous methylation patterns for participants with cancer inAtty. Docket No. 32917-65573 / WO 38 P0252-WOtraining. These filtered sets of fragments with anomalous methylation patterns may be used for the downstream analyses as described below in Sections III.B & III.C.

[0147] In some embodiments, the analytics system uses 360 a sliding window to determine possibilities of methylation state vectors and calculate p-values. Rather than enumerating possibilities and calculating p-values for entire methylation state vectors, the analytics system can enumerate possibilities and calculates p-values for only a window of sequential CpG sites, where the window is shorter in length (of CpG sites) than at least some fragments (otherwise, the window w ould serve no purpose). The window length may be static, user determined, dynamic, or otherwise selected.

[0148] In calculating p-values for a methylation state vector larger than the w indow, the window can identify the sequential set of CpG sites from the vector within the window starting from the first CpG site in the vector. The analytics system can calculate a p-value score for the window' including the first CpG site. The analytics system can then “slide” the window' to the second CpG site in the vector, and calculates another p-value score for the second window. Thus, for a window size I and methylation vector length m, each methylation state vector can generate m-l+1 p-value scores. After completing the p-value calculations for each portion of the vector, the lowest p-value score from all sliding windows can be taken as the overall p-value score for the methylation state vector. In other embodiments, the analytics system aggregates the p-value scores for the methylation state vectors to generate an overall p-value score.

[0149] Using the sliding window can help to reduce the number of enumerated possibilities of methylation state vectors and their corresponding probability calculations that would otherwise need to be performed. To give a realistic example, it can be for fragments to have upwards of 54 CpG sites. Instead of computing probabilities for 2A54 (~1.8xlOA16) possibilities to generate a single p-score, the analytics system can instead use a window of size 5 (for example) which results in 50 p-value calculations for each of the 50 windows of the methylation state vector for that fragment. Each of the 50 calculations can enumerate 2A5 (32) possibilities of methylation state vectors, which total results in 50x2A5 (1.6xlOA3) probability calculations. This can result in a vast reduction of calculations to be performed, with no meaningful hit to the accurate identification of informative fragments.

[0150] In embodiments with indeterminate states, the analytics system may calculate a p-value score summing out CpG sites with indeterminates states in a fragment’s methylation state vector. The analytics system can identify all possibilities that have consensus with the all methylation states of the methylation state vector excluding theAtty. Docket No. 32917-65573 / WO 39 P0252-WOindeterminate states. The analytics system may assign the probability to the methylation state vector as a sum of the probabilities of the identified possibilities. As an example, the analytics system can calculate a probability of a methylation state vector of < Mi, h, U3 > as a sum of the probabilities for the possibilities of methylation state vectors of < Mi, M2, U3 > and < Mi, U2, U3 > since methylation states for CpG sites 1 and 3 are observed and in consensus with the fragment’s methylation states at CpG sites 1 and 3. This method of summing out CpG sites with indeterminate states can use calculations of probabilities of possibilities up to 2Ai, wherein i denotes the number of indeterminate states in the methylation state vector. In additional embodiments, a dynamic programming algorithm may be implemented to calculate the probability of a methylation state vector with one or more indeterminate states. Advantageously, the dynamic programming algorithm operates in linear computational time.

[0151] In some embodiments, the computational burden of calculating probabilities and / or p-value scores may be further reduced by caching at least some calculations. For example, the analytics system may cache in transitory or persistent memory’ calculations of probabilities for possibilities of methylation state vectors (or windows thereof). If other fragments have the same CpG sites, caching the possibility probabilities can allow for efficient calculation of p-score values without needing to re-calculate the underlying possibility probabilities. Equivalently, the analytics system may calculate p-value scores for each of the possibilities of methylation state vectors associated with a set of CpG sites from vector (or window thereof). The analytics system may cache the p-value scores for use in determining the p-value scores of other fragments including the same CpG sites. Generally, the p-value scores of possibilities of methylation state vectors having the same CpG sites may¬ be used to determine the p-value score of a different one of the possibilities from the same set of CpG sites.

[0152] One or more nucleic acid methylation fragments can be fdtered prior to training region models or cancer classifier. Filtering nucleic acid methylation fragments can comprise removing, from the corresponding plurality of nucleic acid methylation fragments, each respective nucleic acid methylation fragment that fails to satisfy one or more selection criteria (e.g., below or above one selection criteria). The one or more selection criteria can comprise a p-value threshold. The output p-value of the respective nucleic acid methylation fragment can be determined, at least in part, based upon a comparison of the corresponding methylation pattern of the respective nucleic acid methylation fragment to a corresponding distribution of methylation patterns of those nucleic acid methylation fragments in a healthyAtty. Docket No. 32917-65573 / WO 40 P0252-WOnoncancer cohort dataset that have the corresponding plurality' of CpG sites of the respective nucleic acid methylation fragment.

[0153] Filtering a plurality of nucleic acid methylation fragments can comprise removing each respective nucleic acid methylation fragment that fails to satisfy a p-value threshold. The filter can be applied to the methylation pattern of each respective nucleic acid methylation fragment using the methylation patterns observed across the first plurality of nucleic acid methylation fragments. Each respective methylation patern of each respective nucleic acid methylation fragment (e.g. , Fragment One, ... , Fragment N) can comprise a corresponding one or more methylation sites (e.g., CpG sites) identified with a methylation site identifier and a corresponding methylation patern, represented as a sequence of 1 's and 0's, where each “1” represents a methylated CpG site in the one or more CpG sites and each “0” represents an unmethylated CpG site in the one or more CpG sites. The methylation paterns observed across the first plurality of nucleic acid methylation fragments can be used to build a methylation state distribution for the CpG site states collectively represented by the first plurality of nucleic acid methylation fragments (e.g. , CpG site A, CpG site B, ... , CpG site ZZZ). Further details regarding processing of nucleic acid methylation fragments are disclosed in U.S. Patent Application No. 17 / 191,914, titled ‘‘Systems and Methods for Cancer Condition Determination Using Autoencoders,” filed March 4, 2021, which is hereby incorporated herein by reference in its entirety .

[0154] The respective nucleic acid methylation fragment may fail to satisfy a selection criterion in the one or more selection criteria when the respective nucleic acid methylation fragment has an anomalous methylation score that is less than an anomalous methylation score threshold. In this situation, the anomalous methylation score can be determined by a mixture model. For example, a mixture model can detect an anomalous methylation patern in a nucleic acid methylation fragment by determining the likelihood of a methylation state vector (e.g., a methylation patern) for the respective nucleic acid methylation fragment based on the number of possible methylation state vectors of the same length and at the same corresponding genomic location. This can be executed by generating a plurality of possible methylation states for vectors of a specified length at each genomic location in a reference genome. Using the plurality of possible methylation states, the number of total possible methylation states and subsequently the probability of each predicted methylation state at the genomic location can be determined. The likelihood of a sample nucleic acid methylation fragment corresponding to a genomic location within the reference genome can then be determined by matching the sample nucleic acid methylation fragment toAtty. Docket No. 32917-65573 / WO 41 P0252-WOa predicted (e.g., possible) methylation state and retrieving the calculated probability of the predicted methylation state. An anomalous methylation score can then be calculated based on the probability of the sample nucleic acid methylation fragment.

[0155] The respective nucleic acid methylation fragment can fail to satisfy a selection criterion in the one or more selection criteria when the respective nucleic acid methylation fragment has less than a threshold number of residues. 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. The respective nucleic acid methylation fragment may fail to satisfy a selection criterion in the one or more selection criteria when the respective nucleic acid methylation fragment has less than a threshold number of CpG sites. The threshold number of CpG sites can be 4. 5, 6, 7. 8, 9, or 10. The respective nucleic acid methylation fragment can fail to satisfy a selection criterion in the one or more selection criteria when a genomic start position and a genomic end position of the respective nucleic acid methylation fragment indicates that the respective nucleic acid methylation fragment represents less than a threshold number of nucleotides in a human genome reference sequence.

[0156] The filtering can remove a nucleic acid methylation fragment in the corresponding plurality of nucleic acid methylation fragments that has the same corresponding methylation pattern and the same corresponding genomic start position and genomic end position as another nucleic acid methylation fragment in the corresponding plurality of nucleic acid methylation fragments. This filtering step can remove redundant fragments that are exact duplicates, including, in some instances, PCR duplicates. The filtering can remove a nucleic acid methylation fragment that has the same corresponding genomic start position and genomic end position and less than a threshold number of different methylation states as another nucleic acid methylation fragment in the corresponding plurality of nucleic acid methylation fragments. The threshold number of different methylation states used for retention of a nucleic acid methylation fragment can be 1, 2, 3, 4, 5, or more than 5. For example, a first nucleic acid methylation fragment having the same corresponding genomic start and end position as a second nucleic acid methylation fragment but having at least 1, at least 2, at least 3, at least 4, or at least 5 different methylation states at a respective CpG site (e.g., aligned to a reference genome) is retained. As another example, a first nucleic acid methylation fragment having the same methylation state vector (e.g, methylation pattern) but different corresponding genomic start and end positions as a second nucleic acid methylation fragment is also retained.Atty. Docket No. 32917-65573 / WO 42 P0252-WO

[0157] The filtering can remove assay artifacts in the plurality' of nucleic acid methylation fragments. The removal of assay artifacts can comprise removing sequence reads obtained from sequenced hybridization probes and / or sequence reads obtained from sequences that failed to undergo conversion during bisulfite conversion. The filtering can remove contaminants (e.g., due to sequencing, nucleic acid isolation, and / or sample preparation).

[0158] The filtering can remove a subset of methylation fragments from the plurality of methylation fragments based on mutual information filtering of the respective methylation fragments against the cancer state across the plurality of training subjects. For example, mutual information can provide a measure of the mutual dependence between two conditions of interest sampled simultaneously. Mutual information can be determined by selecting an independent set of CpG sites (e.g. , within all or a portion of a nucleic acid methylation fragment) from one or more datasets and comparing the probability of the methylation states for the set of CpG sites between two sample groups (e.g., subsets and / or groups of genotypic datasets, biological samples, and / or subjects). A mutual information score can denote the probability of the methylation pattern for a first condition versus a second condition at the respective region in the respective frame of the sliding window, thus indicating the discriminative power of the respective region. A mutual information score can be similarly calculated for each region in each frame of the sliding window as it progresses across the selected sets of CpG sites and / or the selected genomic regions. Further details regarding mutual information filtering are disclosed in U.S. Patent Application 17 / 119,606, titled “Cancer Classification using Patch Convolutional Neural Networks,” filed December 11, 2020, which is hereby incorporated herein by reference in its entirety.III. A.n. HYPERMETHYLATED FRAGMENTS AND HYPOMETHYLATED FRAGMENTS

[0159] In some embodiments, the analytics system identifies 770 hypomethylated fragments or hypermethylated fragments from the filtered set as informative fragments. The analytics system identifies hypermethylated fragments having over a threshold number of CpG sites and over a threshold percentage of the CpG sites methylated. The analytics system identifies hypomethylated fragments having over the threshold number of CpG sites and over a threshold percentage of CpG sites unmethylated. Example thresholds for length of fragments (or CpG sites) include more than 3, 4, 5, 6, 7, 8, 9, 10, etc. Example percentage thresholds of methylation or unmethylation include more than 80%, 85%, 90%, or 95%, or any other percentage within the range of 50%-100%.Atty. Docket No. 32917-65573 / WO 43 P0252-WOIII.B. TRAINING OF CANCER CLASSIFIER

[0160] FIG. 4A is a flowchart describing a process 400 of training a cancer classifier, according to an embodiment. The analytics system obtains 410 a plurality of training samples each having a set of informative fragments and a label of a cancer signal origin. The plurality7of training samples can include any combination of samples from healthy subjects with a general label of “non-cancer,” samples from subjects with a general label of “cancer” or a specific label (e.g., “breast cancer,” “lung cancer,” etc.). The training samples from subjects for one cancer signal origin may be termed a cohort for that cancer signal origin or a cancer signal origin cohort.

[0161] The analytics system determines 420, for each training sample, a feature vector based on the set of informative fragments of the training sample. The analytics system can calculate an anomaly score for each CpG site in an initial set of CpG sites. The initial set of CpG sites may be all CpG sites in the human genome or some portion thereof - which may be on the order of 104, 105, 106, 107, 108, etc. In one embodiment, the analytics system defines the anomaly score for the feature vector with a binary scoring based on whether there is an informative fragment in the set of informative fragments that encompasses the CpG site. In another embodiment, the analytics system defines the anomaly score based on a count of informative fragments overlapping the CpG site. In one example, the analytics system may use a trinary scoring assigning a first score for lack of presence of informative fragments, a second score for presence of a few informative fragments, and a third score for presence of more than a few informative fragments. For example, the analytics system counts 5 informative fragment in a sample that overlap the CpG site and calculates an anomaly score based on the count of 5.

[0162] Once all anomaly scores are determined for a training sample, the analytics system can determine the feature vector as a vector of elements including, for each element, one of the anomaly scores associated with one of the CpG sites in an initial set. The analytics system can normalize the anomaly scores of the feature vector based on a coverage of the sample. Here, coverage can refer to a median or average sequencing depth over all CpG sites covered by the initial set of CpG sites used in the classifier, or based on the set of informative fragments for a given training sample.

[0163] As an example, reference is now made to FIG. 4B illustrating a matrix of training feature vectors 422. In this example, the analytics system has identified CpG sites [K] 426 for consideration in generating feature vectors for the cancer classifier. The analytics system selects training samples [N] 424. The analytics system determines a first anomalyAtty. Docket No. 32917-65573 / WO 44 P0252-WOscore 428 for a first arbitrary CpG site [kl] to be used in the feature vector for a training sample [nl], The analytics system checks each informative fragment in the set of informative fragments. If the analytics system identifies at least one informative fragment that includes the first CpG site, then the analytics system determines the first anomaly score 428 for the first CpG site as 1, as illustrated in FIG. 4B. Considering a second arbitrary CpG site [k2], the analytics system similarly checks the set of informative fragments for at least one that includes the second CpG site [k2]. If the analytics system does not find any such informative fragment that includes the second CpG site, the analytics system determines a second anomaly score 429 for the second CpG site [k2] to be 0, as illustrated in FIG. 4B. Once the analytics system determines all the anomaly scores for the initial set of CpG sites, the analytics system determines the feature vector for the first training sample [nl] including the anomaly scores with the feature vector including the first anomaly score 428 of 1 for the first CpG site [kl] and the second anomaly score 429 of 0 for the second CpG site [k2] and subsequent anomaly scores, thus forming a feature vector [1, 0, ...].

[0164] Additional approaches to featurization of a sample can be found in: 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. The various featurization approaches may generate distinct features to be included in a sample’s feature vector.

[0165] In some embodiments, each classifier may be trained on different matrices of training feature vectors. For example, one classifier may be trained on a first matrix covering a first set of genomic regions, whereas another classifier may be trained on a second matrix covering a second, differing set of genomic regions. In another example, one classifier may be trained on a first matrix with features determined according to one featurization approach, whereas another classifier may be trained on a second matrix with features determined according to another featurization approach.

[0166] The analytics system may further limit the CpG sites considered for use in the cancer classifier. The analytics system computes 430, for each CpG site in the initial set of CpG sites, an information gain based on the feature vectors of the training samples. From step 420. each training sample has a feature vector that may contain an anomaly score all CpG sites in the initial set of CpG sites which could include up to all CpG sites in the humanAtty. Docket No. 32917-65573 / WO 45 P0252-WOgenome. However, some CpG sites in the initial set of CpG sites may not be as informative as others in distinguishing between cancer signal origins, or may be duplicative with other CpG sites.

[0167] In one embodiment, the analytics system computes 430 an information gain for each cancer signal origin and for each CpG site in the initial set to determine whether to include that CpG site in the classifier. The information gain is computed for training samples with a given cancer signal origin compared to all other samples. For example, two random variables ‘informative fragment’ (‘IF’) and ‘cancer signal’ (‘CS’) are used. In one embodiment, IF is a binary’ variable indicating whether there is an informative fragment overlapping a given CpG site in a given samples as determined for the anomaly score / feature vector above. CS is a random variable indicating whether the cancer signal has a particular origin, for example a particular organ or organ group, or a particular cancer biology7. The analytics system computes the mutual information with respect to CS given IF. That is, how many bits of information about the cancer signal origin are gained if it is known whether there is an informative fragment overlapping a particular CpG site. In practice, for a first cancer signal origin, the analytics system computes pairwise mutual information gain against each other cancer signal origin and sums the mutual information gain across all the other cancer signal origins.

[0168] For a given cancer signal origin, the analytics system can use this information to rank CpG sites based on how cancer specific they are. This procedure can be repeated for all cancer signal origins under consideration. If a particular region is commonly anomalously methylated in training samples of a given cancer but not in training samples of other cancer signal origins or in healthy training samples, then CpG sites overlapped by those informative fragments can have high information gains for the given cancer signal origin. The ranked CpG sites for each cancer signal origin can be greedily added (selected) 440 to a selected set of CpG sites based on their rank for use in the cancer classifier.

[0169] In additional embodiments, the analy tics system may consider other selection criteria for selecting informative CpG sites to be used in the cancer classifier. One selection criterion may be that the selected CpG sites are above a threshold separation from other selected CpG sites. For example, the selected CpG sites are to be over a threshold number of base pairs aw ay from any other selected CpG site (e.g., 100 base pairs), such that CpG sites that are w ithin the threshold separation are not both selected for consideration in the cancer classifier.Atty. Docket No. 32917-65573 / WO 46 P0252-WO

[0170] In one embodiment, according to the selected set of CpG sites from the initial set. the analytics system may modify 450 the feature vectors of the training samples as needed. For example, the analytics system may truncate feature vectors to remove anomaly scores corresponding to CpG sites not in the selected set of CpG sites.

[0171] With the feature vectors of the training samples, the analytics system may¬ train the cancer classifier in any of a number of ways. The feature vectors may correspond to the initial set of CpG sites from step 420 or to the selected set of CpG sites from step 450. In one embodiment, the analytics system trains 460 a binary cancer classifier to distinguish between cancer and non-cancer based on the feature vectors of the training samples. In this manner, the analytes system uses training samples that include both non-cancer samples from healthy subjects and cancer samples from subjects. Each training sample can have one of the two labels '‘cancer” or “non-cancer.” In this embodiment, the classifier outputs a cancer prediction indicating the likelihood of the presence or absence of cancer.

[0172] In another embodiment, the analytics system trains 470 a multiclass cancer classifier to distinguish between many cancer signal origins (also referred to as CSO labels). CSO can include one or more organs, organ groups, classes for cancer biology, cellular lineage of a cancer cell, or cancer drivers like a viral status. To do so, the analytics system can use the cancer signal origin cohorts and may also include or not include a non-cancer cohort. In this multiclass embodiment, the cancer classifier is trained to determine a cancer prediction (or, more specifically, a CSO prediction) that comprises a prediction value for each of the cancer signal origins being classified for. The prediction values may correspond to a likelihood that a given training sample (and during inference, a test sample) has each of the cancer signal origins. In one implementation, the prediction values are scored between 0 and 100, wherein the cumulation of the prediction values equals 100. For example, the cancer classifier returns a cancer prediction including a prediction value for the affected organ or organ group being breast, lung, and non-cancer. For example, the classifier can return a cancer prediction that a test samples signal origin is 65% likelihood of breast, 25% likelihood of lung, and 10% likelihood of non-cancer. The analytics system may further evaluate the prediction values to generate a prediction of a presence of one or more cancers in the sample, also may be referred to as a CSO prediction indicating one or more CSO labels, e.g., a first CSO prediction with the highest prediction value, a second CSO prediction with the second highest prediction value, etc. Continuing with the example above and given the percentages, in this example the system may determine that the sample's cancer signal origin is breast given that the CSO breast has the highest likelihood.Atty. Docket No. 32917-65573 / WO 47 P0252-WO

[0173] In both embodiments, the analytics system trains the cancer classifier by inputting sets of training samples with their feature vectors into the cancer classifier and adjusting classification parameters so that a function of the classifier accurately relates the training feature vectors to their corresponding label. The analytics system may group the training samples into sets of one or more training samples for iterative batch training of the cancer classifier. After inputting all sets of training samples including their training feature vectors and adjusting the classification parameters, the cancer classifier can be sufficiently trained to label test samples according to their feature vector within some margin of error. The analytics system may train the cancer classifier according to any one of a number of methods. As an example, the binary cancer classifier may be a L2-regularized logistic regression classifier that is trained using a log-loss function. As another example, the multicancer classifier may be a multinomial logistic regression. In practice either type of cancer classifier may be trained using other techniques. These techniques are numerous including potential use of kernel methods, random forest classifier, a mixture model, an autoencoder model, machine learning algorithms such as multilayer neural networks, etc. In some embodiments, such models may include 1,000 or more parameters, 2,000 or more parameters, 3,000 or more parameters, 4,000 or more parameters, 5,000 or more parameters, 10,000 or more parameters, 15,000 or more parameters, 20,000 or more parameters, 25,000 or more parameters, 50,000 or more parameters, 100,000 or more parameters, 150,000 or more parameters, 200,000 or more parameters, 250.000 or more parameters, 300,000 or more parameters, 350,000 or more parameters, 400,000 or more parameters, 450,000 or more parameters, or 500,000 or more parameters.

[0174] The classifier can include a logistic regression algorithm, a neural network algorithm, a support vector machine algorithm, a Naive Bayes algorithm, a nearest neighbor algorithm, a boosted trees algorithm, a random forest algorithm, a decision tree algorithm, a multinomial logistic regression algorithm, a linear model, or a linear regression algorithm.III. C. DEPLOYMENT OF CANCER CLASSIFIER

[0175] During use of the cancer classifier, the analytics system can obtain a test sample from a subject of unknown cancer signal origin. The analytics system may process the test sample comprised of DNA molecules with any combination of the processes 200 and 330 to achieve a set of informative fragments. The analytics system can determine a test feature vector for use by the cancer classifier according to similar principles discussed in the process 400. The analytics system can calculate an anomaly score for each CpG site in a plurality of CpG sites in use by the cancer classifier. For example, the cancer classifierAtty. Docket No. 32917-65573 / WO 48 P0252-WOreceives as input feature vectors inclusive of anomaly scores for 1,000 selected CpG sites. The analytics system can thus determine a test feature vector inclusive of anomaly scores for the 1,000 selected CpG sites based on the set of informative fragments. The analytics system can calculate the anomaly scores in the same manner as the training samples. In some embodiments, the analytics system defines the anomaly score as a binary score based on whether there is a hypermethylated or hypomethylated fragment in the set of informative fragments that encompasses the CpG site.

[0176] The analytics system can then input the test feature vector into the cancer classifier. The function of the cancer classifier can then generate a cancer prediction based on the classification parameters trained in the process 400 and the test feature vector. In the first manner, the cancer prediction can be binary’ and selected from a group consisting of “cancer” or non-cancer;” in the second manner, the cancer prediction is selected from a group of many potential cancer signal origins that could identify affected organ or organ group and at the same time the underlying cancer biology like a histologic fype, the cellular lineage of the cancer cell or origin, or the presence of a cancer driver like infection with an oncovirus and “non-cancer.” In additional embodiments, the cancer prediction has prediction values for each of the many cancer signal origins. Moreover, the analytics system may determine that the test sample's cancer signal has most likely one of the cancer signal origins. Following the example above with the cancer prediction for a test sample as 65% likelihood of originating from the organ breast, 25% likelihood of originating from the organ lung, and 10% likelihood of non-cancer, the analytics system may determine that the test sample is most likely to have a cancer of origin breast. In another example, where the cancer prediction is binary’ as 60% likelihood of non-cancer and 40% likelihood of cancer, the analytics system determines that the test sample is most likely to not have a cancer signal. In additional embodiments, the cancer prediction with the highest likelihood may still be compared against a threshold (e.g., 40%, 50%, 60%, 70%) in order to call the test subject as having a cancer signal or that cancer signal origin present. If the cancer prediction with the highest likelihood does not surpass that threshold, the analytics system may return an inconclusive result, or report the presence of a cancer signal, while not reporting a cancer signal origin

[0177] In additional embodiments, the analytics system chains a cancer classifier trained in step 460 of the process 400 with one or more additional cancer classifiers trained in step 470 of the process 400. The analytics system can input the test feature vector into the cancer classifier trained as a binary classifier in step 460 of the process 400. The analytics system can receive an output of a cancer prediction. The cancer prediction may be binary asAtty. Docket No. 32917-65573 / WO 49 P0252-WOto whether the test subject likely has or likely does not have cancer and if a cancer signal is detected in the sample. In other implementations, the cancer prediction includes prediction values that describe likelihood of cancer and likelihood of non-cancer. For example, the cancer prediction has a cancer prediction value of 85% and the non-cancer prediction value of 15%. The analytics system may determine the test subject to likely have cancer. Once the analytics system determines a test subject is likely to have cancer, the analytics system may input the test feature vector into a multiclass cancer classifier trained to distinguish between different cancer signal origins. The multiclass cancer classifier can receive the test feature vector and returns one or more cancer signal predictions of a cancer signal origin of the plurality of cancer signals in one or multiple categories of cancer signal origins. For example, the multiclass cancer classifier provides a cancer prediction specifying that the test subject is most likely to have a cancer signal originating in a predicted organ or organ group like ovaries or Fallopian tube cancer while at the same time providing a cancer signal origin prediction that this cancer signal has a cell of Mullerian lineage as origin. In another implementation, the multiclass cancer classifier provides a prediction value for each cancer signal origin of the plurality of cancer signal origins and cancer signal origin classifiers. For example, a cancer prediction may include a cancer signal origin for head and neck of 40%, a cancer signal origin for anus of 20%, and cancer signal origin of cervix of 20% for the organ of organ group, and in parallel a prediction of a cancer signal origin being 80% a carcinoma associated with a Human Papilloma Virus (HPV) infection. 10% having origin in a squamous cell carcinoma, and 10% having origin in a cancer of cells of Mullerian lineage.

[0178] According to generalized embodiment of binary cancer classification, the analytics system can determine a cancer score for a test sample based on the test sample's sequencing data (e g., methylation sequencing data, SNP sequencing data, other DNA sequencing data, RNA sequencing data, etc.). The analytics system can compare the cancer score for the test sample against a binary' threshold cutoff for predicting whether the test sample likely has cancer. The binary7threshold cutoff can be tuned using CSO thresholding based on one or more CSO predictions. The analytics system may further generate a feature vector for the test sample for use in the multiclass cancer classifier to determine a cancer signal origin prediction indicating one or more likely organ or organ groups as cancer signal origins, and in parallel indicating one or more histologic types, cellular lineages, or oncogenic drives as cancer signal origin.

[0179] The classifier may be used to determine the disease state of a test subject, e.g., a subject whose disease status is unknown. The method can include obtaining a test genomicAtty. Docket No. 32917-65573 / WO 50 P0252-WOdata construct (e.g., single time point test data), in electronic form, that includes a value for each genomic characteristic in the plurality of genomic characteristics of a corresponding plurality of nucleic acid fragments in a biological sample obtained from a test subject. The method can then include applying the test genomic data construct to the test classifier to thereby determine the state of the disease condition in the test subject. The test subject may not be previously diagnosed with the disease condition.

[0180] In some embodiments, the label for a subject’s disease state can be binary, i.e., a positive label representing that the subject has disease, or a negative label representing the subject does not have the disease. In other embodiments, the label for the disease state can be probabilistic, i.e., a likelihood of having the disease. In other embodiments, the label for the disease state can be multiclass. In one example, the label can represent progression of a disease state, e.g.. Stage I cancer. State II cancer. Stage III cancer. Stage IV cancer, etc. In one example, the label can represent a particular tissue of origin implicated in the disease (e.g., cancer), e.g., lung cancer, breast cancer, head / neck cancer, etc.

[0181] The classifier can be a temporal classifier that uses at least (i) a first test genomic data construct generated from a first biological sample acquired from a test subject at a first point in time, and (ii) a second test genomic data construct generated from a second biological sample acquired from a test subject at a second point in time.

[0182] The trained classifier can be used to determine the disease state of a test subject, e.g., a subject whose disease status is unknown. In this case, the method can include obtaining a test time-series data set, in electronic form, for a test subject, where the test timeseries data set includes, for each respective time point in a plurality of time points, a corresponding test genoty pic data construct including values for the plurality of genotypic characteristics of a corresponding plurality of nucleic acid fragments in a corresponding biological sample obtained from the test subject at the respective time point, and for each respective pair of consecutive time points in the plurality of time points, an indication of the length of time between the respective pair of consecutive time points. The method can then include applying the test genotypic data construct to the test classifier to thereby determine the state of the disease condition in the test subject. The test subject may not be previously diagnosed with the disease condition.IV. AUGMENTED CANCER CLASSIFIER TRAINING WITH WBC DNA

[0183] In one or more embodiments, the analytics system may augment cancer classifier training with WBC DNA samples. The WBC DNA samples may be obtained fromAtty. Docket No. 32917-65573 / WO 51 P0252-WOa buffy coat sample isolated from other components of a blood sample. Augmentation of the cancer classifier with WBC DNA information provides for an improvement to the cancer classifier. The augmentation can yield improvement in classifier sensitivity for solid cancers and / or hematological-based cancers.

[0184] In alternative embodiments, the analytics system may leverage other genomic DNA sample sources, e.g.. other than WBC DNA that is nonetheless matched to the cfDNA samples. As such, the description below can be readily adapted to genomic DNA samples, not specifically WBC DNA samples.IV. A. WBC -DNA FEATURIZATION IN CANCER CLASSIFICATION

[0185] FIG. 6 illustrates a flowchart describing a process of augmenting a cancer classifier with features derived from WBC DNA samples, according to one or more embodiments.

[0186] The analytics system obtains 610 cfDNA samples and WBC DNA samples, with each sample comprising sequence reads and a label of the subject's disease state. The disease state may indicate whether the subject has cancer, a cancer type, a stage of cancer, a prognosis, another characteristic to the subject’s health state, etc. The sequence reads may be methylation sequence reads, e.g., from a targeted methylation sequencing assay.

[0187] The analytics system identifies 620 a first set of cfDNA derived features from the cfDNA samples and a second set of WBC DNA derived features from the WBC DNA samples. To identify each set of features, the analytics system may determine mutual information gain from all features in an initial feature set. The mutual information gain may indicate a discriminatory power of a feature in classifying between two labels. For example, the analytics sy stem may leverage a cohort of cfDNA samples from healthy subjects (or those without a cancer diagnosis) and another cohort of cfDNA samples from subjects with cancer. The analytics system may determine mutual information gain from each feature in totality of the available sequence reads. The analytics system may identify the first set of cfDNA derived features as those features with high mutual information gain. The analytics system may perform likewise to identify’ the second set of WBC DNA derived features, leveraging the WBC DNA samples.

[0188] In one or more embodiments, the analytics system may leverage the second set of WBC DNA derived features for redaction of features from the first set of cfDNA derived features. Such feature redaction may remove confounding features for hematological disorders. In one or more embodiments, the analytics system may perform subject-specific feature redaction, removing confounding features in a subject’s cfDNA sample informed byAtty. Docket No. 32917-65573 / WO 52 P0252-WOthe subject’s WBC DNA sample. Redaction of WBC signal is further described in U.S. Patent Application No. 18 / 594,860 entitled "‘Redacting Cell-Free DNA From Test Samples for Classification by a Mixture Model” filed on March 4, 2024.

[0189] In some embodiments, the analytics system receives informative feature sets from cfDNA samples and WBC DNA samples, ranks features within each matrix by¬ informativeness using supervised and unsupervised criteria such as mutual information, sparsity- -promoting regression weights, and stability across cross-validation folds, and selects an informative set for each modality. The system normalizes features to a common genomic coordinate and ontology- space that encodes locus, strand, context, feature type (e.g., methylation state, sequence variant, fragment end motif), or some combination thereof. The analytics system identifies the overlap between the cfDNA and gDNA informative sets by intersection of coordinates and type-aware matching rules. This overlapping set is used in general cancer classification.

[0190] In some embodiments, the analytics system may identify- hematological features from the informative feature set derived from the WBC DNA samples. In such embodiments, the analytics system may compute information gain for each informative feature in identifying hematological cancer types or other hematological diseases. The analytics system may- curate a grouping of informative features for keying into the hematological cancer types or other hematological diseases. This feature set can be used in a submodel of the cancer classification workflow to predict presence of such cancers. In other example implementations, the cancer classification performs tissue of origin prediction across a plurality- of cancer ty pes, including the hematological cancer t pes, leveraging the informative feature set specific to the hematological cancer types or other hematological diseases.

[0191] The analytics system featurizes 630 the cfDNA samples and / or the WBC DNA samples with the first set of cfDNA derived features and the second set of WBC DNA derived features. Featurization entails generating a feature vector for each sample based on the identified feature sets. As an example with one sample, the analytics system would determine a first feature vector for the first set of cfDNA derived features based on the sample’s sequence reads and a second feature vector for the second set of WBC DNA derived features based on the sample’s sequence reads. The analytics system may concatenate the two feature-set-specific feature vectors to form a singular feature vector for training.

[0192] The analytics system trains 640 the cancer classifier with the concatenated feature vectors for the cfDNA samples and / or the concatenated feature vectors for the WBCAtty. Docket No. 32917-65573 / WO 53 P0252-WODNA samples. Here, the training may leverage principles described in FIGs. 4A & 4B. The analytics system may perform supervised machine-learning training of the cancer classifier as a machine-learning model. The analytics system may leverage solely the cfDNA samples, or may further include the WBC DNA samples when training the cancer classifier. This embodiment is also referred to as an early fusion classifier, fusing the separately -identified feature sets prior to singular input into a cancer classifier.

[0193] FIG. 7 illustrates the augmentation of the cancer classifier with features derived from WBC DNA samples, e.g., as described in FIG. 6. The cfDNA samples 710 and the WBC DNA samples 720 are leveraged to identify cfDNA derived features and WBC DNA derived features, respectively. Upon identifying the informative features, the analytics system determines feature vectors for the cfDNA samples based on the cfDNA derived features, represented by Matrix XI 712. The analytics system also determines features vectors for the cfDNA samples based on the WBC DNA derived features, represented by Matrix X2714. The analytics system does similarly for the WBC DNA samples, yielding Matrix W1 722 for the cfDNA derived features and Matrix W2724 for the WBC DNA derived features. The analytics system concatenates the feature vectors for each sample, yielding Matrix XI, X2 716 for the cfDNA samples 710 and Matrix Wl, W2726 for the WBC DNA samples 720. The concatenated features XI, X2 716 and / or the concatenated features Wl, W2726 are used to train the cancer classifier 730.

[0194] During deployment, for any given type of sample, the analytics system determines the concatenated feature vector inclusive of both the cfDNA derived features and the WBC derived features. The analytics system inputs the concatenated feature vector into the cancer classifier 730 to output the cancer prediction.IV. B. PAIRED CFDNA & WBC DNA CANCER CLASSIFICATION

[0195] FIG. 8 illustrates a flowchart describing a process of architecting a late fusion cancer classifier, according to one or more embodiments.

[0196] The analytics system obtains 810 cfDNA samples and WBC DNA samples, with each sample comprising sequence reads and a label of the subject's disease state. The disease state may indicate whether the subject has cancer, a cancer type, a stage of cancer, a prognosis, another characteristic to the subject’s health state, etc. The sequence reads may be methylation sequence reads, e.g., from a targeted methylation sequencing assay.

[0197] The analytics system identifies 820 a first set of cfDNA derived features from the cfDNA samples and a second set of WBC DNA derived features from the WBC DNA samples. To identify' each set of features, the analytics system may determine mutualAtty. Docket No. 32917-65573 / WO 54 P0252-WOinformation gain from all features in an initial feature set. The mutual information gain may indicate a discriminatory power of a feature in classifying between two labels. For example, the analytics system may leverage a cohort of cfDNA samples from healthy subjects (or those without a cancer diagnosis) and another cohort of cfDNA samples from subjects with cancer. The analytics system may determine mutual information gain from each feature in totality7of the available sequence reads.

[0198] The analytics system may identify the first set of cfDNA derived features as those features with high mutual information gain. The analytics system may perform another information gain analysis to identify the second set of WBC DNA derived features, leveraging the WBC DNA samples.

[0199] In some embodiments, the analytics system may identify an aggregate feature set based on the two sets of informative features derived from the cfDNA samples and the WBC DNA samples. In one or more embodiments, the analytics system may identifies the union of all features from both sets. As such, the aggregate set of features may include one or more features from the first set of cfDNA derived features and one or more features from the second set of cfDNA derived features. In other embodiments, the analytics system may compute an aggregate measure of information gain for a feature across both sets, and select the features for the aggregate set based on the aggregate measure of information gain. In other embodiments, the analytics system identifies the overlap of the two feature sets as the aggregate feature set. The aggregate set of features may exclude one or more features from either the first set of cfDNA derived features or the second set of WBC DNA derived features.

[0200] The analytics system featurizes 830 the cfDNA samples and / or the WBC DNA samples with the first set of cfDNA derived features. Featurization entails generating a feature vector for each sample based on the specific feature set. As an example with one sample, the analytics system would determine a feature vector for the first set of cfDNA derived features based on the sample’s sequence reads.

[0201] The analytics system trains 840 a cfDNA feature classifier with the first feature vectors for the cfDNA samples and / or the first features vectors for the WBC DNA samples. The cfDNA feature classifier is trained to output a cfDNA feature-specific prediction on disease state for the subject. In one or more embodiments, the WBC DNA samples may be paired to at least a subset of the cfDNA samples. In such embodiments, the feature vectors from the paired samples may be concatenated together in training the cfDNA feature classifier.Atty. Docket No. 32917-65573 / WO 55 P0252-WO

[0202] The analytics system featurizes 850 the cfDNA samples and / or the WBC DNA samples with the second set of WBC DNA derived features. Featurization entails generating a feature vector for each sample based on the specific feature set. As an example with one sample, the analytics system would determine a second feature vector for the second set of WBC DNA derived features based on the sample’s sequence reads.

[0203] The analytics system trains 860 a WBC DNA feature classifier with the second feature vectors for the cfDNA samples and / or the second features vectors for the WBC DNA samples. The WBC DNA feature classifier is trained to output a WBC DNA feature-specific prediction on disease state for the subject. In one or more embodiments, the WBC DNA samples may be paired to at least a subset of the cfDNA samples. In such embodiments, the feature vectors from the paired samples may be concatenated together in training the WBC DNA feature classifier.

[0204] The analytics system then trains 870 a logistic layer fusing outputs from the cfDNA feature classifier and the WBC DNA feature classifier into an aggregate prediction on disease state for the subject. This embodiment is also referred to as a late fusion classifier, fusing the separately -identified predictions from separately-trained feature-specific classifiers prior to singular input into a cancer classifier.

[0205] FIG. 9 illustrates the late fusion cancer classifier w ith separately -trained feature-specific classifiers, e.g., as described in FIG. 8. The cfDNA samples 910 and the WBC DNA samples 920 are leveraged to identify cfDNA derived features and WBC DNA derived features, respectively. Upon identifying the informative features, the analytics system determines feature vectors for the cfDNA samples based on the cfDNA derived features, represented by Matrix XI 912. The analytics system also determines features vectors for the cfDNA samples based on the WBC DNA derived features, represented by Matrix X2914. The analytics system does similarly for the WBC DNA samples, yielding Matrix W1 922 for the cfDNA derived features and Matrix W2924 for the WBC DNA derived features. The analytics system trains two separate feature-specific classifiers. The analytics system trains the cfDNA feature classifier 930 with the feature vectors for the cfDNA derived features, e.g., Matrix XI 912 and / or Matrix W1 922. The analytics system trains the WBC DNA feature classifier 940 with the feature vectors for the WBC DNA derived features, e.g., Matrix X2914 and / or Matrix W2924. The analytics system trains the logistic layer 950 which takes the outputs of the two feature-specific classifiers, to output a singular prediction on disease state.Atty. Docket No. 32917-65573 / WO 56 P0252-WO

[0206] During deployment, for any given type of sample, the analytics system determines two feature vectors for the two feature-specific classifiers, i.e., one feature vector for the cfDNA derived features and another feature vector for the WBC derived features. The analytics system inputs each feature vector into the respective feature-specific classifier, yielding two separate predictions. The two separate predictions are input into the logistic layer 950 to yield the aggregate prediction on disease state.

[0207] In one or more embodiments, the architecture of the two feature-specific classifiers and the logistic layer is considered one cancer classifier model. The featurespecific classifiers are parallel in the model architecture. The logistic layer is downstream of the two feature-specific classifiers, i.e., the feature-specific classifiers amount to a first tier in the cancer classifier model, with the logistic layer as a second tier in the cancer classifier model.IV. C. REFLEXASSAY

[0208] In one or more embodiments, the analytics system may leverage a tiered workflow for cancer classification with a blood sample. A blood draw from an individual yields a blood sample (e.g., up to 5 rnL). The blood sample may be separated into plasma and buffy coat fractions. An aliquot of plasma is processed to extract circulating cell-free DNA (cfDNA). The cfDNA is sequenced yielding sequencing data. Remaining plasma and buffy coat fractions are stored at collection to permit deferred reflex testing, e.g., without an additional draw.

[0209] At a first tier of the cancer classification, the analytics system analyzes the sequencing data from the cfDNA sample to detect presence of cancer signal. The analytics system may apply a cfDNA cancer classifier to output a cancer prediction based on the cfDNA-derived sequencing data. The cfDNA cancer classifier may leverage both cfDNA-dervied features and WBC-DNA-derived features. In one or more embodiments, the cfDNA cancer classifier is configured for a high sensitivity, low specificity operating point. The first-tier engine produces a binary or probabilistic prediction. For example, if the result falls below a predefined threshold, the sample is reported as negative. In response to the negative result, the analytics system may report out the negative result, e.g., to a healthcare provider, a clinician, or the individual.

[0210] If the first-tier prediction exceeds the threshold, the analytics system initiates a second tier of classification on the matched buffy coat material. Genomic DNA (gDNA) is extracted from leukocytes and sequenced to yield gDNA sequencing data. The analytics system featurizes the cfDNA sequencing data and the WBC DNA sequencing data. TheAtty. Docket No. 32917-65573 / WO 57 P0252-WOanalytics system applies a paired classifier to the sample’s features to output a cancer prediction. The paired classifier can be tuned to a target specificity, target sensitivity¬ operating point. The analytics system may provide the output of the second tier of classification, alone or in combination with the results of the first tier of classification.

[0211] The operating characteristics of the two tiers may be selected to minimize total sequencing burden while maximizing clinical sensitivity-. In one embodiment, the first tier is set such that approximately 95% of tested samples are resolved as negative without reflex, with the remaining ~5% routed to the second tier. Other thresholds may be used to accommodate different prevalence or utility- models. The storage of both plasma and buffy coat fractions at the outset permits asynchronous reflex without recontacting the patient and ensures that the paired genomic context is temporally matched to the plasma draw.This two-step architecture yields improved overall sensitivity relative to a single-stage test while controlling the number of samples that require intensive analysis. By resolving most negatives after the first tier and using the matched leukocyte gDNA to confirm or refute first-tier positives, the workflow reduces false-positive calls and associated follow-up. The approach is compatible with a range of classification models and feature sets and can incorporate updates to the cfDNA feature-based cancer classifier or the paired classifier without changes to specimen handling. The process is thus well suited for population-scale screening, reflex testing in clinical laboratories, or other implementations where resource efficiency and high specificity must be balanced against detection sensitivity.IV. D. HEMATOLOGICAL CANCER TYPE SORTING

[0212] In one or more embodiments, the analytics system may implement a cancer classification schema for classifying between hematological cancer ty pes, solid tumor cancer types, and non-cancer type. The analytics system may identify WBC noise by evaluating paired features derived from plasma cfDNA and matched WBC DNA for training subject. The analytics system computes a similarity metric between the cfDNA and WBC DNA features. Training samples derived from subject with solid tumors exhibit lowcfDNA-to-WBC similarity, as the cancer signal originates primarily from solid tissue tumor cells. Training samples with hematologic malignancies exhibit high similarity because the malignant clone resides in WBC. Training samples with non-cancer conditions often show mid-to-high similarity due to clonal hematopoiesis and other immune or precursor processes that manifest in both cfDNA and WBC DNA profiles. The analytics sy stem may leverage this pattern to separate solid cancers, heme cancers, and non-cancers. and to estimate the relative contribution of WBC-derived signal within each sample.Atty. Docket No. 32917-65573 / WO 58 P0252-WO

[0213] In one embodiment, the analytics system configures the classifier by presorting the different partitions when training. The training pipeline may entail feature selection with a cross-validated scheme that maximizes the correlation differential between the partitions. The resulting similarity score and feature subset are incorporated as covariates or constraints in the cancer classifier that outputs class probabilities and a WBC-signal burden estimate. This approach improves discrimination of solid cancers, hematologic cancers or diseases, and non-cancers.IV. E. EXAMPLE RESULTS OF AUGMENTED CANCER CLASSIFICATION

[0214] FIGs. 10A-10C illustrate receiver operating characteristic (ROC) curves for augmentation of an early fusion classifier. Each figure plots one curve for a cancer classifier trained without WBC DNA derived features (titled “cfDNA only chiron mf ’) and another curve for a cancer classifier trained with WBC DNA derived features (titled“early _fusion_mr_64_no_wbc”). FIG. 10A illustrates the ROC curves for a validation set of samples with solid cancers and samples with hematological-based cancers. There was a 4% sensitivity improvement. FIG. 10B illustrates the ROC curves for only the samples with hematological-based cancers. There was a 6% sensitivity improvement. FIG. 10C illustrates the ROC curves for only the samples with solid cancers. There was a 4% sensitivity improvement.

[0215] FIGs. 11 A-l 1C illustrate receiver operating characteristic (ROC) curves for augmentation of a late fusion classifier. Each figure plots one curve for a cancer classifier trained without WBC DNA derived features (titled “cfDNA_only_chiron_mr”) and another curve for a cancer classifier trained with WBC DNA derived features (titled “late_fusion_mr_with_wbc’'). FIG. 11 A illustrates the ROC curves for a validation set of samples with solid cancers and samples with hematological-based cancers. There was a 2% sensitivity improvement. FIG. 1 IB illustrates the ROC curves for only the samples with hematological-based cancers. There was a 4% sensitivity improvement. FIG. 11C illustrates the ROC curves for only the samples with solid cancers. There was a 1% sensitivity improvement.

[0216] FIGs. 12A-12C illustrate receiver operating characteristic (ROC) curves for augmentation of a late fusion classifier trained with paired cfDNA and WBC DNA samples. Each figure plots one curve for a cancer classifier trained without WBC DNA derived features (titled “cfDNA_only_chiron_mr’‘) and another curve for a cancer classifier trained with WBC DNA derived features (titled "wbc paired earl v no heme cal ib"). FIG. 12A illustrates the ROC curves for a validation set of samples with solid cancers and samples withAtty. Docket No. 32917-65573 / WO 59 P0252-WOhematological-based cancers. There was a 10% sensitivity setback. FIG. 12B illustrates the ROC curves for only the samples with hematological-based cancers. There was a 14% sensitivity improvement. FIG. 12C illustrates the ROC curves for only the samples with solid cancers. There was a 16% sensitivity setback. This tended to show a tradeoff in improved hematological-based cancer sensitivity at the cost of solid tumor type cancer sensitivity'.

[0217] FIG. 13 illustrates a plot revealing the correlation between cfDNA derived features and WBC DNA derived features. As noted in FIG. 10 A, there is an overall 4% improvement in sensitivity from leveraging augmentation with WBC DNA derived features. The system can detect -50% of undetected solid tumor cancers with the non-augmented cancer classifier, with the potential boost in sensitivity with the augmented cancer classifier. There was potential for -12% boost in sensitivity for hematological-based cancers.

[0218] FIG. 14 illustrates the spread of cancer predictions for early stage cancer samples. The right line around 0.7 is the current non-augmented classifier threshold for positive calling. Roughly 10% of the samples in the validation set had a predicted cancer likelihood between 0.4 and 0.7, from the non-augmented classifier. This 10% stands to be a potential improvement via augmentation with WBC DNA derived features.V. APPLICATIONS

[0219] In some embodiments, the methods, analytics systems and / or classifier of the present invention can be used to detect the presence of cancer, monitor cancer progression or recurrence, monitor therapeutic response or effectiveness, determine a presence or monitor minimum residual disease (MRD), or any combination thereof. For example, as described herein, a classifier can be used to generate a probability score (e.g., from 0 to 1) or scaled score or percent (e.g., from 0 to 100) describing a likelihood that a test feature vector is from a subject with cancer. In some embodiments, the probability score is compared to a threshold probability to determine whether or not the subject has a detectable cancer signal.

[0220] In one or more embodiments, the likelihood or probability score can be assessed at multiple different time points (e.g., before or after treatment) to monitor disease progression or to monitor treatment effectiveness (e.g., therapeutic efficacy). For example, the individual being monitored may be subject to recurrent sample draws. The biological sample draw n from the individual may be sequenced, e.g., by one or more sequencing devices, to measure sequencing data from the sample. The analytics system may perform various analyses, including any of the cancer classification analyses described elsewhere. In the context of monitoring progression, recurrent sample collection, sequencing, and analysisAtty. Docket No. 32917-65573 / WO 60 P0252-WOcan track change in cancer signal in the individual over time. If cancer signal is increasing over time, the analytics system may infer the tumor is malignant and progressing. If the cancer signal plateaus, the analytics system may infer the tumor is benign or stagnant. In the context of treatment assessment, the analytics system may collect samples, sequence the samples, and analyze the samples to predict cancer signal before versus after treatment. If the cancer signal decreases following treatment administration, then the analytics system may infer the treatment is successful. The analytics system may further compare the change in cancer signal across a population to assess efficacy individual-to-individual. In one or more embodiments, upon detecting a significant cancer signal, the analytics system may transmit a notification to the individual (e.g., for viewing on their personal computing device) to visit a healthcare provider for diagnostic workup.

[0221] In still other embodiments, the likelihood or probability score can be used to make or influence a clinical decision (e.g., diagnostic workup of a cancer signal detection to diagnose a cancer, treatment selection, assessment of treatment effectiveness, etc.). For example, in one embodiment, if the probability score exceeds a threshold, a physician can prescribe an appropriate treatment. In other embodiments, the prediction results may include a CSO prediction comprising one or more predicted CSO classes. For example, the CSO prediction may include an organ or organ group, a tumor biology7, a cell of origin, or some combination thereof. Such results (or combination of results) may inform diagnostic workup options, treatment options and follow-up strategies between the healthcare provider and the subject.V . A. DETEC TION OF CANCER

[0222] In some embodiments, the methods and / or classifier of the present invention are used to detect the presence or absence of cancer in a subject unsuspected of having cancer. For example, a classifier (e.g., as described above in Section III) can be used to determine a cancer prediction describing a likelihood that a test feature vector is from a subject that has cancer.

[0223] In one embodiment, a cancer prediction is a likelihood (e.g.. scored between 0 and 100) for whether the test sample has cancer (i.e., binary classification). Thus, the analytics system may determine a threshold for determining whether a test subject has cancer. For example, a cancer prediction of greater than or equal to 60 can indicate that the subject has cancer. In still 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 indicates thatAtty. Docket No. 32917-65573 / WO 61 P0252-WOthe subject has cancer. In other embodiments, the cancer prediction can indicate the severity of disease. For example, a cancer prediction of 80 may indicate a more severe form, or later stage, of cancer compared to a cancer prediction below 80 (e.g., a probability score of 70). Similarly, an increase in the cancer prediction over time (e.g., determined by classifying test feature vectors from multiple samples from the same subject taken at two or more time points) can indicate disease progression or a decrease in the cancer prediction over time can indicate successful treatment.

[0224] In another embodiment, a cancer prediction comprises many prediction values, wherein each of a plurality' of different cancer signal origins being classified (i.e., multiclass classification) for has a prediction value (e.g., scored between 0 and 100). The prediction values may correspond to a likelihood that a given training sample (and during inference, training sample) has each of the cancer signal origins. The analytics system may identify the cancer signal origin that has the highest prediction value and indicate that the test subject likely has a cancer from this signal origin. In other embodiments, the analytics system further compares the highest prediction value to a threshold value (e.g.. 50. 55, 60, 65, 70, 75, 80. 85. etc.) to determine that the test subject likely has a cancer of that signal origin. In other embodiments, a prediction value can also indicate the severity of disease. For example, a prediction value greater than 80 may indicate a more severe form, or later stage, of cancer compared to a prediction value of 60. Similarly, an increase in the prediction value over time (e.g.. determined by classifying test feature vectors from multiple samples from the same subject taken at two or more time points) can indicate disease progression or a decrease in the prediction value over time can indicate successful treatment.

[0225] In some embodiments, the methods and / or classifier of the present invention are used to determine a cancer source of origin prediction in a subject suspected of having cancer. For example, one or more CSO classifiers (e.g., as described above in Section IV) can be used to determine a cancer source of origin prediction to aid in diagnostic workup options.

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

[0227] Examples of cancers that can be detected using the methods, systems and classifiers of the present invention include carcinoma, lymphoma, blastoma, sarcoma, andAtty. Docket No. 32917-65573 / WO 62 P0252-WOleukemia or lymphoid malignancies. More particular examples of such cancers include, but are not limited to, squamous cell cancer (e.g., epithelial squamous cell cancer), skin carcinoma, melanoma, lung cancer, including small-cell lung cancer, non-small cell lung cancer (“NSCLC”), adenocarcinoma of the lung and squamous carcinoma of the lung, cancer of the peritoneum, gastric or stomach cancer including gastrointestinal cancer, pancreatic cancer (e g., pancreatic ductal adenocarcinoma), cervical cancer, ovarian cancer (e.g., high grade serous ovarian carcinoma), liver cancer (e.g., hepatocellular carcinoma (HCC)), hepatoma, hepatic carcinoma, bladder cancer (e g., urothelial bladder cancer), testicular (germ cell tumor) cancer, 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 or uterine carcinoma, salivary gland carcinoma, kidney or renal cancer (e.g., renal cell carcinoma, nephroblastoma or Wilms’ tumor), prostate cancer, vulval cancer, thyroid cancer, anal carcinoma, penile carcinoma, head and neck cancer, esophageal carcinoma, and nasophary ngeal carcinoma (NPC).Additional examples of cancers include, without limitation, retinoblastoma, thecoma, arrhenoblastoma, hematological malignancies, including but not limited to non-Hodgkin's lymphoma (NEIL), multiple myeloma and acute hematological malignancies, endometriosis, fibrosarcoma, choriocarcinoma, lary ngeal carcinomas, Kaposi's sarcoma, Schwannoma, oligodendroglioma, neuroblastomas, rhabdomyosarcoma, osteogenic sarcoma, leiomyosarcoma, and urinary tract carcinomas.

[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 & neck cancer, hepatobiliary cancer, leukemia, lung cancer, lymphoma, melanoma, multiple myeloma, ovarian cancer, pancreatic cancer, prostate cancer, renal cancer, thyroid cancer, uterine cancer, or any combination thereof.

[0229] In some embodiments, the one or more cancer can be a “high-signal” cancer (defined as cancers with greater than 50% 5-year cancer-specific mortality ), such as anorectal, colorectal, esophageal, head & neck, hepatobiliary, lung, ovarian, and pancreatic cancers, as well as lymphoma and multiple myeloma. High-signal cancers tend to be more aggressive and typically have an above-average cell-free nucleic acid concentration in test samples obtained from a patient.V.B. CANCER AND TREATMENT MONITORING

[0230] In some embodiments, the cancer prediction can be assessed at multiple different time points (e.g., or before or after treatment) to make a patient prognosis, predictAtty. Docket No. 32917-65573 / WO 63 P0252-WOresponse to a candidate treatment, to monitor disease progression or to monitor treatment effectiveness (e.g.. therapeutic efficacy). For example, the present invention includes methods that involve 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 therefrom (as described herein), obtaining a second test sample (e.g., a second plasma cfDNA sample) from the cancer patient at a second time point, and determining a second cancer prediction therefrom (as described herein).

[0231] In certain embodiments, the first time point is before a cancer treatment (e.g., before a resection surgery or a therapeutic intervention), and the second time point is after a cancer treatment (e.g., after a resection surgery' or therapeutic inters ention), and the classifier is utilized to monitor the effectiveness of the treatment. For example, if the second cancer prediction decreases compared to the first cancer prediction, then the treatment is considered to have been successful. However, if the second cancer prediction increases compared to the first cancer prediction, then the treatment is considered to have not been successful. In other embodiments, both the first and second time points are before a cancer treatment (e.g., before a resection surgery or a therapeutic intervention). In still other embodiments, both the first and the second time points are after a cancer treatment (e.g., after a resection surgery or a therapeutic intervention). In still other embodiments, cfDNA samples may be obtained from a cancer patient at a first and second time point and analyzed, e.g., to monitor cancer progression, to determine if a cancer is in remission (e.g.. after treatment), to monitor or detect residual disease or recurrence of disease, or to monitor treatment (e.g., therapeutic) efficacy.

[0232] Those of skill in the art will readily appreciate that test samples can be obtained from a cancer patient over any desired set of time points and analyzed in accordance with the methods of the invention to monitor a cancer state in the patient. In some embodiments, the first and second time points are separated by an amount of time that ranges from about 15 minutes up 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, 1.5, 2, 2.5, 3, 3.5, 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, test samples can be obtained from the patient at least once every 5 months, at least once every' 6 months, at least once a year, at least onceAtty. Docket No. 32917-65573 / WO 64 P0252-WOevery 2 years, at least once every' 3 years, at least once every' 4 years, or at least once every 5 years.V.C. TREATMENT

[0233] In still another embodiment, the cancer prediction can be used to make or influence a clinical decision (e.g., diagnosis of cancer, treatment selection, assessment of treatment effectiveness, etc.). For example, in one embodiment, if the cancer prediction (e.g., for cancer or for a particular cancer signal origin) exceeds a threshold, a physician can prescribe an appropriate treatment (e.g., a resection surgery, radiation therapy, chemotherapy, and / or immunotherapy).

[0234] A classifier (as described herein) can be used to determine a cancer prediction that a sample feature vector is from a subject that has cancer. In one embodiment, an appropriate treatment (e.g., resection surgery or therapeutic) is prescribed when the cancer prediction exceeds a threshold. For example, in one embodiment, if the cancer prediction is greater than or equal to 60 one or more appropriate 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 appropriate treatments are prescribed. In other embodiments, the cancer prediction can indicate the severity of disease. An appropriate treatment matching the severity of the disease may then be prescribed.

[0235] In some embodiments, the treatment is one or more cancer therapeutic agents selected from the group consisting of a chemotherapy agent, a targeted cancer therapy agent, a differentiating therapy agent, a hormone therapy agent, and an immunotherapy agent. For example, the treatment can be one or more chemotherapy agents selected from the group consisting of alkylating agents, antimetabolites, anthracy clines, anti-tumor antibiotics, cytoskeletal disruptors (taxans), topoisomerase inhibitors, mitotic inhibitors, corticosteroids, kinase inhibitors, nucleotide analogs, platinum-based agents and any combination thereof. In some embodiments, the treatment is one or more targeted cancer therapy agents selected from the group consisting of signal transduction inhibitors (e.g., tyrosine kinase and growth factor receptor inhibitors), histone deacetylase (HD AC) inhibitors, retinoic receptor agonists, proteosome inhibitors, angiogenesis inhibitors, and monoclonal antibody conjugates. In some embodiments, the treatment is one or more differentiating therapy agents including retinoids, such as tretinoin, alitretinoin and bexarotene. In some embodiments, the treatment is one or more hormone therapy agents selected from the group consisting of anti-estrogens, aromatase inhibitors, progestins, estrogens, anti-androgens, and GnRH agonists or analogs.Atty. Docket No. 32917-65573 / WO 65 P0252-WOIn one embodiment, the treatment is one or more immunotherapy agents selected from the group comprising monoclonal antibody therapies such as rituximab (RITUXAN) and alemtuzumab (CAMPATH), non-specific immunotherapies and adjuvants, such as BCG, interleukin-2 (IL-2), and interferon-alfa, immunomodulating drugs, for instance, thalidomide and lenalidomide (REVLIMID). It is within the capabilities of a skilled physician or oncologist to select an appropriate cancer therapeutic agent based on characteristics such as the type of tumor, cancer stage, previous exposure to cancer treatment or therapeutic agent, and other characteristics of the cancer.V.D. KIT IMPLEMENTATION

[0236] Also disclosed herein are kits for performing the methods described above including the methods relating to the cancer classifier. The kits may include one or more collection vessels for collecting a sample from the subject comprising genetic material. The sample can include blood, plasma, serum, urine, fecal, saliva, other ty pes of bodily fluids, or any combination thereof. Such kits can include reagents for isolating nucleic acids from the sample. The reagents can further include reagents for sequencing the nucleic acids including buffers and detection agents. In one or more embodiments, the kits may include one or more sequencing panels comprising probes for targeting particular genomic regions, particular mutations, particular genetic variants, or some combination thereof. In other embodiments, samples collected via the kit are provided to a sequencing laboratory that may use the sequencing panels to sequence the nucleic acids in the sample. The WBC contamination detection may be applied to various configurations of the kit, to minimize WBC contamination potentially originating from components of the kit. For example, experiments may be run comparing types of collection vessels. WBC contamination can be assessed and compared between the types of collection vessels to identify an optimal type that minimizes WBC contamination.

[0237] A kit can further include instructions for use of the reagents included in the kit. For example, a kit can include instructions for collecting the sample, extracting the nucleic acid from the test sample. Example instructions can be the order in which reagents are to be added, centrifugal speeds to be used to isolate nucleic acids from the test sample, how to amplify' nucleic acids, how to sequence nucleic acids, or any combination thereof. The instructions may further illuminate how to operate a computing device as the analytics system 200, for the purposes of performing the steps of any of the methods described.

[0238] In addition to the above components, the kit may’ include computer-readable storage media storing computer software for performing the various methods describedAtty. Docket No. 32917-65573 / WO 66 P0252-WOthroughout the disclosure. One form in which these instructions can be present is as printed information on a suitable medium or substrate, e.g., a piece or pieces of paper on which the information is printed, in the packaging of the kit, in a package insert. Yet another means would be a computer readable medium, e.g., diskette, CD, hard-drive, network data storage, on which the instructions have been stored in the form of computer code. Y et another means that can be present is a website address or QR code which can be used via the internet to access the information at a removed site.VI. ADDITIONAL CONSIDERATIONS

[0239] The foregoing detailed description of embodiments refers to the accompanying drawings, which illustrate specific embodiments of the present disclosure. Other embodiments having different structures and operations do not depart from the scope of the present disclosure. The term “the invention” or the like is used with reference to certain specific examples of the many alternative aspects or embodiments of the applicants’ invention set forth in this specification, and neither its use nor its absence is intended to limit the scope of the applicants’ invention or the scope of the claims.

[0240] Embodiments of the invention may also relate to an apparatus for performing the operations herein. This apparatus may be specially constructed for the required purposes, and / or it may comprise a general-purpose computing device selectively activated or reconfigured by a computer program stored in the computer. Such a computer program may be stored in a non-transitory, tangible computer readable storage medium, or any type of media suitable for storing electronic instructions, which may be coupled to a computer system bus. Furthermore, any computing systems referred to in the specification may include a single processor or may be architectures employing multiple processor designs for increased computing capability.

[0241] Any of the steps, operations, or processes described herein as being performed by the analytics system may be performed or implemented with one or more hardware or software modules of the apparatus, alone or in combination with other computing devices. In one embodiment, a software module is implemented with a computer program product comprising a computer-readable medium containing computer program code, which can be executed by a computer processor for performing any or all of the steps, operations, or processes described.Atty. Docket No. 32917-65573 / WO 67 P0252-WO

Claims

CLAIMSWlIAT IS CLAIMED IS:

1. A method for augmentation of a cell-free (cf) DNA cancer classifier with white blood cell (WBC) DNA. the method comprising:accessing cfDNA samples and WBC DNA samples, with each sample comprising sequence reads and a label of an originating subject’s disease state; identifying a first set of cfDNA derived features from the sequence reads of the cfDNA samples and a second set of WBC DNA derived features from the sequence reads of the WBC DNA samples;featurizing each cfDNA sample based on the first set of cfDNA derived features and the second set of WBC DNA derived features yielding a feature vector;training a cancer classifier with the feature vectors for the cfDNA samples, the cancer classifier trained to output a prediction on disease state for a test cfDNA sample.

2. The method of claim 1, wherein accessing the cfDNA samples and the WBC DNA samples comprises:accessing a blood sample drawn from a subject;isolating the cfDNA sample and the WBC DNA sample from the blood sample:andperforming sequencing of nucleic acid fragments in the cfDNA sample to yield the sequence reads for the cfDNA sample and sequencing of nucleic acid fragments in the WBC DNA sample to yield the sequence reads for the WBC DNA sample.

3. The method of claim 1, wherein the sequence reads for each sample comprise methylation sequence reads obtained from targeted methylation sequencing of a plurality of CpG sites.

4. The method of claim 1 , wherein identifying the first set of cfDNA derived features comprises:Atty. Docket No. 32917-65573 / WO 68 P0252-WOcomputing information gain for each feature of an initial set of features in discrimination of cfDNA samples having a positive label for the disease state and cfDNA samples having a negative label for the disease state; and selecting the cfDNA derived features based on the information gain computed with the cfDNA samples.

5. The method of claim 1. wherein identifying the second set of WBC DNA derived features comprises:computing information gain for each feature of the initial set of features in discrimination of WBC DNA samples having the positive label for the disease state and WBC DNA samples having the negative label for the disease state; andselecting the WBC DNA derived features based on the information gain computed with the WBC DNA samples.

6. The method of claim 1 , further comprising:featurizing each WBC DNA sample with the first set of cfDNA derived features and the second set of WBC DNA derived features yielding a feature vector;wherein training the cancer classifier comprises training with the feature vectors for the WBC DNA samples.

7. The method of claim 1. further comprising:identifying an aggregate set of features including one or more features from the first set of cfDNA derived features and one or more features from the second set of cfDNA derived features, wherein the aggregate set of features excludes one or more features from either the first set of cfDNA derived features or the second set of WBC DNA derived features, wherein featurizing each cfDNA sample yielding the feature vector comprises featurizing each cfDNA sample based on the aggregate set of features.

8. The method of claim 1, further comprising:identifying an aggregate set of features from the first set of cfDNA derived features and the second set of WBC DNA derived features by:Atty. Docket No. 32917-65573 / WO 69 P0252-WOidentifying one or more features overlapping the first set and the second set, andidentifying the aggregate set of features by removing the overlapping features from the first set of cfDNA derived features, wherein featurizing each cfDNA sample yielding the feature vector comprises featurizing each cfDNA sample based on the aggregate set of features.

9. The method of claim 1. wherein training the cancer classifier comprises performing supervised machine-learning training to predict the label of the disease state for each cfDNA sample based on the feature vector of the cfDNA sample.

10. The method of claim 1, wherein the samples comprise a label of one or more cancer types from a plurality of cancer types, wherein training the cancer classifier comprises training the cancer classifier to output a multiclass cancer prediction.

11. A method for augmentation of a cell-free (cf) DNA cancer classifier with white blood cell (WBC) DNA, the method comprising:accessing cell-free (cf) DNA samples and white blood cell (WBC) DNA samples, with each sample comprising sequence reads and a label of an originating subject’s disease state;identifying a first set of cfDNA derived features from the sequence reads of the cfDNA samples and a second set of WBC DNA derived features from the sequence reads of the WBC DNA samples;featurizing each cfDNA sample with the first set of cfDNA derived features yielding a cfDNA feature-specific feature vector;training a cfDNA feature classifier with the cfDNA feature-specific feature vectors for the cfDNA samples;featurizing each cfDNA sample with the second set of WBC DNA derived features yielding a WBC DNA feature-specific feature vector; training a WBC DNA feature classifier with the WBC DNA feature-specific feature vectors for the cfDNA samples; andtraining a logistic layer configured to input predictions by the cfDNA feature classifier and the WBC DNA feature classifier to yield an aggregate prediction on disease state for a test cfDNA sample.Atty. Docket No. 32917-65573 / WO 70 P0252-WO12. The method of claim 11, wherein accessing the cfDNA samples and the WBC DNA samples comprises:accessing a blood sample drawn from a subject;isolating the cfDNA sample and the WBC DNA sample from the blood sample;andperforming sequencing of nucleic acid fragments in the cfDNA sample to yield the sequence reads for the cfDNA sample and sequencing of nucleic acid fragments in the WBC DNA sample to yield the sequence reads for the WBC DNA sample.

13. The method of claim 11, wherein the sequence reads for each sample comprise methylation sequence reads obtained from targeted methylation sequencing of a plurality of CpG sites.

14. The method of claim 11, wherein identifying the first set of cfDNA derived features comprises:computing information gain for each feature of an initial set of features in discrimination of cfDNA samples having a positive label for the disease state and cfDNA samples having a negative label for the disease state; and selecting the cfDNA derived features based on the information gain computed with the cfDNA samples.

15. The method of claim 11, wherein identifying the second set of WBC DNA derived features comprises:computing information gain for each feature of the initial set of features in discrimination of WBC DNA samples having the positive label for the disease state and WBC DNA samples having the negative label for the disease state; andselecting the WBC DNA derived features based on the information gain computed with the WBC DNA samples.

16. The method of claim 11, further comprising:featurizing each WBC DNA sample with the first set of cfDNA derived features yielding a cfDNA feature-specific feature vector,Atty. Docket No. 32917-65573 / WO 71 P0252-WOwherein training the cfDNA feature classifier comprises training with the cfDNA feature-specific feature vectors for the WBC DNA samples.

17. The method of claim 11, further comprising:featurizing each WBC DNA sample with the second set of WBC DNA derived features yielding a WBC DNA feature-specific feature vector; wherein training the WBC DNA feature classifier comprises training with the WBC DNA feature-specific feature vectors for the WBC DNA samples.

18. The method of claim 11, wherein training the logistic layer comprises performing supervised machine-learning training to predict the label of the disease state for each sample based on the predictions by the cfDNA feature classifier and the WBC DNA feature classifier for the cfDNA sample.

19. The method of claim 11, wherein training the cfDNA feature classifier, the WBC DNA feature classifier, or the logistic layer comprises training on concatenated feature vectors from each paired cfDNA sample and WBC DNA sample.

20. The method of claim 11, wherein the samples each comprise a label of one or more cancer types from a plurality of cancer types, wherein training the logistic layer comprises training the cancer classifier to output a multiclass cancer prediction.

21. A non-transitory computer-readable storage medium storing instructions for execution of a cancer classifier trained by the method of any one of claims 1-20.

22. A method for prediction of a disease state for a subject, the method comprising:accessing a test sample comprising sequence reads;featurizing the test sample based on the first set of cfDNA derived features and the second set of WBC DNA derived features yielding a feature vector for the test sample;applying the cancer classifier to the feature vector for the test sample to output a prediction on disease state for the test sample, wherein the cancer classifier is trained by the method of any one of claims 1-20.Atty. Docket No. 32917-65573 / WO 72 P0252-WO23. The method of claim 22, wherein the test sample comprises sequence reads derived from sequencing cell-free DNA (cfDNA) fragments and sequence reads derived from sequencing white blood cell (WBC) DNA fragments.

24. A method for prediction of a disease state for a subject, the method comprising:accessing a cfDNA sample comprising sequence reads derived from sequencing cell-free DNA (cfDNA) fragments;featurizing the cfDNA sample based on a first set of cfDNA derived features yielding a first feature vector;applying a first cancer classifier to the first feature vector for the test sample to output a prediction on disease state for the test sample;responsive to detection of positive signal in the disease state, accessing a white blood cell (WBC) DNA sample from a reflex assay, the WBC DNA sample comprising sequence reads derived from sequencing WBC DNA fragments;featurizing the WBC DNA sample based on a second set of WBC DNA derived features yielding a second feature vector; andapplying a second cancer classifier to the second feature vector, and optionally the first feature vector, to output a second prediction on disease state for the test sample, wherein the second cancer classifier is trained by the method of any one of claims 1-20.Atty. Docket No. 32917-65573 / WO 73 P0252-WO

Images