Methods and compositions for analysis of cell-free biomarkers

Abstract

In various aspects, the present disclosure provides methods for detection of various cancer types, comprising measuring the level of target molecules in a sample. In some embodiments, the one or more target molecules include a cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer, and a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types. Methods for training a classifier for detecting target molecules from a cancer are also provided.

Description

METHODS AND COMPOSITIONS FOR ANALYSIS OF CELL-FREE BIOMARKERSCROSS-REFERENCE

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 578,347, filed August 23, 2023, and U.S. Provisional Application No. 63 / 549,406, filed February 2, 2024, which applications are incorporated herein by reference in their entireties for all purposes.BACKGROUND

[0002] Cancer represents a prominent worldwide public health problem. Screening programs and early diagnosis have an important impact in improving disease-free survival and reducing mortality in cancer patients. As noninvasive approaches for early diagnosis foster patient compliance, they can be included in screening programs.

[0003] DNA methylation plays an important role in regulating gene expression. Aberrant DNA methylation has been implicated in many disease processes, including cancer. DNA methylation profiling using methylation sequencing (e.g., whole genome bisulfite sequencing (WGBS)) is increasingly recognized as a valuable diagnostic tool for detection, diagnosis, and / or monitoring of cancer. For example, specific patterns of differentially methylated regions may be useful as molecular markers for various diseases. However, WGBS is not ideally suitable for a product assay, as only a few percent of the genome is likely to be useful in classification because the majority of the genome either is not differentially methylated in cancer, or has a local CpG density that is too low to provide a robust signal.

[0004] Cancer remains a frequent cause of death worldwide. Over the last several decades, treatment options have improved, yet survival rates remain low. The success of treatment by surgical resection and drug-based approaches is strongly dependent on identification of early- stage tumors. However, many current detection methodologies frequently cannot identify tumors until the more advanced stages of the disease.SUMMARY

[0005] In view of the foregoing, there remains a need for non-invasive detection modalities that can identify disease at the earliest stages, when therapeutic interventions have a greater chance of success. Aspects of the present disclosure address this need, and provide other advantages as well. For example, some aspects provided herein involve methods combining the detection of polypeptides in combination with assessment of methylation patterns in cfDNA in a non-invasive, cost-efficient manner for detecting cancer-associated biomarkers in a sample from a subject. Byanalyzing both data relating to methylation patterns in cfDNA and polypeptide levels, higher sensitivity for detecting cancer biomarkers can be achieved at specificity equal to or greater than using either alone. Such improvement in detection allows for identifying true positive cancer samples that may been missed by analysis of either analyte alone.

[0006] In one aspect, the present disclosure provides a method of detecting cancer in a subject. In some embodiments, the method of detecting cancer in a subject includes: (a) measuring levels of first target molecules from a first sample of the subject; (b) measuring levels of second target molecules from a second sample of the subject; (c) applying a trained classifier to the measured levels of the first and second target molecules to assign an aggregate probability score for the cancer; and (d) detecting the cancer by identifying that the aggregate probability score is above a threshold for presence of the cancer. In some embodiments, the first target molecules include cell- free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer types. In some embodiments, the second target molecules include a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types. In some embodiments, applying the trained classifier includes: (i) applying a first trained model to the measured levels of the first target molecules to assign a first probability score for the cancer; (ii) applying a second trained model to the measured levels of the second target molecules to assign a second probability score for the cancer; and (iii) aggregating the first probability score and the second probability score. In some embodiments, the first sample, and the second sample are the same.

[0007] In some embodiments, the trained classifier was trained using reference first probability scores from the first trained model, reference second probability scores from the second trained model, and reference aggregated probability scores aggregating the reference first probability scores and reference second probability scores, for reference samples from (1) reference subjects having known cancers, and (2) reference subjects without cancer. In some embodiments, the trained classifier assigns an aggregate probability score for each of a plurality of different cancer types, and detecting the cancer includes identifying the cancer type as the cancer type with the highest aggregate probability score. In some embodiments, aggregating the first probability score and the second probability score includes calculating a product of the first and second probability scores for the cancer. In some embodiments, aggregating the first probability score and the second probability score includes combining the first and second probability scores for the cancer in a linear model.

[0008] In some embodiments, the plurality of different target genomic regions includes at least 1000, 5000, 10000, 20000, or 30000 target genomic regions. In some embodiments, the plurality of target genomic regions includes a total collective length of at least 50 kb, 100 kb, 500 kb, or1000 kb. In some embodiments, each of the plurality of different target genomic regions includes at least five methylation sites. In some embodiments, measuring the first target molecules includes sequencing converted cfDNA from the plurality of different target genomic regions, or amplification products thereof, wherein the converted cfDNA includes cfDNA treated with a deaminating agent. In some embodiments, the method further including treating the cfDNA with the deaminating agent, optionally wherein the deaminating agent is a cytosine deaminase or bisulfite. In some embodiments, the sequencing produces at least 100,000 sequencing reads.

[0009] In some embodiments, measuring the first target molecules includes enriching for the converted cfDNA or amplification products thereof, to produce an enriched sample of polynucleotides. In some embodiments, the enriching includes capturing the converted cfDNA or amplification products thereof with a plurality of corresponding bait oligonucleotides. In some embodiments, the plurality of different target genomic regions for enrichment by the bait oligonucleotides are genomic regions identified by the first trained model as differentially methylated in the at least one of a plurality of cancer types relative to non-cancer tissue or relative to cancer of a different type.

[0010] In some embodiments, the plurality of different polypeptides include at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides. In some embodiments, the plurality of different polypeptides comprise: (a) polypeptides identifying proteins selected from List 1; (b) polypeptides identifying proteins selected from any one of Lists 2-19; (c) polypeptides identifying proteins selected from List 20; or (d) polypeptides identifying one or more of CHAD, KRT19, MMP12, PTN, SERPINA3, and SPP1.

[0011] In some embodiments, the trained classifier discriminates a subject with cancer from a subject without cancer with a defined specificity for each of the plurality of cancer types. In some embodiments, the trained classifier has a higher sensitivity for cancer detection than each of the first trained model and the second trained model; optionally wherein the trained classifier has a specificity for cancer detection that is equal to or greater than each of the first trained model and the second trained model. In some embodiments, the trained classifier is a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier.

[0012] In some embodiments, the first trained model and / or the second trained model is a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier. In some embodiments, the first trained model binarizes measured levels of the first target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected. In some embodiments, the second trained model log-transforms measured levels of the second target molecules normalized against controlprotein present in a known amount. In some embodiments, (a) the first trained model is trained using measured levels of the first target molecules for first reference samples, (b) the second trained model is trained using measured levels of the second target molecules for second reference samples, and (c) the first and second reference samples comprise samples from reference subjects having known cancers, and reference subjects without cancer.

[0013] In some embodiments, the first sample and / or second sample includes a biological fluid. In some embodiments, the biological fluid includes blood, plasma, serum, urine, saliva, pleural fluid, pericardial fluid, cerebrospinal fluid (CSF), peritoneal fluid, or any combination thereof. In some embodiments, the biological fluid includes blood, a blood fraction, plasma, or serum. In some embodiments, the first sample and / or second sample is a plasma sample.

[0014] In some embodiments, the plurality of cancer types include at least 10 cancer types. In some embodiments, the plurality of cancer types includes one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0015] In some embodiments of any of the methods provided herein, the method further includes treating the subject for the cancer type. In some embodiments, the treating includes surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

[0016] In one aspect, also provided herein is a method of training a classifier for detecting target molecules from a cancer. In some embodiments, the method of training a classifier for detecting target molecules from a cancer includes (a) receiving first measured levels of first target molecules for first samples of reference subjects; (b) training a first model to generate a first probability score for the presence of cancer in a subject by applying a first machine learning algorithm to the first measured levels; (c) receiving second measured levels of second target molecules for second samples of the reference subjects; (d) training a second model to generate a second probability score for the presence of cancer in a subject by applying a second machine learning algorithm to the second measured levels; (e) generating reference first cancer probability scores for the first samples using the trained first model; (f) generating reference second cancer probability scores for the second samples using the trained second model; (g) generating reference aggregated cancer probability scores for a plurality of the reference subjects by aggregating the reference first cancer probability score and reference second cancer probability score for each respective reference subject; and (h) training a classifier to generate an aggregate cancer probability score for a subject by applying a third machine learning algorithm to the reference first cancer probability scores, reference second cancer probability scores, and reference aggregated cancer probability scores. In some embodiments, the first target molecules include cell-free DNA (cfDNA) from a plurality ofdifferent target genomic regions that are differentially methylated in at least one of a plurality of cancer types. In some embodiments, the reference subjects include first subjects having known cancer types, and second subjects without cancer. In some embodiments, the second target molecules include a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types.

[0017] In some embodiments, aggregating the first cancer probability score and the second cancer probability score includes calculating a product of the first and second probability scores for the cancer. In some embodiments, aggregating the first probability score and the second probability score includes combining the first and second probability scores for the cancer in a linear model. In some embodiments, the first machine learning algorithm, second machine learning algorithm, and / or third machine learning algorithm is an LI -regularized logistic regression, an L2-regularized logistic regression, a generalized linear model (GLM), a random forest, a multinomial logistic regression, a multilayer perceptron, a support vector machine, or a neural network. In some embodiments, the first trained model binarizes measured levels of the first target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected. In some embodiments, the second trained model log-transforms measured levels of the second target molecules normalized against control protein present in a known amount. In some embodiments, the third machine learning algorithm is a logistic regression. In some embodiments, (a) the first trained model is trained using measured levels of the first target molecules for first reference samples, (b) the second trained model is trained using measured levels of the second target molecules for second reference samples, and (c) the first and second reference samples comprise samples from reference subjects having known cancers, and reference subjects without cancer. In some embodiments, at least some of the first reference samples and second reference samples are from the same reference subjects.

[0018] In some embodiments, the plurality of different target genomic regions includes at least 1000, 5000, 10000, 20000, or 30000 target genomic regions. In some embodiments, the plurality of target genomic regions includes a total collective length of at least 50 kb, 100 kb, 500 kb, or 1000 kb. In some embodiments, each of the plurality of different target genomic regions includes at least five methylation sites. In some embodiments, the first measured levels include sequencing results for the cfDNA or amplicons thereof. In some embodiments, the sequencing results include at least 100,000 reads for each of the first samples.

[0019] In some embodiments, the plurality of different polypeptides include at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides. In some embodiments, the plurality of different polypeptides comprise: (a) polypeptides identifying proteins selected from List 1; (b) polypeptides identifying proteins selected from any one of Lists 2-19; (c)polypeptides identifying proteins selected from List 20; or (d) polypeptides identifying one or more of CHAD, KRT19, MMP12, PTN, SERPINA3, and SPP1.

[0020] In some embodiments, the plurality of cancer types include at least 10 cancer types. In some embodiments, the plurality of cancer types includes one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0021] In one aspect, provided herein is a method of screening for cancer in a subject comprising (a) measuring levels of first target molecules from a first sample of the subject, wherein the first target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of a plurality of cancer types; (b) applying a first trained model to the measured levels of the first target molecules to assign a first probability score for each of the plurality of cancer types; wherein (i) the first trained model has a first specificity for cancer detection, and (ii) the first probability score for at least one of the cancer types is above a first threshold for the presence of cancer; (c) measuring levels of second target molecules from a second sample of the subject, wherein the second target molecules comprise cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of the plurality of cancer types; (d) applying a second trained model to the measured levels of the second target molecules to assign a second probability score for the cancer; wherein (ii) the second trained model has a second specificity for cancer detection, and (ii) the second specificity is higher than the first specificity; and (e) detecting the cancer by identifying that the second probability score is above a threshold for the presence of cancer. In some embodiments, (a) the first trained model is trained using measured levels of the first target molecules for first reference samples, (b) the second trained model is trained using measured levels of the second target molecules for second reference samples, and (c) the first and second reference samples comprise samples from reference subjects having known cancers, and reference subjects without cancer. In some embodiments, the first sample and the second sample are the same. In some embodiments, the method further comprises treating the subject for the cancer type (e.g., by surgical resection, radiation therapy, chemotherapy, and / or immunotherapy).

[0022] In some embodiments, the plurality of different target genomic regions comprises at least 1000, 5000, 10000, 20000, or 30000 target genomic regions. In some embodiments, the plurality of target genomic regions comprises a total collective length of at least 50 kb, 100 kb, 500 kb, or 1000 kb. In some embodiments, each of the plurality of different target genomic regions comprises at least five methylation sites. In some embodiments, measuring the second target molecules comprises sequencing converted cfDNA from the plurality of different target genomic regions, oramplification products thereof, wherein the converted cfDNA comprises cfDNA treated with a deaminating agent. In some embodiments, the sequencing produces at least 100,000 sequencing reads. In some embodiments, measuring the first target molecules comprises enriching for the converted cfDNA or amplification products thereof, to produce an enriched sample of polynucleotides. In some embodiments, the enriching comprises capturing the converted cfDNA or amplification products thereof with a plurality of corresponding bait oligonucleotides. In some embodiments, the plurality of different target genomic regions for enrichment by the bait oligonucleotides are genomic regions identified by the second trained model as differentially methylated in the at least one of a plurality of cancer types relative to non-cancer tissue or relative to cancer of a different type. In some embodiments, the plurality of different polypeptides comprise at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides. In some embodiments, the plurality of different polypeptides comprise: (a) polypeptides identifying proteins selected from List 1; (b) polypeptides identifying proteins selected from any one of Lists 2-19; (c) polypeptides identifying proteins selected from List 20; or (d) polypeptides identifying one or more of CHAD, KRT19, MMP12, PTN, SERPINA3, and SPP1. In some embodiments, the first trained model and / or the second trained model is a neural network classifier, a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier. In some embodiments, the second trained model binarizes measured levels of the second target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected. In some embodiments, the first trained model log-transforms measured levels of the first target molecules normalized against a control polypeptide present in a known amount. In some embodiments, (a) the first sample and / or second sample comprises a biological fluid; optionally where the biological fluid comprises blood, plasma, serum, urine, saliva, pleural fluid, pericardial fluid, cerebrospinal fluid (CSF), peritoneal fluid, or any combination thereof; (b) the first sample and / or second sample is a plasma sample. In some embodiments, (a) the plurality of cancer types comprise at least 10 cancer types; and / or (b) the plurality of cancer types comprises one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0023] In one aspect, also provided herein is a method of treating cancer in a subject. In some embodiments, the method for treating a cancer in a subject includes selecting a subject based on the results of a detection or screening assay, and treating the subject for the cancer, wherein: (a) the detection or screening assay includes a method of detecting or screening for cancer in a subjectaccording to any of the various aspects or embodiments described herein; and (b) the treating includes surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

[0024] In one aspect, provided herein is a non-transitory computer-readable medium with instructions stored thereon, that when executed by one or more processors, perform one or more steps in a method of detecting cancer in a subject according to any of the various aspects or embodiments described herein.

[0025] In one aspect, provided herein is a non-transitory computer-readable medium with instructions stored thereon, that when executed by one or more processors, perform a method of training a classifier for detecting target molecules from a cancer according to any of the various aspects or embodiments provided herein.INCORPORATION BY REFERENCE

[0026] All publications, patents, and patent applications mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference.BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The novel features of the disclosure are set forth with particularity in the appended claims. A better understanding of the features and advantages of the present disclosure will be obtained by reference to the following detailed description that sets forth illustrative embodiments, in which the principles of the disclosure are utilized, and the accompanying drawings of which:

[0028] FIG. 1A is a flowchart describing a process of sequencing a fragment of cell-free DNA (cfDNA), according to an embodiment.

[0029] FIG. IB is an illustration of a process of FIG. 1A of sequencing a fragment of cfDNA to obtain a methylation state vector, according to an embodiment.

[0030] FIG. 2 illustrates a process for developing single-analyte models for cancer detection using cfDNA methylation or protein level in a sample.

[0031] FIG. 3 provide illustrative plots of cancer probability scores based on 1 -dimensional analysis of cfDNA methylation (left plot), 1 -dimensional analysis protein levels (bottom plot), and 2-dimensional analysis with cancer probability based on protein as the x-axis and cancer probability based on cfDNA methylation as the y-axis (middle plot). The shaded areas designate samples below a selected threshold that are designated as non-cancer by the respective models. Dots corresponding to samples in the white area between the two dashed lines in the middle plot represent additional cancer calls gained by analyzing both cfDNA and protein, as compared to cfDNA alone at the same specificity. All of these gained calls correspond to samples from cancersubjects. In the left plot, samples from subjects without cancer (non-cancer) and samples from subjects with cancer are plotted on the left and right, respectively. In the bottom plot, samples from subjects without cancer (non-cancer) and samples from subjects with cancer are plotted on the bottom and top, respectively.

[0032] FIG. 4 illustrates a process for developing a multi-omics classifier for cancer detection by combined assessment of cfDNA methylation and protein level in a sample.

[0033] FIG. 5A is an illustrative Receiver Operator Characteristic (ROC) plot comparing cancer detection by analysis of cfDNA methylation alone or in combination with analysis of protein levels in a sample. The bottom line in each pair of lines corresponds to analysis of cfDNA methylation alone, while the top line corresponds to analysis of cfDNA methylation in combination with analysis of protein markers.

[0034] FIG. 5B is a graph comparing the sensitivity at 99.4% specificity for cancer detection by analysis of cfDNA methylation alone or in combination with analysis of protein levels in a sample. As indicated in the graph, mean sensitivity at 99.4% specificity for analysis of cfDNA methylation in combination with protein markers was 0.534, while mean sensitivity at 99.4% specificity for analysis of cfDNA methylation alone was 0.478.

[0035] FIG. 6 is a graph comparing the sensitivity at 99.4% specificity for cancer detection by analysis of cfDNA methylation in combination with analysis of the indicated number of protein markers.DETAILED DESCRIPTION

[0036] Before the present invention is described in greater detail, it is to be understood that this invention is not limited to particular embodiments described, as such may, of course, vary. It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only, and is not intended to be limiting, since the scope of the present invention will be limited only by the appended claims.

[0037] Where a range of values is provided, it is understood that each intervening value, to the tenth of the unit of the lower limit, unless the context clearly dictates otherwise, between the upper and lower limit of that range and any other stated or intervening value in that stated range, as well as each of the provided endpoints of the range, is encompassed within the invention. The upper and lower limits of these smaller ranges may independently be included in the smaller ranges encompassed within the invention, subject to any specifically excluded limit in the stated range.

[0038] Unless defined otherwise, technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0039] As used herein, the term “about” means a range of values including the specified value, which a person of ordinary skill in the art would consider reasonably similar to the specified value. In embodiments, about means within a standard deviation using measurements generally acceptable in the art. In embodiments, about means a range extending to + / - 10% of the specified value. In embodiments, about includes the specified value.

[0040] The term “methylation” as used herein refers to a process by which a methyl group is added to a DNA molecule. For example, a hydrogen atom on the pyrimidine ring of a cytosine base can be converted to a methyl group, forming 5-methylcytosine. The term also refers to a process by which a hydroxymethyl group is added to a DNA molecule, for example by oxidation of a methyl group on the pyrimidine ring of a cytosine base. Methylation and hydroxymethylation tend to occur at dinucleotides of cytosine and guanine referred to herein as “CpG sites.” The principles described herein are also applicable for the detection of methylation in a non-CpG context, including non-cytosine methylation. In such embodiments, a wet laboratory assay used to detect methylation may vary from any described herein. 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).

[0041] The term “methylation” can also refer to the methylation status of a CpG site. A CpG site with a 5-methylcytosine moiety is methylated. A CpG site with a hydrogen atom on the pyrimidine ring of the cytosine base is unmethylated.

[0042] The term “methylation site” as used herein refers to a region of a DNA molecule where a methyl group can be added. “CpG” sites are the most common methylation site, but methylation sites are not limited to CpG sites. For example, DNA methylation may occur in cytosines in CHG and CHH, where H is adenine, cytosine or thymine. Cytosine methylation in the form of 5- hydroxymethylcytosine may also assessed (see, e.g., US20110236894A1 and US20110301045A1, which are incorporated herein by reference), and features thereof, using the methods and procedures disclosed herein.

[0043] The term “CpG site” as used herein refers to a region of a DNA molecule where a cytosine nucleotide is followed by a guanine nucleotide in the linear sequence of bases along its 5' to 3' direction. “CpG” is a shorthand for 5'-C-phosphate-G-3' that is cytosine and guanine separated by only one phosphate group. Cytosines in CpG dinucleotides can be methylated to form 5-methylcytosine.

[0044] In some embodiments, oligonucleotide probes described herein comprise one or more CpG detection sites. The term “CpG detection site” as used herein refers to a region in a probe that is configured to hybridize to a CpG site of a target DNA molecule. The CpG site on the target DNA molecule can comprise cytosine and guanine separated by one phosphate group, wherecytosine is methylated or unmethylated. The CpG site on the target DNA molecule can comprise uracil and guanine separated by one phosphate group, where the uracil is generated by the conversion of unmethylated cytosine.

[0045] The term “UpG” is a shorthand for 5'-U-phosphate-G-3' that is uracil and guanine separated by only one phosphate group. UpG can be generated by a bisulfite treatment of a DNA that converts unmethylated cytosines to uracils. Cytosines can be converted to uracils by other methods such as chemical modification, synthesis, or enzymatic conversion.

[0046] The terms “hypomethylated” or “hypermethylated” as used herein refer to a methylation status of a DNA molecule containing multiple CpG sites (e.g., more than 2, 3, 4, 5, 6, 7, 8, 9, 10, etc.) where a high percentage of the CpG sites (e.g., more than 80%, 85%, 90%, or 95%, or any other percentage within the range of 50%-100%, 70% or more, 75% or more, 80% or more, 85% or more, 90% or more, 95% or more, 97.5% or more, 98% or more, 99% or more, 99.9% or more, or any other numerical percentage within the range of 50%-100% or more, wherein the provided range includes the range limits of 50% and 100%) are unmethylated or methylated, respectively. For example, “hypomethylated” nucleic acid, e.g., cfDNA, fragments can be fragments having a number, e.g., 3 or more, 4 or more, 5 or more, 6 or more, 7 or more, 9 or more, 10 or more, of CpG sites with a percentage, e.g., 70% or more, 75% or more, 80% or more, 85% or more, 90% or more, or 95% or more, or 97.5% or more, 98% or more, 99% or more, 99.9% or more, of the CpG sites being unmethylated. Likewise, “hypermethylated” nucleic acid, e.g., cfDNA, fragments can be fragments having a number, e.g., 3 or more, 4 or more, 5 or more, 6 or more, 7 or more, 9 or more, 10 or more, of CpG sites with a percentage, e.g., 70% or more, 75% or more, 80% or more, 85% or more, 90% or more, or 95% or more, or 97.5% or more, 98% or more, 99% or more, 99.9% or more of the CpG sites being methylated. In some embodiments, a hypomethylated DNA molecule comprises a plurality of CpG sites, at least 80% of which are unmethylated. In some embodiments, a hypermethylated DNA molecule comprises a plurality of CpG sites, at least 80% of which are methylated. In some embodiments, the level of a first target molecule is the presence, number, or proportion of cfDNA molecules from a target genomic region having a particular level of methylation, or the presence, number, or proportion of sequencing reads thereof.

[0047] The terms “methylation state vector” or “methylation status vector” as used herein refer to a vector comprising multiple elements, where each element indicates the methylation status of a methylation site in a DNA molecule comprising multiple methylation sites, in the order they appear from 5' to 3' in the DNA molecule. For example, < Mx, Mx+i, Mx+2 >, < Mx, Mx+i, Ux+2 >, . . ., < Ux, Ux+i, Ux+2 > can be methylation vectors for DNA molecules comprising three methylation sites, where M represents a methylated methylation site and U represents an unmethylated methylation site. In some embodiments, the level of a first target molecule is thepresence, number, or proportion of cfDNA molecules from a target genomic region having a particular methylation state vector, or the presence, number, or proportion of sequencing reads thereof.

[0048] The terms “abnormal methylation pattern” and “anomalous methylation pattern” as used herein refer to a methylation pattern of a nucleic acid molecule (e.g., a cfDNA molecule) or a methylation state vector thereof that is found in a sample at a higher frequency than expected for a healthy, e.g., non-cancer, sample. In various embodiments, the probability of finding such a methylation pattern in a healthy (e.g., non-cancer) sample is lower than a threshold value. As such, for example, the terms “abnormally methylated” and “anomalously methylated” as used herein describe a nucleic acid, e.g., DNA such as a cfDNA, molecule or a methylation state vector exhibiting an abnormal methylation pattern. Where an abnormal methylation pattern differentiates cancer from non-cancer, and / or one cancer type from another cancer type, the respective target genomic region may be referred to as “differentially methylated.” Whether a target genomic region is differentially methylated can be used as an indicator for a determination of healthy, e.g., non-cancer, as opposed to diseased, e.g., cancer, in referring to the health of a subject from which a subject sample was originated. In some embodiments provided herein, the finding and / or expectedness of finding a specific methylation state vector in a healthy control group including healthy individuals is represented by a p-value. In some embodiments, a low p-value score corresponds to a methylation state vector which is relatively unexpected in comparison to other methylation state vectors within samples from healthy individuals, such as individuals in a healthy control group. In some versions, a high p-value score corresponds to a methylation state vector which is relatively more expected in comparison to other methylation state vectors found in samples from healthy individuals such as those in the healthy control group. In various embodiments, a methylation state vector having an abnormal / anomalous methylation pattern is a methylation state vector having a p-value at and / or lower than a threshold value (e.g., 0.1, 0.01, 0.001, 0.0001, etc.), such as a threshold value corresponding with a healthy, e.g., non-cancer, sample. In various embodiments, the methods include associating a methylation state vector from a sample and having a p-value at and / or lower than a threshold value (e.g., 0.1 or smaller, 0.01 or smaller, 0.001 or smaller, 0.0001 or smaller, etc.) with a determination that the sample is not a healthy sample, e.g., is a sample from a subject having cancer. In various embodiments, the threshold value is applied as a filter in that application of a smaller threshold value (e.g., 0.001, 0.0001, etc.) is associated with a higher expectation that a methylation state vector is from a sample that is not healthy, e.g., is a sample from an individual with cancer. Various methods can be used to calculate a p-value or expectedness of a methylation pattern or a methylation state vector. Exemplary methods provided herein involve use of a Markov chain probability that assumesmethylation statuses of CpG sites to be dependent on methylation statuses of neighboring CpG sites. Alternate methods provided herein calculate the expectedness of observing a specific methylation state vector in healthy individuals by utilizing a mixture model including multiple mixture components, each being an independent-sites model where methylation at each CpG site is assumed to be independent of methylation statuses at other CpG sites. In some versions, the subject methods include determining whether a nucleic acid, e.g., DNA, molecule or a methylation state vector is abnormally methylated. In various embodiments of the methods, a generated p- value is compared, such as by an analytics system, against a threshold to identify vectors, e.g., nucleic acids such as cfDNA fragments, that are abnormally methylated relative to a control group, such as a group associated with one or more healthy, e.g., non-cancer, samples. In addition, abnormal methylation, e.g., cfDNA methylation, can be hypermethylation and / or hypomethylation, both of which can be indicative of a non-healthy, e.g., cancer, status. Accordingly, the methods include determining healthy or diseased, e.g., non-cancer or cancer, status based at least in part on a p-value, such as a relatively low p-value, such as a p-value below a threshold, wherein the p-value can in various aspects be indicative of abnormal methylation, e.g., hypermethylation and / or hypomethylation. A low p-value, e.g., a p-value at or below a threshold value (e.g., 0.1, 0.01, 0.001, 0.0001, etc.), can be indicative of abnormal methylation, e.g., hypermethylation and / or hypomethylation in a sample. In various embodiments, the methods include determining healthy or diseased, e.g., non-cancer or cancer, status of a sample based on a nucleic acid, e.g., nucleic acid fragment, or methylation vector from the sample having a low p- value, e.g., equal to or less than 0.1, 0.01 or 0.001, and being both hypermethylated and hypomethylated or hypermethylated or hypomethylated. In various aspects, the methods include determining healthy or diseased, e.g., non-cancer or cancer, status of a sample based at least in part on whether a nucleic acid, e.g., nucleic acid fragment, or methylation vector from the sample, is both hypermethylated and hypomethylated. In some variations, determining a vector, e.g., sample fragment, to be anomalously methylated based on a generated p-value score includes determining whether the generated score for the vector is below a threshold score, wherein the threshold score is a degree of confidence that the vector is anomalously methylated.

[0049] The term “amplicon” as used herein means the product of a polynucleotide amplification reaction; that is, a clonal population of polynucleotides, which may be single stranded or double stranded, which are replicated from one or more starting sequences. The one or more starting sequences may be one or more copies of the same sequence, or they may be a mixture of different sequences. Preferably, amplicons are formed by the amplification of a single starting sequence. Amplicons may be produced by a variety of amplification reactions whose products comprise replicates of the one or more starting, or target, nucleic acids. In one aspect, amplification reactionsproducing amplicons are “template-driven” in that base pairing of reactants, either nucleotides or oligonucleotides, have complements in a template polynucleotide that are required for the creation of reaction products. In one aspect, template-driven reactions are primer extensions with a nucleic acid polymerase, or oligonucleotide ligations with a nucleic acid ligase. Such reactions include, but are not limited to, polymerase chain reactions (PCRs), linear polymerase reactions, nucleic acid sequence-based amplification (NASBAs), rolling circle amplifications, and the like, disclosed in the following references, each of which are incorporated herein by reference herein in their entirety: Mullis et al, U.S. Pat. Nos. 4,683,195; 4,965,188; 4,683,202; 4,800,159 (PCR); Gelfand et al, U.S. Pat. No. 5,210,015 (real-time PCR with “taqman” probes); Wittwer et al, U.S. Pat. No. 6,174,670; Kacian et al, U.S. Pat. No. 5,399,491 (“NASBA”); Lizardi, U.S. Pat. No. 5,854,033; Aono et al, Japanese patent publ. JP 4-262799 (rolling circle amplification); and the like. In one aspect, amplicons of the invention are produced by PCRs. An amplification reaction may be a “real-time” amplification if a detection chemistry is available that permits a reaction product to be measured as the amplification reaction progresses, e.g., “real-time PCR”, or “real-time NASBA” as described in Leone et al, Nucleic Acids Research, 26: 2150-2155 (1998), and like references.

[0050] The term “enrich” as used herein means to increase a proportion of one or more target molecule (e.g., a target nucleic acid or a target polypeptide) in a sample. For instance, an “enriched” sample or sequencing library is therefore a sample or sequencing library in which a proportion of one of more target nucleic acids has been increased with respect to non-target nucleic acids in the sample. In some embodiments, enrichment comprises physical separation of target molecules from non-target molecules.

[0051] The term “cancerous sample” as used herein refers to a sample comprising genomic DNAs and / or polypeptides from an individual with cancer. The genomic DNAs can be, but are not limited to, cfDNA fragments or chromosomal DNAs from a subject with cancer. The genomic DNAs can be sequenced and their methylation status can be assessed by various methods, for example, bisulfite sequencing. When genomic sequences are obtained from public database (e.g., The Cancer Genome Atlas (TCGA)) or experimentally obtained by sequencing a genome of an individual diagnosed with cancer, cancerous sample can refer to genomic DNAs or cfDNA fragments having the genomic sequences. The polypeptides can be detected by any suitable method known in the art. The term “cancerous samples” as a plural refers to samples comprising genomic DNAs and / or polypeptides from multiple individuals, each individual being an individual with cancer. In various embodiments, cancerous samples from more than 100, 300, 500, 1,000, 2,000, 5,000, 10,000, 20,000, 40,000, 50,000, or more individuals diagnosed with cancer are used.

[0052] The term “non-cancerous sample” or “healthy sample” as used herein refers to a sample comprising genomic DNAs and / or polypeptides from a healthy individual or an individual notdiagnosed with cancer. The genomic DNAs can be, but are not limited to, cfDNA fragments or chromosomal DNAs from a subject without cancer. The genomic DNAs can be sequenced and their methylation status can be assessed by various methods, for example, bisulfite sequencing. When genomic sequences are obtained from public database (e.g., The Cancer Genome Atlas (TCGA)) or experimentally obtained by sequencing a genome of an individual without cancer, non-cancerous sample can refer to genomic DNAs or cfDNA fragments having the genomic sequences. The polypeptides can be detected by any suitable method known in the art. The term “non-cancerous samples” as a plural refers to samples comprising genomic DNAs and / or polypeptides from multiple individuals, each individual is without cancer. In various embodiments, cancerous samples from more than 100, 300, 500, 1,000, 2,000, 5,000, 10,000, 20,000, 40,000, 50,000, or more individuals without cancer are used. In various embodiments, cancerous samples from 100 or more, 300 or more, 500 or more, 1,000 or more, 2,000 or more, 5,000 or more, 10,000 or more, 20,000 or more, 40,000 or more, or 50,000 or more individuals without cancer are used.

[0053] The term “training sample” as used herein refers to a sample used to train a model or classifier described herein, select one or more target genomic regions for capture / analysis, and / or to determine a probability score for cancer detection. The training samples can comprise genomic DNAs or a modification thereof, and / or polypeptides, from one or more healthy subjects and from one or more subjects having a disease condition (e.g., cancer, a specific type of cancer, a specific stage of cancer, etc.). The genomic DNAs can be, but are not limited to, cfDNA fragments or chromosomal DNAs. The genomic DNAs can be sequenced and their methylation status can be assessed by various methods, for example, bisulfite sequencing. When genomic sequences are obtained from a public database (e.g., The Cancer Genome Atlas (TCGA)) or experimentally obtained by sequencing a genome of an individual, a training sample can refer to genomic DNAs or cfDNA fragments having the genomic sequences. The polypeptides can be detected by any suitable method known in the art.

[0054] The term “test sample” as used herein refers to a sample from a subject, whose health condition was, has been or will be tested using a classifier and / or an assay panel described herein. The test sample can comprise genomic DNAs or a modification thereof, and / or polypeptides. The genomic DNAs can be, but are not limited to, cfDNA fragments or chromosomal DNAs.

[0055] The term “target genomic region” as used herein refers to a region in a genome selected for analysis in test samples. An assay panel is generated with oligonucleotide probes designed to hybridize to (and optionally pull down) nucleic acid fragments derived from the target genomic region or a fragment thereof. Oligonucleotide probes directed to target regions are also referred to herein as “bait oligonucleotides.” A nucleic acid fragment derived from the target genomicregion refers to a nucleic acid fragment generated by degradation, cleavage, bisulfite conversion, or other processing of the DNA from the target genomic region. In some embodiments, a plurality of different bait oligonucleotides are designed to hybridize across a single target genomic region (e.g., overlapping probes tiled across a target genomic region). In general, when referring to a plurality of target genomic regions, no target genomic region of the plurality is wholly contained within another target genomic region. Different target genomic regions is a plurality of target genomic regions may be overlapping, but will at least have different termini. In one embodiment, each target genomic region in a plurality of target genomic regions is separate from and does not overlap with any other target genomic region in the plurality.

[0056] Various target genomic regions are described according to their chromosomal location in the sequence listing filed herewith. Chromosomal DNA is double-stranded, so a target genomic region includes two DNA strands: one with the sequence provided in the listing and a second that is a reverse complement to the sequence in the listing. Probes can be designed to hybridize to one or both sequences. Optionally, probes hybridize to converted sequences resulting from, for example, treatment with sodium bisulfite.

[0057] The term “off-target genomic region” as used herein refers to a region in a genome which has not been selected for analysis in test samples but has sufficient homology to a target genomic region to potentially be bound and pulled down by a probe designed to target the target genomic region. In one embodiment, an off-target genomic region is a genomic region that aligns to a probe along at least 45 bp with at least a 90% match rate.

[0058] The terms “converted DNA molecules,” “converted cfDNA molecules,” and “modified fragment obtained from processing of the cfDNA molecules” refer to DNA molecules obtained by processing DNA or cfDNA molecules in a sample for the purpose of differentiating a methylated nucleotide and an unmethylated nucleotide in the DNA or cfDNA molecules. For example, in some embodiments, the sample can be treated with bisulfite ion (e.g., using sodium bisulfite), to convert unmethylated cytosines (“C”) to uracils (“U”). In another embodiment, the conversion of unmethylated cytosines to uracils is accomplished using an enzymatic conversion reaction, for example, using a cytidine deaminase (such as APOBEC). After treatment, converted DNA molecules or cfDNA molecules include additional uracils which are not present in the original cfDNA sample. Replication by DNA polymerase of a DNA strand comprising a uracil results in addition of an adenine to the nascent complementary strand instead of the guanine normally added as the complement to a cytosine or methylcytosine.

[0059] In general, the terms “cell-free,” “circulating,” and “extracellular” as applied to polynucleotides (e.g. “cell-free DNA” or “cfDNA”) are used interchangeably to refer to polynucleotides present in a sample from a subject or portion thereof that can be isolated orotherwise manipulated without applying a lysis step to the sample as originally collected (e.g., as in lysis for the extraction from cells or viruses). Cell-free polynucleotides are thus unencapsulated or “free” from the cells or viruses from which they originate, even before a sample of the subject is collected. Cell-free polynucleotides may be produced as a byproduct of cell death (e.g. apoptosis or necrosis) or cell shedding, releasing polynucleotides into surrounding body fluids or into circulation. Accordingly, cell-free nucleic acids may be isolated from a non-cellular fraction of blood (e.g. serum or plasma), from other bodily fluids (e.g. urine), or from non-cellular fractions of other types of samples. In some embodiments, cfDNA refers to deoxyribonucleic acid molecules that circulate in a subject’s body (e.g., bloodstream) and may originate from one or more healthy cells and / or from one or more cancer cells.

[0060] The term “circulating tumor DNA” or “ctDNA” refers to nucleic acid fragments that originate from tumor cells, which may be released into an individual’s bloodstream as result of biological processes such as apoptosis or necrosis of dying cells or actively released by viable tumor cells.

[0061] The term “fragment” as used herein can refer to a fragment of a nucleic acid molecule. For example, in one embodiment, a fragment can refer to a cfDNA molecule in a blood or plasma sample, or a cfDNA molecule that has been extracted from a blood or plasma sample. An amplification product of a cfDNA molecule may also be referred to as a “fragment.” In another embodiment, the term “fragment” refers to a sequence read, or set of sequence reads, that have been processed for subsequent analysis (e.g., for machine-learning based classification), as described herein. For example, raw sequence reads can be aligned to a reference genome and matching paired end sequence reads assembled into a longer fragment for subsequent analysis.

[0062] The terms “polypeptide”, “peptide” and “protein” are used interchangeably herein to refer to polymers of amino acids of any length. The terms also encompass an amino acid polymer that has been modified; for example, disulfide bond formation, glycosylation, lipidation, acetylation, phosphorylation, or any other manipulation, such as conjugation with a labeling component. As used herein the term “amino acid” includes natural and / or unnatural or synthetic amino acids, including glycine and both the D or L optical isomers, and amino acid analogs and peptidomimetics. In some embodiments, a polypeptide is encoded by a target polynucleotide, or a portion thereof. In some embodiments, the polypeptide is a polypeptide fragment.

[0063] The terms “individual” and “subject” refer to a human individual. The term “healthy individual” refers to an individual presumed not to have a cancer or disease. In some embodiments, a subject is an individual whose DNA is being analyzed. For example, a subject may be a test subject whose DNA is be evaluated using a targeted panel as described herein to evaluate whether the person has cancer or another disease. In some embodiments, a subject is partof a control group known to have (or not have) cancer or another disease (also referred to as a “reference subject”). Control and cancer / disease groups may be used to assist in designing or validating the targeted panel.

[0064] The term “sequence read” as used herein refers to a string of nucleotides as determined for part of, or all of, a nucleic acid molecule by a nucleic acid sequencing process. A sequence read may be a short string of nucleotides (e.g., 20-150) sequenced from a nucleic acid fragment, a short string of nucleotides at one or both ends of a nucleic acid fragment, or the sequencing of the entire nucleic acid fragment that exists in the biological sample. Sequence reads can be obtained through various methods provided herein or by other methods known in the art.

[0065] The term “sequencing depth” as used herein refers to the count of the number of times a given target nucleic acid within a sample has been sequenced (e.g., the count of sequence reads at a given target region), or is on average actually or expected to be sequenced based on the amount of nucleic acid subjected to sequencing and the total read length generated by a given sequencing process (e.g., an average of the read depth for all sequenced regions from a given sequencing run). Increasing sequencing depth can reduce required amounts of nucleic acids required to assess a disease state (e.g., cancer or cancer tissue of origin).

[0066] “Treating” or “treatment” as used herein includes any approach for obtaining beneficial or desired results in a subject’s condition, including clinical results. Beneficial or desired clinical results can include, but are not limited to, alleviation or amelioration of one or more symptoms or conditions, diminishment of the extent of a disease, stabilizing (z.e., not worsening) the state of disease, prevention of a disease’s transmission or spread, delay or slowing of disease progression, amelioration or palliation of the disease state, diminishment of the reoccurrence of disease, and remission, whether partial or total and whether detectable or undetectable. In other words, “treatment” as used herein includes any cure, amelioration, or prevention of a disease. Treatment may prevent the disease from occurring; inhibit the disease’s spread; relieve the disease’s symptoms, fully or partially remove the disease’s underlying cause, shorten a disease’s duration, or do a combination of these things.

[0067] “Treating” and “treatment” as used herein includes prophylactic treatment. Treatment methods include administering to a subject a therapeutically effective amount of an active agent. The administering step may consist of a single administration or may include a series of administrations. The length of the treatment period depends on a variety of factors, such as the severity of the condition, the age of the patient, the concentration of active agent, the activity of the compositions used in the treatment, or a combination thereof. It will also be appreciated that the effective dosage of an agent used for the treatment or prophylaxis may increase or decrease over the course of a particular treatment or prophylaxis regime. Changes in dosage may result andbecome apparent by standard diagnostic assays known in the art. In some instances, chronic administration may be required. For example, the compositions are administered to the subject in an amount and for a duration sufficient to treat the patient. In embodiments, the treating or treatment is not prophylactic treatment.

[0068] The term “prevent”, as pertains to a disease or condition of a subject, refers to a decrease in the occurrence of one or more corresponding symptoms in the subject. As indicated above, the prevention may be complete (no detectable symptoms) or partial, such that fewer symptoms are observed, and / or with lower incidence, than would likely occur absent treatment.

[0069] “Anti-cancer agent” and “anticancer agent” are used in accordance with their plain ordinary meanings and refer to a composition (e.g. compound, drug, antagonist, inhibitor, modulator) having antineoplastic properties or the ability to inhibit the growth or proliferation of cells. In some embodiments, an anti-cancer agent is a chemotherapeutic. In some embodiments, an anti-cancer agent is an agent identified herein having utility in methods of treating cancer. In some embodiments, an anti-cancer agent is an agent approved by the FDA or similar regulatory agency of a country other than the USA, for treating cancer. Examples of anti-cancer agents include, but are not limited to, MEK inhibitors, alkylating agents, anti-metabolites, plant alkaloids, topoisomerase inhibitors, antitumor antibiotics, platinum-based compounds, inhibitors of mitogen- activated protein kinase signaling, and others known to those skilled in the art.

[0070] In some embodiments, the anti-cancer agent is an epigenetic inhibitor. An “epigenetic inhibitor” as used herein, refers to an inhibitor of an epigenetic process, such as DNA methylation (a DNA methylation Inhibitor) or modification of histones (a Histone Modification Inhibitor). An epigenetic inhibitor may be a histone-deacetylase (HDAC) inhibitor, a DNA methyltransferase (DNMT) inhibitor, a histone methyltransferase (HMT) inhibitor, a histone demethylase (HDM) inhibitor, or a histone acetyltransferase (HAT). Examples of HDAC inhibitors include Vorinostat, romidepsin, CI-994, Belinostat, Panobinostat , Givinostat, Entinostat, Mocetinostat, SRT501, CUDC-101, JNJ-26481585, or PCI24781. Examples of DNMT inhibitors include azacitidine and decitabine. Examples of HMT inhibitors include EPZ-5676. Examples of HDM inhibitors include pargyline and tranylcypromine. Examples of HAT inhibitors include CCT077791 and garcinol.

[0071] In some embodiments, the anti-cancer agent is a multi-kinase inhibitor. A “multi-kinase inhibitor” is a small molecule inhibitor of at least one protein kinase, including tyrosine protein kinases and serine / threonine kinases. A multi-kinase inhibitor may include a single kinase inhibitor. Multi-kinase inhibitors may block phosphorylation. Multi-kinases inhibitors may act as covalent modifiers of protein kinases. Multi-kinase inhibitors may bind to the kinase active site or to a secondary or tertiary site inhibiting protein kinase activity. A multi-kinase inhibitor may be an anti-cancer multi-kinase inhibitor. Exemplary anti-cancer multi-kinase inhibitors includedasatinib, sunitinib, erlotinib, bevacizumab, vatalanib, vemurafenib, vandetanib, cabozantinib, poatinib, axitinib, ruxolitinib, regorafenib, crizotinib, bosutinib, cetuximab, gefitinib, imatinib, lapatinib, lenvatinib, mubritinib, nilotinib, panitumumab, pazopanib, trastuzumab, or sorafenib.Methods of Detecting Cancer

[0072] In one aspect, the present disclosure provides a method of detecting cancer in a subject, the method comprising: (a) measuring levels of first target molecules from a first sample of the subject; (b) measuring levels of second target molecules from a second sample of the subject; (c) applying a trained classifier to the measured levels of the first and second target molecules to assign an aggregate probability score for the cancer; and (d) detecting the cancer by identifying that the aggregate probability score is above a threshold for presence of the cancer. In some embodiments, the first target molecules comprise cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer types. In some embodiments, the second target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types. In some embodiments, applying the trained classifier comprises: (i) applying a first trained model to the measured levels of the first target molecules to assign a first probability score for the cancer; (ii) applying a second trained model to the measured levels of the second target molecules to assign a second probability score for the cancer; and (iii) aggregating the first probability score and the second probability score.Trained Classifiers

[0073] Aspects of the disclosure are directed to trained classifiers. For example, a machine learning or deep learning model (e.g., a trained classifier) can be used to determine a disease state based on the levels of first and second target molecules. In various embodiments, the output of the trained classifier is a probability score of a cancer. For instance, the trained classifier can determine a first probability score based on the levels of the first target molecule, and a second probability score based on the levels of the second target molecule. In some embodiments, the first target molecules are cfDNA molecules from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer types. In some embodiments, the second target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types. In some embodiments, trained classifier aggregates the first and second probability scores to generate an aggregate probability score for the cancer. For instance, in some embodiments, the trained classifier calculates the product of first and second probability scores for the cancer to aggregate the firstand second probability scores, where the first and second probability scores were determined by first and second trained models, respectively. In some embodiments, aggregating the first probability score and the second probability score includes combining the first and second probability scores for the cancer in a linear model. Furthermore, the trained classifier can incorporate or otherwise be used along with thresholding to determine whether a sample is to be called as cancer or non-cancer based on whether the aggregate probability score is above a threshold or not.

[0074] For determining each of the first and second probability scores, the trained classifier can apply trained models to the measured levels of the first target molecule and the second target molecule. For instance, in some embodiments, the trained classifier applies a first trained model to measured levels of the first target molecule to assign a first probably score for the cancer, and applies a second trained model to measured levels of the second target molecule to assign a second probably score for the cancer. In some embodiments, the first trained model binarizes measured levels of the first target molecules by assigning a first value if a target genomic region (e.g., a target genomic region having a particular methylation level, methylation pattern, or methylation state vector) is detected, and a second value if a target genomic region is not detected. In some embodiments, the second trained model log-transforms measured levels of the second target molecules normalized against control protein present in a known amount.

[0075] In some embodiments, the trained classifier assigns an aggregate probability score for each of a plurality of different cancer types. In some embodiments, the cancer type with the highest aggregate probability score is identified as the detected cancer in a sample. In some embodiments, the trained classifier discriminates a subject with cancer from a subject without cancer with a defined specificity for each of the plurality of cancer types. In some embodiments, the plurality of cancer types include at least 10 cancer types. In some embodiments, the plurality of cancer types includes one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0076] The trained classifier and trained models can apply any of a variety of types of modeling. In some embodiments, the trained classifier is a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier. In some embodiments, the first trained model and / or the second trained model is a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier. In some embodiments, the first and the second trained model are the same. In some embodiments, the first and the second trained models are different. In some embodiments, the trained classifier has ahigher sensitivity for cancer detection than each of the first trained model and the second trained model. In some embodiments, the trained classifier has a specificity for cancer detection that is equal to or greater than each of the first trained model and the second trained model. In some embodiments, the first and / or second trained model have a defined specificity for cancer detection that is 0.900 or higher (e.g., at least 0.950, 0.975, 0.980, 0.985, 0.990, 0.995, or higher). In some embodiments, the trained classifier has a defined specificity for cancer detection that is 0.900 or higher (e.g., at least 0.950, 0.975, 0.980, 0.985, 0.990, 0.995, or higher). In some embodiments, the trained classifier has a defined specificity for cancer detection that is at least 0.99 or higher. In some embodiments, the application of the trained classifier comprises a sensitivity of at least 30% (e.g., at least 40%, 50%, 60%, 70%, 80%, or higher).

[0077] To train the cancer type classifier, an analytics system can obtain data for a set of training samples. In some embodiments, the data includes measurements of the level of the first and second target molecules, and a label corresponding to a type of cancer or non-cancer status of the sample. The analytics system can utilize the training set to train the classifier to predict the cancer state of a test sample. For example, a first model may be trained to generate a first probability score for the presence of cancer in a subject by applying a first machine learning algorithm to the first measured levels, a second model may be trained to generate a second probability score for the presence of cancer in a subject by applying a second machine learning algorithm to the second measured levels, and the classifier may then be trained on aggregate cancer probability scores obtained by aggregating scores from the first and second models.

[0078] Furthermore, in some embodiments, the analytics system can split the training set into K subsets or folds to be used in a K-fold cross-validation. In some examples, the folds can be balanced for cancer / non-cancer status, type of cancer, cancer stage, age (e.g., grouped in 10-year buckets), and / or smoking status. In some examples, the training set is split into 5 folds, whereby 5 separate classifiers are trained, in each case training on 4 / 5 of the training samples and using the remaining 1 / 5 for validation.

[0079] During training with the training set, the analytics system can fit a probabilistic model to the level of the first target molecules in a sample, or the level of the second target molecules in a sample. As used herein a “probabilistic model” is any mathematical model capable of assigning a probability to the measured level of a target molecule. During training, the analytics system receives the levels of the first and second target analytes from one or more samples from subjects having a known cancer state and can be used to determine probabilities indicative of a cancer state. The trained classifier can then be trained based on the trained models. For instance, in some embodiments, the trained classifier can be trained using reference first probability scores from the first trained model, reference second probability scores from the second trained model, andreference aggregated probability scores aggregating the reference first probability scores and reference second probability scores, for reference samples. In some embodiments, the reference samples are from reference subjects having known cancers, and from reference subjects without cancer.

[0080] In some examples, the probabilistic model is a “mixture model” fitted using a mixture of components from underlying models. In some examples, the analytics system performs fits separately for each cancer type in a plurality of cancer types. According to various aspects of the subject disclosure, other means can be used to fit the probabilistic models or to identify parameters that maximize the log-likelihood of a given level (e.g., sequence reads or protein level) derived from the reference samples. For example, in some examples, Bayesian fitting (using e.g., Markov chain Monte Carlo), in which each parameter is not assigned a single value but instead is associated to a distribution, is used. In some examples, gradient-based optimization, in which the gradient of the likelihood (or log-likelihood) with respect to the parameter values is used to step through parameter space towards an optimum, is used. In some embodiments, expectation-maximization, in which a set of latent parameters (such as identities of the mixture component from which each of a plurality of cfDNA fragments is derived) are set to their expected values under the previous model parameters, and then the model’s parameters are assigned to maximize the likelihood conditional on the assumed values of those latent variables. The two-step process is then repeated until convergence.

[0081] In some examples, the analytics system trains a multinomial logistic regression classifier on the training data for a fold, and generates predictions for the held-out data. For example, for each of the K folds, one logistic regression can be trained for each combination of hyperparameters. Such hyperparameters can include L2 penalty and / or topK (e.g., the number of high-ranking regions to keep per tissue type pair (including non-cancer), as ranked by the mutual information procedure outlined above). For each set of hyperparameters, performance is evaluated on the cross-validated predictions of the full training set, and the set of hyperparameters with the best performance is selected for retraining on the full training set. In some examples, the analytics system uses log-loss as a performance metric, whereby the log-loss is calculated by taking the negative logarithm of the prediction for the correct label for each sample, and then summing over samples (i.e. a perfect prediction of 1.0 for the correct label would give a log-loss of 0).

[0082] To generate predictions for a new sample, feature values are calculated using the same method described above, but restricted to features (region / positive class combinations) selected under the chosen topK value. Generated features are then used to create a prediction using the trained logistic regression model.

[0083] In some examples, the analytics trains a two-stage classifier. For example, the analytics system trains a binary cancer classifier to distinguish between the labels, cancer and non-cancer, based on feature vectors of the training samples. In this case, the binary classifier outputs a probability score indicating the likelihood of the presence or absence of cancer. In another example, the analytics system trains a multiclass cancer classifier to distinguish between many cancer types. In this multiclass cancer classifier, the cancer classifier is trained to determine a cancer prediction that comprises a prediction value for each of the cancer types in a plurality of cancer types being classified for. The prediction values can correspond to a likelihood that a given sample has each of the cancer types. For example, the cancer classifier returns a cancer prediction including a prediction value for breast cancer, lung cancer, and non-cancer. For example, the cancer classifier may return a cancer prediction for a test sample including a prediction score for breast cancer, lung cancer, and / or no cancer.

[0084] The analytics system can 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 classifier may be a multinomial logistic regression classifier. In practice either type of cancer classifier may be trained using other techniques, including, but not limited to, LI -regularized logistic regression, a generalized linear model (GLM), random forests, multilayer perceptron, support vector machines, and neural networks. These techniques are numerous including potential use of kernel methods, machine learning algorithms such as multilayer neural networks, etc. In particular, methods as described in PCT / US2019 / 022122 and US20190287652A1 which are incorporated by reference in their entireties herein can be used for various embodiments, particularly those relating to training models based on methylated cfDNA molecules.

[0085] In one specific embodiment, the training a classifier comprises (a) receiving first measured levels of first target molecules for first samples of reference subjects; (b) training a first model to generate a first probability score for the presence of cancer in a subject by applying a first machine learning algorithm to the first measured levels; (c) receiving second measured levels of second target molecules for second samples of the reference subjects; (d) training a second model to generate a second probability score for the presence of cancer in a subject by applying a second machine learning algorithm to the second measured levels; (e) generating reference first cancer probability scores for the first samples using the trained first model; (f) generating reference second cancer probability scores for the second samples using the trained second model; (g) generating reference aggregated cancer probability scores for a plurality of the reference subjects by aggregating the reference first cancer probability score and reference second cancer probability score for each respective reference subject; and (h) training the classifier to generate an aggregatecancer probability score for a subject by applying a third machine learning algorithm to the reference first cancer probability scores, reference second cancer probability scores, and reference aggregated cancer probability scores. In some embodiments, the reference subjects include first subjects having known cancer types, and second subjects without cancer. In some embodiments, aggregating the first cancer probability score and the second cancer probability score comprises calculating a product of the first and second probability scores for the cancer.

[0086] In some embodiments, aggregating the first probability score and the second probability score includes combining the first and second probability scores for the cancer in a linear model. In some embodiments, an aggregate probability of cancer (y) is determined using a linear model according to the following formula:In the above formula: y = final probability of cancer output given DNA and Protein p_DNA = cancer prediction from DNA model p_prot = cancer prediction from Protein model Po = Bias term, constant / intercept Pi = Learned coefficient for DNA probability P2 = Learned coefficient for Protein probability P3 = Learned coefficient for DNA probability and Protein probability product

[0087] The first and the second trained models can apply the machine learning algorithm differently to the levels of each target molecule being analyzed. For instance, the first trained model can binarize the measured levels of the first target molecules, while the second trained model can log-transform the measured levels of the second target molecule. In some embodiments, the first trained model binarizes the measured levels of first target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected. In some embodiments, the first target molecules include cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer types. In some embodiments, the second trained model logtransforms measured levels of the second target molecules normalized against control protein present in a known amount. In some embodiments, the second target molecules include a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types.

[0088] The machine learning algorithms used for training the classifier can be any suitable machine learning algorithm known in the art. In some embodiments, the first machine learningalgorithm, the second machine learning algorithm, and / or the third machine learning algorithm is an LI -regularized logistic regression, an L2-regularized logistic regression, a generalized linear model (GLM), a random forest, a multinomial logistic regression, a multilayer perceptron, a support vector machine, or a neural network. In some embodiments, the third machine learning algorithm is a logistic regression.Trained Models

[0089] The present disclosure provides models for assigning probability scores based on levels of target molecules, methods of training the same, and methods of using the same in conjunction with a trained classifier. In some embodiments, a model is trained using levels of cfDNA (e.g., presence, number, or proportion of cfDNA from target genomic regions having a particular methylation level or methylation state vector, or sequencing reads thereof) for reference samples of subjects with a known cancer diagnosis (e.g., cancer of any type, cancer of a particular type, or healthy / non-cancer). In some embodiments, the model is trained using measured levels of polypeptides for reference samples of subjects with a known cancer diagnosis (e.g., cancer of any type, cancer of a particular type, or healthy / non-cancer). Additional details concerning embodiments for training of a model based on cfDNA methylation levels, and selection of target genomic regions based thereon, follows below. Similar considerations may be applied to the analysis of polypeptide levels, and the selection of target polypeptides based thereon.Generation of data structure

[0090] To create a healthy control group data structure, an analytics system obtains information related to methylation status of a plurality of CpG sites on sequence reads derived from a plurality of DNA molecules or fragments from a plurality of healthy subjects. Methods provided herein for creating a healthy control group data structure can be performed similarly for subjects with cancer, subjects with cancer of a particular tissue of origin (TOO), subjects with a known cancer type, or subjects with another known disease state. A methylation state vector is generated for each DNA molecule or fragment, for example via the process illustrated in FIG. IB.

[0091] In some embodiments, an analytics system subdivides the methylation state vector of each DNA fragment into strings of CpG sites. In one embodiment, the analytics system subdivides 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 7 being subdivided into strings of length less than or equal to 4 would result in 4 strings of length 4, 5 strings of length 3, 6 strings of length2, and 7 strings of length 1. If the methylation state vector resulting from a DNA fragment 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 CpG sites of the vector.

[0092] In some embodiments, the analytics system tallies 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 a string length of three at a given CpG site, there are 2A3 or 8 possible string configurations. For each CpG site, the analytics system tallies how many occurrences of each possible methylation state vector appear in the control group. 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 a data structure storing the tallied counts for each starting CpG site and string possibility at each starting CpG.

[0093] 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 there are at most 2A4 numbers to tally at every CpG. Increasing the maximum string length to 5 doubles the possible number of methylation states to tally. Reducing string size helps reduce the computational and data storage burden of the data structure. In some embodiments, the string size is 3. In some embodiments, the string size is 4. A second reason to limit the maximum string length is to avoid overfitting downstream models. Calculating probabilities based on long strings of CpG sites can be problematic if the long CpG strings do not have a strong biological effect on the outcome (e.g., predictions of anomalousness that predictive of the presence of cancer), as it requires a significant amount of data that may not be available, and would thus be too sparse for a model to perform appropriately. For example, calculating a probability of anomalousness / cancer conditioned on the prior 100 CpG sites would require 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 will be insufficient data to determine whether a given string of length of 100 in a test sample is anomalous or not.Validation of data structure

[0094] Once the data structure has been created, the analytics system may seek to validate the data structure and / or any downstream models making use of the data structure.

[0095] This first type of validation ensures that potential cancerous samples are removed from the healthy control group so as to not affect the control group’s purity. This type of validationchecks consistency within the control group’s data structure. For example, the healthy control group may contain a sample from an individual with an undiagnosed cancer that contains a plurality of anomalously methylated fragments. The analytics system may perform various calculations to determine whether to exclude data from a subject with apparently undiagnosed cancer.

[0096] A second type of validation checks the probabilistic model used to calculate p-values with the counts from the data structure itself (e.g., from the healthy control group). Once the analytics system generates a p-value for the methylation state vectors in the validation group, the analytics system builds a cumulative density function (CDF) with the p-values. With the CDF, the analytics system may perform various calculations on the CDF to validate the control group’s data structure. One test uses the fact that the CDF should ideally be at or below an identity function, such that CDF(x) < x. On the converse, being above the identity function reveals some deficiency within the probabilistic model used for the control group’s data structure. For example, if 1 / 100 of fragments have a p-value score of 1 / 1000 meaning CDF(l / 1000) = 1 / 100 > 1 / 1000, then the second type of validation fails indicating an issue with the probabilistic model. See e.g., U.S. Publ. No. 2019 / 0287652, which is hereby incorporated by reference in its entirety.

[0097] A third type of validation uses a healthy set of validation samples separate from those used to build the data structure. This tests if the data structure is properly built and the model works. The third type of validation can quantify how well the healthy control group generalizes the distribution of healthy samples. If the third type of validation fails, then the healthy control group does not generalize well to the healthy distribution.

[0098] A fourth type of validation tests with samples from a non-healthy validation group. The analytics system calculates p-values and builds the CDF for the non-healthy validation group. With a non-healthy validation group, the analytics system expects to see the CDF(x) > x for at least some samples or, stated differently, the converse of what was expected in the second type of validation and the third type of validation with the healthy control group and the healthy validation group. If the fourth type of validation fails, then this is indicative that the model is not appropriately identifying the anomalousness that it was designed to identify.

[0099] In embodiments including a step of validating the data structure, the analytics system performs the fourth type of validation test as described above which utilizes a validation group with a supposedly similar composition of subjects, samples, and / or fragments as the control group. For example, if the analytics system selected healthy subjects without cancer for the control group, then the analytics system also uses healthy subjects without cancer in the validation group.

[0100] In some embodiments, the analytics system takes the validation group and generates a set of methylation state vectors. The analytics system performs a p-value calculation for eachmethylation state vector from the validation group. For each possible methylation state vector, the analytics system calculates a probability from the control group’s data structure. Once the probabilities are calculated for the possibilities of methylation state vectors, the analytics system calculates a p-value score for that methylation state vector based on the calculated probabilities. The p-value score represents an expectedness of finding that specific methylation state vector and other possible methylation state vectors having even lower probabilities in the control group. A low p-value score, thereby, generally corresponds to a methylation state vector which is relatively unexpected in comparison to other methylation state vectors within the control group, whereas a high p-value score generally corresponds to a methylation state vector which is relatively more expected in comparison to other methylation state vectors found in the control group. Once the analytics system generates a p-value score for the methylation state vectors in the validation group, the analytics system builds a cumulative density function (CDF) with the p-value scores from the validation group. The analytics system validates consistency of the CDF as described above in the fourth type of validation tests.Anomalously methylated fragments

[0101] Anomalously methylated fragments having abnormal methylation patterns in cancer patient samples, subjects with cancer of a TOO, subjects with a known cancer type, or subjects with another known disease state, are selected as target genomic regions, according to an embodiment. In some embodiments, an analytics system generates methylation state vectors from cfDNA fragments of the sample. The analytics system may handle each methylation state vector as follows.

[0102] For a given methylation state vector, the analytics system enumerates all possibilities of methylation state vectors having the same starting CpG site and same length (a set of CpG sites) in the methylation state vector. As each methylation state may be methylated or unmethylated there are only two possible states at each CpG site, and thus the count of distinct possibilities of methylation state vectors depends on a power of 2, such that a methylation state vector of length n would be associated with 2npossibilities of methylation state vectors.

[0103] The analytics system calculates the probability of observing each possibility of methylation state vector for the identified starting CpG site / methylation state vector length by accessing the healthy control group data structure. In one embodiment, calculating the probability of observing a given possibility uses a Markov chain probability to model the joint probability calculation. In some embodiments, calculation methods other than Markov chain probabilities are used to determine the probability of observing each possibility of methylation state vector.

[0104] The analytics system calculates a p-value score for the methylation state vector using the calculated probabilities for each possibility. In one embodiment, this includes identifying the calculated probability corresponding to the possibility that matches the methylation state vector in question. Specifically, this is 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 sums the calculated probabilities of any possibilities having probabilities less than or equal to the identified probability to generate the p-value score.

[0105] The p-value represents 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, thereby, generally corresponds to a methylation state vector which is rare in a healthy subject, and which causes the fragment to be labeled abnormally methylated, relative to the healthy control group. A high p-value score generally relates to a methylation state vector 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 indicates that the fragment is abnormally methylated relative to the non-cancer group, and therefore possibly indicative of the presence of cancer in the test subject.

[0106] As above, the analytics system calculates 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 abnormally methylated, the analytics system may filter the set of methylation state vectors based on their p-value scores. In one embodiment, 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 could be on the order of 0.1, 0.01, 0.001, 0.0001, or similar.P-value score calculation

[0107] To calculate a p-value score given a test methylation state vector, an analytics system takes that test methylation state vector and enumerates possibilities of methylation state vectors. For example, a test methylation state vector of < M23, M24, M25, U26 > has a length of 4, with 2A4 possibilities of methylation state vectors encompassing the indicated CpG sites 23 - 26. In a generic example, the number of possibilities of methylation state vectors is 2An, where n is the length of the test methylation state vector or alternatively the length of the sliding window (described further below).

[0108] The analytics system calculates probabilities for the enumerated possibilities of methylation state vectors. As methylation is conditionally dependent on methylation status of nearby CpG sites, one way to calculate the probability of observing a given methylation state vector possibility is to use Markov chain model. Generally, a methylation state vector such as <Si,S2, . . . , Sn>, where S denotes the methylation state whether methylated (denoted as M), unmethylated (denoted as U), or indeterminate (denoted as I), has a joint probability that can be expanded using the chain rule of probabilities as: / ’(< 51,52,Markov chain model can be used to make the calculation of the conditional probabilities of each possibility more efficient. In one embodiment, the analytics system selects a Markov chain order k which corresponds to how many prior CpG sites in the vector (or window) to consider in the conditional probability calculation, such that the conditional probability is modeled as P(Sn | Si, . . . , Sn-l ) ~ P(Sn | Sn-k-2, . . . , Sn-1 ).

[0109] To calculate each Markov modeled probability for a possibility of methylation state vector, the analytics system accesses the control group’s data structure, specifically the counts of various strings of CpG sites and states. To calculate P(Mn| Sn-k-2, . . . , Sn-i ), the analytics system takes a ratio of the stored count of the number of strings from the data structure matching < Sn-k-2, . . . , Sn-i, Mn> divided by the sum of the stored count of the number of strings from the data structure matching < Sn-k-2, . . . , Sn-i, Mn> and < Sn-k-2, . . . , Sn-i, Un >. Thus, P(Mn| Sn-k-2, . . . , Sn-i), is calculated ratio having the form:_ # of < Sn-k-2, ..., Sn-1, Mn> _# of < Sn-k-2, ... , Sn-1, Mn> + # of < Sn-k-2, ..., Sn-1, Un>

[0110] The calculation may additionally implement a smoothing of the counts by applying a prior distribution. In one embodiment, the prior distribution is a uniform prior as in Laplace smoothing. As an example of this, a constant is added to the numerator and another constant (e.g., twice the constant in the numerator) is added to the denominator of the above equation. In other embodiments, an algorithmic technique such as Knesser-Ney smoothing is used.

[0111] The above formulas are applied to a test methylation state vector. Once the calculated probabilities are completed, the analytics system calculates a p-value score that sums the probabilities that are less than or equal to the probability of possibility of methylation state vector matching the test methylation state vector.

[0112] In one embodiment, the computational burden of calculating probabilities and / or p-value scores may be further reduced by caching at least some calculations. For example, the analytic 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 allows for efficient calculation of p-value scores 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.Sliding window

[0113] In one embodiment, an analytics system uses 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 enumerates 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 would serve no purpose). The window length may be static, user determined, dynamic, or otherwise selected.

[0114] In calculating p-values for a methylation state vector larger than the window, the window identifies the sequential set of CpG sites from the vector within the window starting from the first CpG site in the vector. The analytic system calculates a p-value score for the window including the first CpG site. The analytics system then “slides” 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 will generate m l+l p-value scores. After completing the p-value calculations for each portion of the vector, the lowest p-value score from all sliding windows is taken as the overall p-value score for the methylation state vector. In another embodiment, the analytics system aggregates the p-value scores for the methylation state vectors to generate an overall p-value score.

[0115] Using the sliding window helps to reduce the number of enumerated possibilities of methylation state vectors and their corresponding probability calculations that would otherwise need to be performed. In general, the number of possibilities of methylation state vectors increases exponentially by a factor of 2 with the size of the methylation state vector. To give a realistic example, it is possible for fragments to have upwards of 54 CpG sites. Instead of computing probabilities for 2A54 (~1.8x 10Al 6) possibilities to generate a single p-value, 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 enumerates 2A5 (32) possibilities of methylation state vectors, which total results in 50*2A5(1.6>< 10A3) probability calculations. This results in a vast reduction of calculations to be performed, with no meaningful hit to the accurate identification of anomalous fragments. This additional step can also be applied when validating the control group with the validation group’s methylation state vectors.Identifying fragments indicative of cancer

[0116] In some embodiments, the analytics system identifies DNA fragments indicative of cancer from the filtered set of anomalously methylated fragments. In some embodiments, fragments identified as indicative of cancer are used in the selection of target genomic regions for analysis of subsequent samples. For example, data used in training may include information for a set of genomic regions, and a subset of these regions are selected for analysis in test samples based on how informative they are determined to be. Selection of a target genomic region for analysis may performed by a sample processing step (e.g., selective capture of cfDNA fragments from the target genomic regions using bait oligonucleotides), or computationally (e.g., by ignoring sequencing reads for cfDNA fragments originating outside of a set of target genomic regions).Hypomethylated and hypermethylated fragments

[0117] In some embodiments, an analytics system may identify DNA fragments that are deemed hypomethylated or hypermethylated as fragments indicative of cancer from a filtered set of anomalously methylated fragments. Hypomethylated and hypermethylated fragments can be defined as fragments of a certain length of CpG sites (e.g., more than 3, 4, 5, 6, 7, 8, 9, 10, etc.) with a high percentage of methylated CpG sites (e.g., more than 80%, 85%, 90%, or 95%, or any other percentage over 50%) or a high percentage of unmethylated CpG sites (e.g., more than 80%, 85%, 90%, or 95%, or any other percentage over 50%).Probabilistic models

[0118] In some embodiments, an analytics system identifies fragments indicative of cancer utilizing probabilistic models of methylation patterns fitted to each cancer type and non-cancer type. The analytics system calculates log-likelihood ratios for a sample using DNA fragments in the genomic regions considering the various cancer types with the fitted probabilistic models for each cancer type and non-cancer type. The analytics system may determine a DNA fragment to be indicative of cancer based on whether at least one of the log-likelihood ratios considered against the various cancer types is above a threshold value.

[0119] In some embodiments, the analytics system partitions the genome into regions by multiple stages. In a first stage, the analytics system separates the genome into blocks of CpG sites.Each block is defined when there is a separation between two adjacent CpG sites that meets and / or exceeds some threshold, e.g., greater than 200 bp, 300 bp, 400 bp, 500 bp, 600 bp, 700 bp, 800 bp, 900 bp, or 1,000 bp. From each block, the analytics system subdivides at a second stage each block into regions of a certain length, e.g., 500 bp, 600 bp, 700 bp, 800 bp, 900 bp, 1,000 bp, 1,100 bp, 1,200 bp, 1,300 bp, 1,400 bp, or 1,500 bp. The analytics system may further overlap adjacent regions by a percentage of the length, e.g., 10%, 20%, 30%, 40%, 50%, or 60%, or 10% or more, 20% or more, 30% or more, 40% or more, 50% or more, or 60% or more.

[0120] The analytics system may analyze sequence reads derived from DNA fragments for each region. The analytics system may process samples from tissue and / or high-signal cfDNA. High- signal cfDNA samples may be determined by a binary classification model, by cancer stage, or by another metric.

[0121] For each cancer type and non-cancer, the analytics system fits a separate probabilistic model for fragments. In one example, each probabilistic model is mixture model comprising a combination of a plurality of mixture components with each mixture component being an independent-sites model where methylation at each CpG site is assumed to be independent of methylation statuses at other CpG sites.

[0122] In some embodiments, calculation is performed with respect to each CpG site. Specifically, a first count is determined that is the number of cancerous samples (cancer count) that include an anomalously methylated DNA fragment overlapping that CpG, and a second count is determined that is the total number of samples containing fragments overlapping that CpG (total) in the set. Genomic regions can be selected based on the numbers, for example, based on criteria positively correlated to the number of cancerous samples (cancer count) that include a DNA fragment overlapping that CpG, and inversely correlated to the total number of samples containing fragments overlapping that CpG (total) in the set.

[0123] In some embodiments, cancer of various types having different TOO are selected from anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0124] The analytics system can further calculate log-likelihood ratios (“R”) for a fragment indicating a likelihood of the fragment being indicative of cancer considering the various cancer types with the fitted probabilistic models for each cancer type and non-cancer type, or for a cancer TOO. The two probabilities may be taken from probabilistic models fitted for each of the cancer types and the non-cancer type, the probabilistic models defined to calculate a likelihood of observing a methylation pattern on a fragment given each of the cancer types and the non-cancertype. For example, the probabilistic models may be defined fitted for each of the cancer types and the non-cancer type.Selection of genomic regions indicative of cancer

[0125] In some embodiments, the analytics system can identify genomic regions indicative of cancer. To identify these informative regions, the analytics system calculates an information gain for each genomic region or more specifically each CpG site that describes an ability to distinguish between various outcomes.

[0126] A method for identifying genomic regions capable of distinguishing between cancer type and non-cancer type utilizes a trained classification model that can be applied on the set of anomalously methylated DNA molecules or fragments corresponding to or derived from a cancerous or non-cancerous group. The trained classification model can be trained to identify any condition of interest that can be identified from the methylation state vectors.

[0127] In one embodiment, the trained classification model is a binary model trained based on methylation states for cfDNA fragments or genomic sequences obtained from a subj ect cohort with cancer or a cancer TOO, and a healthy subject cohort without cancer, and is then used to classify a test subject probability of having cancer, a cancer TOO, or not having cancer, based on anomalously methylation state vectors. In other embodiments, different models may be trained using subject cohorts known to have particular cancer (e.g., breast, lung, prostrate, etc.); known to have cancer of particular TOO where the cancer is believed to originate; or known to have different stages of particular cancer (e.g., breast, lung, prostrate, etc.). In these embodiments, different models may be trained using sequence reads obtained from samples enriched for tumor cells from subject cohorts known to have particular cancer (e.g., breast, lung, prostrate, etc.). Each genomic region’s ability to distinguish between cancer type and non-cancer type in the classification model is used to rank the genomic regions from most informative to least informative in classification performance. The analytics system may identify genomic regions from the ranking according to information gain in classification between non-cancer type and cancer type.Computing information gain from hypomethylated and hypermethylated fragments indicative of cancer

[0128] In some embodiments, an analytics system trains a model using fragments indicative of cancer. The process accesses two training groups of samples - a non-cancer group and a cancer group - and obtains a non-cancer set of methylation state vectors and a cancer set of methylation state vectors comprising anomalously methylated fragments.

[0129] The analytics system may determine, for each methylation state vector, whether the methylation state vector is indicative of cancer. Here, fragments indicative of cancer may be defined as hypermethylated or hypomethylated fragments determined if at least some number of CpG sites have a particular state (methylated or unmethylated, respectively) and / or have a threshold percentage of sites that are the particular state (again, methylated or unmethylated, respectively). In one example, cfDNA fragments are identified as hypomethylated or hypermethylated, respectively, if the fragment overlaps at least 5 CpG sites, and at least 80%, 90%, or 100% of its CpG sites are methylated or at least 80%, 90%, or 100% are unmethylated.

[0130] In some embodiments, the analytics system considers portions of the methylation state vector and determines whether the portion is hypomethylated or hypermethylated, and may distinguish that portion to be hypomethylated or hypermethylated. This alternative resolves missing methylation state vectors which are large in size but contain at least one region of dense hypomethylation or hypermethylation. In some embodiments, the fragments indicative of cancer may be defined according to likelihoods outputted from trained probabilistic models.

[0131] In one embodiment, the analytics system generates a hypomethylation score (Phypo) and a hypermethylation score (Phyper) per CpG site in the genome. To generate either score at a given CpG site, the model takes four counts at that CpG site - (1) count of (methylations state) vectors of the cancer set labeled hypomethylated that overlap the CpG site; (2) count of vectors of the cancer set labeled hypermethylated that overlap the CpG site; (3) count of vectors of the noncancer set labeled hypomethylated that overlap the CpG site; and (4) count of vectors of the noncancer set labeled hypermethylated that overlap the CpG site. Additionally, the process may normalize these counts for each group to account for variance in group size between the non-cancer group and the cancer group. In some embodiments wherein fragments indicative of cancer are more generally used, the scores may be more broadly defined as counts of fragments indicative of cancer at each genomic region and / or CpG site.

[0132] In some embodiments, to generate the hypomethylation score at a given CpG site, the process takes a ratio of (1) over (1) summed with (3). Similarly, the hypermethylation score may be calculated by taking a ratio of (2) over (2) and (4). Additionally, these ratios may be calculated with an additional smoothing technique as discussed above. The hypomethylation score and the hypermethylation score relate to an estimate of cancer probability given the presence of hypomethylation or hypermethylation of fragments from the cancer set.

[0133] In some embodiments, the analytics system generates an aggregate hypomethylation score and an aggregate hypermethylation score for each anomalous methylation state vector. The aggregate hyper and hypo methylation scores are determined based on the hyper and hypo methylation scores of the CpG sites in the methylation state vector. In one embodiment, theaggregate hyper and hypo methylation scores are assigned as the largest hyper and hypo methylation scores of the sites in each state vector, respectively. In some embodiments, the aggregate scores could be based on means, medians, or other calculations that use the hyper / hypo methylation scores of the sites in each vector.

[0134] In some embodiments, the analytics system ranks all of that subject’s methylation state vectors by their aggregate hypomethylation score and by their aggregate hypermethylation score, resulting in two rankings per subject. The process selects aggregate hypomethylation scores from the hypomethylation ranking and aggregate hypermethylation scores from the hypermethylation ranking. With the selected scores, the model generates a single feature vector for each subject. In one embodiment, the scores selected from either ranking are selected with a fixed order that is the same for each generated feature vector for each subject in each of the training groups. As an example, in one embodiment the model selects the first, the second, the fourth, and the eighth aggregate hyper methylation score, and similarly for each aggregate hypo methylation score, from each ranking and writes those scores in the feature vector for that subject.

[0135] In some embodiments, the analytics system trains a binary model to distinguish feature vectors between the cancer and non-cancer training groups. Generally, any one of a number of classification techniques may be used. In one embodiment the model is a non-linear classifier. In a specific embodiment, the classifier is a non-linear classifier utilizing a L2-regularized kernel logistic regression with a Gaussian radial basis function (RBF) kernel.

[0136] Specifically, in one embodiment, the number of non-cancer samples or different cancer type(s) (nother) and the number of cancer samples or cancer type(s) (ncancer) having an anomalously methylated fragment overlapping a CpG site are counted. Then the probability that a sample is cancer is estimated by a score (“S”) that positively correlates to ncancer and inversely correlated to nother. The score can be calculated using the equation; (ncancer + 1) / (ncancer + nother + 2) or (n cancer) / (ncancer + nother). The analytics system computes an information gain for each cancer type and for each genomic region or CpG site to determine whether the genomic region or CpG site is indicative of cancer. The information gain is computed for training samples with a given cancer type compared to all other samples. For example, two random variables ‘anomalous fragment’ (‘AF’) and ‘cancer type’ (‘CT’) are used. In on embodiment, AF is a binary variable indicating whether there is an anomalous fragment overlapping a given CpG site in a given samples as determined for the anomaly score / feature vector above. CT is a random variable indicating whether the cancer is of a particular type. The analytics system computes the mutual information with respect to CT given AF. That is, how many bits of information about the cancer type are gained if it is known whether there is an anomalous fragment overlapping a particular CpG site.

[0137] For a given cancer type, the analytics system may use this information to rank CpG sites based on how cancer specific they are. This procedure is repeated for all cancer types 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 types or in healthy training samples, then CpG sites overlapped by those anomalous fragments will tend to have high information gains for the given cancer type. The ranked CpG sites for each cancer type are greedily added (selected) to a selected set of CpG sites based on their rank for use in the trained model.Computing pairwise information gain from fragments indicative of cancer identified from probabilistic models

[0138] With fragments indicative of cancer identified, the analytics may then be used to identify genomic regions. The analytics system defines a feature vector for each sample, for each region, for each cancer type by a count of DNA fragments that have a calculated log-likelihood ratio that the fragment is indicative of cancer above a plurality of thresholds, wherein each count is a value in the feature vector. In some embodiments, the analytics system counts the number of fragments present in a sample at a region for each cancer type with log-likelihood ratios above one or a plurality of possible threshold values. In some embodiments, the analytics system defines a feature vector for each sample, by a count of DNA fragments for each genomic region for each cancer type that provides a calculated log-likelihood ratio for the fragment above a plurality of thresholds, wherein each count is a value in the feature vector. The analytics system may use the defined feature vectors to calculate an informative score for each genomic region describing that genomic region’s ability to distinguish between each pair of cancer types. For each pair of cancer types, the analytics system ranks regions based on the informative scores. The analytics system may select regions based on the ranking according to informative scores.

[0139] In some embodiments, the analytics system calculates an informative score for each region describing that region’s ability to distinguish between each pair of cancer types. For each pair of distinct cancer types, the analytics system may specify one type as a positive type and the other as a negative type. In some embodiments, a region’s ability to distinguish between the positive type and the negative type is based on mutual information, calculated using the estimated fraction of cfDNA samples of the positive type and of the negative type for which the feature would be expected to be non-zero in the final assay, i.e., at least one fragment of that tier that would be sequenced in a targeted methylation assay. Those fractions are estimated using the observed rates at which the feature occurs in healthy cfDNA, and in high-signal cfDNA and / or tumor samples of each cancer type. For example, if a feature occurs frequently in healthy cfDNA, then it will also be estimated to occur frequently in cfDNA of any cancer type and would likelyresult in a low informative score. The analytics system may choose a certain number of regions for each pair of cancer types from the ranking.

[0140] In some embodiments, the analytics system further identifies predominantly hypermethylated or hypomethylated regions from the ranking of regions. The analytics system may load the set of fragments in the positive type(s) for a region that was identified as informative. The analytics system, from the loaded fragments, evaluates whether the loaded fragments are predominantly hypermethylated or hypomethylated. If the loaded fragments are predominately hypermethylated or hypomethylated, the analytics system may select probes corresponding to the predominant methylation pattern. If the loaded fragments are not predominantly hypermethylated or hypomethylated, the analytics system may use a mixture of probes for targeting both hypermethylation and hypomethylation. The analytics system may further identify a minimal set of CpG sites that overlap more than some percentage of the fragments.

[0141] In some embodiments, the analytics system, after ranking the regions based on informative scores, labels each region with the lowest informative ranking across all pairs of cancer types. For example, if a region was the lOth-most-informative region for distinguishing breast from lung, and the 5th-most-informative for distinguishing breast from colorectal, then it would be given an overall label of “5”. The analytics system may design probes starting with the lowest- labeled regions while adding regions to the panel, e.g., until the panel’s size budget has been exhausted.Trained model and featurization

[0142] In some examples, an assay panel is used with a trained model that predicts a disease state for a sample, such as a cancer or non-cancer prediction, a tissue of origin prediction, and / or an indeterminate prediction. In some examples, the cancer type model can generate features based on sequence reads by taking into account methylated or unmethylated fragments of DNA at certain genomic areas of interest. For instance, if the cancer type model determines that a methylation pattern at a fragment resembles that of a certain cancer type, then the cancer type model can set a feature for that fragment as 1, and otherwise if no such fragment is present, then the feature can be set as 0. In this way, the cancer type model can produce a set of binary features (merely by way of example, 30,000 features) for each sample. Further, in some examples, all or a portion of the set of binary features for a sample can be input into the cancer type model to provide a set of probability scores, such as one probability score per cancer type class and for a non-cancer type class. Furthermore, in some examples, the cancer type model can incorporate or otherwise be used in conjunction with thresholding to determine whether a sample is to be called as cancer or non-cancer, and / or indeterminate thresholding to reflect confidence in a specific TOO call. Such methods are described further below.

[0143] To train the cancer type model, the analytics system can obtain a set of training samples. In some examples, each training sample includes fragment file(s) (e.g., file containing sequence read data), a label corresponding to a type of cancer (TOO) or non-cancer status of the sample, and / or sex of the individual of the sample. The analytics system can utilize the training set to train the cancer type classifier to predict the disease state of the sample.

[0144] In some examples, for training, the analytics system divides the genome (e.g., whole genome) or a subset of the genome (e.g., targeted methylation regions) into regions. Merely by way of example, portions of the genome can be separated into “blocks” of CpGs, whereby a new block begins whenever there is a separation between nearest-neighbor CpGs is at least a minimum separation distance (e.g., at least 500 bp). Further, in some examples, each block can be divided into 1000 bp regions and positioned such that neighboring regions have a certain amount (e.g., 50% or 500 bp) of overlap.

[0145] Furthermore, in some examples, the analytics system can split the training set into K subsets or folds to be used in a K-fold cross-validation. In some examples, the folds can be balanced for cancer / non-cancer status, tissue of origin, cancer stage, and / or age (e.g., grouped in lOyr buckets). In some examples, the training set is split into 5 folds, whereby 5 separate classifiers are trained, in each case training on 4 / 5 of the training samples and using the remaining 1 / 5 for validation.

[0146] During training with the training set, the analytics system can, for each cancer type (and for healthy cfDNA), fit a probabilistic model to the fragments deriving from the samples of that type. During training, the analytics system fits sequence reads derived from one or more samples from subjects having a known disease and can be used to determine sequence reads probabilities indicative of a disease state utilizing methylation information or methylation state vectors. In particular, in some cases, the analytics system determines observed rates of methylation for each CpG site within a sequence read. The rate of methylation represents a fraction or percentage of base pairs that are methylated within a CpG site. The trained probabilistic model can be parameterized by products of the rates of methylation. In general, any known probabilistic model for assigning probabilities to sequence reads from a sample can be used. For example, the probabilistic model can be a binomial model, in which every site (e.g., CpG site) on a nucleic acid fragment is assigned a probability of methylation, or an independent sites model, in which each CpG’s methylation is specified by a distinct methylation probability with methylation at one site assumed to be independent of methylation at one or more other sites on the nucleic acid fragment.

[0147] In some examples, the probabilistic model is a Markov model, in which the probability of methylation at each CpG site is dependent on the methylation state at some number of preceding CpG sites in the sequence read, or nucleic acid molecule from which the sequence read is derived. See, e.g., US20190287652A1, which is incorporated by reference in its entirety herein and can be used for various embodiments.

[0148] In some examples, the probabilistic model is a “mixture model” fitted using a mixture of components from underlying models. For example, in some embodiments, the mixture components can be determined using multiple independent sites models, where methylation (e.g., rates of methylation) at each CpG site is assumed to be independent of methylation at other CpG sites. Utilizing an independent sites model, the probability assigned to a sequence read, or the nucleic acid molecule from which it derives, is the product of the methylation probability at each CpG site where the sequence read is methylated and one minus the methylation probability at each CpG site where the sequence read is unmethylated. In accordance with this example, the analytics system determines rates of methylation of each of the mixture components. The mixture model is parameterized by a sum of the mixture components each associated with a product of the rates of methylation. A probabilistic model Pr of n mixture components can be represented as:For an input fragment, m, G {0, 1} represents the fragment’s observed methylation status at position z of a reference genome, with 0 indicating unmethylation and 1 indicating methylation. A fractional assignment to each mixture component k is fk, where fk> 0 and k=i fk=1 The probability of methylation at position z in a CpG site of mixture component k is / 3ki. Thus, the probability of unmethylation is 1 — / 3ki. The number of mixture components n can be 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, etc.

[0149] In some embodiments, the analytics system fits the probabilistic model using maximumlikelihood estimation to identify a set of parameters { / 3ki, fk} that maximizes the log-likelihood of all fragments deriving from a disease state, subject to a regularization penalty applied to each methylation probability with regularization strength r. The maximized quantity for N total fragments can be represented as:

[0150] In some embodiments, the analytics system performs fits separately for each cancer type and for healthy cfDNA. According to various aspects of the subject disclosure, other means can be used to fit the probabilistic models or to identify parameters that maximize the log-likelihood of all sequence reads derived from the reference samples. For example, in some examples, Bayesian fitting (using e.g., Markov chain Monte Carlo), in which each parameter is not assigned a single value but instead is associated to a distribution, is used. In some examples, gradient-based optimization, in which the gradient of the likelihood (or log-likelihood) with respect to the parameter values is used to step through parameter space towards an optimum, is used. In still some examples, expectation-maximization, in which a set of latent parameters (such as identities of the mixture component from which each fragment is derived) are set to their expected values under the previous model parameters, and then the model’s parameters are assigned to maximize the likelihood conditional on the assumed values of those latent variables. The two-step process is then repeated until convergence.

[0151] Further, in some examples, the analytics system can generate features for each sample in the training set. For example, for each sample (regardless of label), in each region, for each of a plurality of cancer types, for each fragment, the analytics system can evaluate the log-likelihood ratio R with the fitted probabilistic models according to:_ / Prfragment | cancer type 4) \Rcancer type A fragment) = In ^pr^ragmen^healthy cfDNA)JNext, for each sample, for each region, for each cancer type, for each of a set of “tier” values, the analytics system can count the number of fragments with Rcancer type > tier and assign those counts as non-negative integer-valued features. For example, the tiers include threshold values of 1, 2, 3, 4, 5, 6, 7, 8, and 9, resulting in each region hosting 9 features per cancer type.

[0152] In some examples, the analytics system can select certain features for inclusion in a feature vector for each sample. For example, for each pair of distinct cancer types, the analytics system can specify one type as the “positive type” and the other as the “negative type” and rank the features by their ability to distinguish those types. In some cases, the ranking is based on mutual information calculated by the analytics system. For example, the mutual information can be calculated using the estimated fraction of samples of the positive type and negative type (e.g., cancer types A and B) for which the feature is expected to be nonzero in a resulting assay. For instance, if a feature occurs frequently in healthy cfDNA, the analytics system determines the feature is unlikely to occur frequently in cfDNA associated with various types of cancer. Consequently, the feature can be a weak measure in distinguishing between disease states. Incalculating mutual information / , the variable is a certain feature (e.g., binary) and variable 7 represents a disease state, e.g., cancer type A or B:p(l|A) = fA+ fH- fHfAThe joint probability mass function of X and Y is p(x,y) and the marginal probability mass functions are p(x) and p(y). The analytics system can assume that feature absence is uninformative and either disease state is equally likely a priori, for example, p(Y = A~) = p(Y = B) = 0.5. The probability of observing (e.g., in cfDNA) a given binary feature of cancer type A is represented by p(l| ), where fAis the probability of observing the feature in ctDNA samples from tumor (or high-signal cfDNA samples) associated with cancer type A, and fHis the probability of observing the feature in a healthy or non-cancer cfDNA sample.

[0153] In some embodiments, only features corresponding to the positive type are included in the ranking, and only when those features’ predicted rate of occurrence is greater in the positive type than in the negative type. For example, if “liver” is the positive type and “breast” is the negative type, then only “liver x” features are considered, and only if their estimated occurrence in liver cfDNA is greater than their estimated occurrence in breast cfDNA. Further, in some examples, for each region, for each cancer type pair (including non-cancer as a negative type), the analytics system keeps only the best performing tier. Further, in some examples, the analytics system transforms feature values by binarization, whereby any feature value greater than 0 is set to 1, such that all features are either 0 or 1.

[0154] In some examples, the analytics system trains a multinomial logistic regression model on the training data for a fold, and generates predictions for the held-out data. For example, for each of the K folds, one logistic regression can be trained for each combination of hyperparameters. Such hyperparameters can include L2 penalty and / or topK (e.g., the number of high-ranking regions to keep per tissue type pair (including non-cancer), as ranked by the mutual information procedure outlined above). For each set of hyperparameters, performance is evaluated on the cross-validated predictions of the full training set, and the set of hyperparameters with the best performance is selected for retraining on the full training set. In some examples, the analytics system uses log-loss as a performance metric, whereby the log-loss is calculated by taking thenegative logarithm of the prediction for the correct label for each sample, and then summing over samples (i.e. a perfect prediction of 1.0 for the correct label would give a log-loss of 0).

[0155] To generate predictions for a new sample, feature values are calculated using the same method described above, but restricted to features (region / positive class combinations) selected under the chosen topK value. Generated features are then used to create a prediction using the logistic regression model trained above.

[0156] In some examples, the analytics trains a two-stage model. For example, the analytics system trains a binary cancer model to distinguish between the labels, cancer and non-cancer, based on the feature vectors of the training samples. In this case, the binary model outputs a prediction score indicating the likelihood of the presence or absence of cancer. In another example, the analytics system trains a multiclass cancer model to distinguish between many cancer types. In this multiclass cancer model, the cancer model is trained to determine a cancer prediction that comprises a prediction value for each of the cancer types being classified for. The prediction values can correspond to a likelihood that a given sample has each of the cancer types. For example, the cancer model returns a cancer prediction or probability score including a prediction value for breast cancer, lung cancer, and non-cancer. For example, the cancer model may return a cancer prediction or probability score for a test sample including a prediction score for breast cancer, lung cancer, and / or no cancer.

[0157] The analytics system can train the model according to any one of a number of methods. As an example, the binary cancer model may be a L2-regularized logistic regression classifier that is trained using a log-loss function. As another example, the multi-cancer (TOO) model may be a multinomial logistic regression. In practice either type of cancer model may be trained using other techniques. These techniques are numerous including potential use of kernel methods, machine learning algorithms such as multilayer neural networks, etc. In particular, methods as described in US20190287652A1, which are incorporated by reference in their entireties herein, can be used for various embodiments. Still further, in some embodiments, the TOO model is trained only on cancer samples that were successfully called as cancer by the binary model, thereby ensuring sufficient cancer signal in the cancer sample. In some examples, the binary model is trained on the training samples regardless of TOO.Detection of Methylated Nucleic Acid

[0158] In some embodiments, methods described herein comprise detecting a methylation pattern in nucleic acid molecules (e.g., cfDNA molecules). Any suitable method known in the art can be used to detect and analyze methylation patterns in nucleic acids.

[0159] FIG. 1A is a flowchart of an exemplary process 100 for processing a nucleic acid sample and generating methylation state vectors for DNA fragments, according to some embodiments. Methods may include, but are not limited to, one or more of the following steps. For example, any step of the method may comprise a quantitation sub-step for quality control or other laboratory assay procedures known to one skilled in the art.

[0160] In step 105, a sample comprising nucleic acid (e.g., cfDNA) is collected from a subject. The sample may be any subset of the human genome, including the whole genome. The sample may include blood, plasma, serum, urine, fecal, saliva, other types of bodily fluids, or any combination thereof. In some embodiments, methods for drawing a blood sample (e.g., syringe or finger prick) may be less invasive than procedures for obtaining a tissue biopsy, which may require surgery. The extracted sample may comprise cfDNA and / or ctDNA. For healthy individuals, the human body may naturally clear out cfDNA and other cellular debris. If a subject has a cancer or disease, cfDNA and / or ctDNA in a sample may be present at a detectable level for detecting the cancer or disease.

[0161] In step 110, the nucleic acids are treated to differentiate methylated nucleotides and an unmethylated nucleotides, thereby generating the converted cfDNA molecules. In some embodiments, the treatment comprises deamination, such as treating the cfDNA molecules with a cytosine deaminase or with bisulfite. In one embodiment, the method uses a bisulfite treatment of DNA (e.g. cfDNA) which converts unmethylated cytosines to uracils without converting 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).

[0162] In step 115, a sequencing library is prepared. In some embodiments, a ssDNA adapter is added to the 3'-OH end of a bisulfite-converted ssDNA molecule using a ssDNA ligation reaction. In one embodiment, the ssDNA ligation reaction uses CircLigase II (Epicentre) to ligate the ssDNA adapter to the 3'-OH end of a bisulfite-converted ssDNA molecule, wherein the 5 '-end of the adapter is phosphorylated and the bisulfite-converted ssDNA has been dephosphorylated (i.e., the 5' phosphate is removed). In another embodiment, the ssDNA ligation reaction uses Thermostable 5' AppDNA / RNA ligase (available from New England BioLabs (Ipswich, MA)) to ligate the ssDNA adapter to the 3'-OH end of a bisulfite-converted ssDNA molecule. In this example, a first adapter is adenylated at the 5'-end and blocked at the 3'-end. In another embodiment, the ssDNA ligation reaction uses a T4 RNA ligase (available from New EnglandBioLabs) to ligate the ssDNA adapter to the 3'-OH end of a bisulfite-converted ssDNA molecule. In a second step, a second strand DNA is synthesized in an extension reaction. For example, an extension primer, that hybridizes to a primer sequence included in the ssDNA adapter, is used in a primer extension reaction to form a double-stranded bisulfite-converted DNA molecule. Optionally, in one embodiment, the extension reaction uses an enzyme that is able to read through uracil residues in the bisulfite-converted template strand. Optionally, in a third step, a dsDNA adapter is added to the double-stranded bisulfite-converted DNA molecule. Finally, the doublestranded bisulfite-converted DNA is amplified to add sequencing adapters. For example, PCR amplification using a forward primer that includes a P5 sequence and a reverse primer that includes a P7 sequence is used to add P5 and P7 sequences to the bisulfite-converted DNA. Optionally, during library preparation, unique molecular identifiers (UMI) may be added to the nucleic acid molecules (e.g., DNA molecules), such as through adapter ligation or primer extension. In general, a UMI is short nucleic acid sequences (e.g., 4-10 base pairs) that serve as a tag that can be used to facilitate identifying sequence reads originating from a specific DNA fragment. In some embodiments, UMIs comprise degenerate base pair positions. During PCR amplification following adapter ligation, the UMIs are replicated along with the attached DNA fragment, which provides a way to identify sequence reads that came from the same original fragment in downstream analysis (either by the UMI alone, or using the UMI in combination with a portion of the sample nucleic acid fragment end sequence, such as the first 2-10 nucleotides).

[0163] In step 120, targeted genomic regions may be enriched from the library. This is used, for example, where a targeted panel assay is being performed on the samples. During enrichment, hybridization probes (also referred to herein as “probes” or “bait oligonucleotides”) are used to target, and optionally pull down, nucleic acid fragments informative for the presence or absence of cancer (or disease), cancer status, or a cancer classification (e.g., cancer type or tissue of origin). In some embodiments, the probes have features specified herein, such as in connection with various other aspects described herein. For a given workflow, the probes may be designed to anneal (or hybridize) to a target (complementary) strand of DNA (e.g., converted DNA molecules). 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. Moreover, the probes may cover overlapping portions of a target genomic region.

[0164] In some embodiments, the bait oligonucleotides are designed to enrich target genomic regions that comprise at least 1000, 5000, 10000, 20000, or 30000, or more target genomic regions. In some embodiments, the target genomic regions comprise 1000 to 30000, 5000 to 25000, 10000 to 20000, or 12500 to 15000 target genomic regions. In some embodiments, the target genomicregions comprise at least 5000 target genomic regions. In some embodiments, the target genomic regions comprise at least 20000 target genomic regions. In some embodiments, the bait oligonucleotides are designed to target genomic regions that have a collective total size of at least 50 kb, 100 kb, 500 kb, or 1000 kb. In some embodiments, the total size of the target genomic regions (e.g., at least 10000 target genomic regions) is 500 kb to 1000 kb, 100 kb to 500 kb, or 50 kb to 100 kb, or 10 kb to 50 kb. In some embodiments, the total size of the target genomic regions is at least 100 kb. In some embodiments, the total size of the target genomic regions is at least 500 kb. In some embodiments, the total size of the target genomic regions is smaller than the combined length of all different bait oligonucleotides in a panel for enriching target genomic regions, such as when the bait oligonucleotides comprise overlapping sequences. In some embodiments, the total size of the plurality of target genomic regions is given by the total length of the different oligonucleotide probes in the panel for enriching target genomic regions when counting overlapping nucleotide positions only once.

[0165] In some embodiments, the bait oligonucleotides are configured to hybridize to converted DNA molecules (e.g., converted cfDNA molecules) corresponding to, or derived from, one or more genomic regions. Accordingly, the bait oligonucleotides can have a sequence different from the targeted genomic region. For example, a DNA containing an unmethylated CpG site can be converted to include UpG instead of CpG by deamination (e.g., by treatment with a cytosine deaminase or bisulfite). As a result, a probe to such a target may be configured to hybridize to a sequence including UpG instead of a naturally-existing unmethylated CpG. Accordingly, a complementary site in the probe to the unmethylated site can comprise CpA instead of CpG, and some probes targeting a hypomethylated site where all methylation sites are unmethylated may have no guanine (G) bases. In some embodiments, at least 3%, 5%, 10%, 15%, or 20% of the probes comprise no CpG sequences. In some embodiments, at least 5% of the probes comprise no CpG sequences. In some embodiments, at least 10% of the probes comprise no CpG sequences.

[0166] In some embodiments, probes range in length from 10s, 100s, 200s, or 300s of base pairs. The probes can comprise at least 50, 75, 100, or 120 nucleotides. The probes can comprise less than 300, 250, 200, or 150 nucleotides. In an embodiment, the probes comprise 100-150 nucleotides. In one particular embodiment, the probes comprise 120 nucleotides.

[0167] In some embodiments, the probes are designed in a “2* tiled” fashion to cover overlapping portions of a target region. Each probe optionally overlaps in coverage at least partially with another probe in the library. In such embodiments, the panel contains multiple pairs of probes, with each probe in a pair overlapping the other by at least 25, 30, 35, 40, 45, 50, 60, 70, 75 or 100 nucleotides. In some embodiments, the overlapping sequence can be designed to be complementary to a target genomic region (or cfDNA derived therefrom) or to be complementaryto a sequence with homology to a target region or cfDNA. Thus, in some embodiments, at least two probes comprise a sequence with complementarity to the same sequence within a target genomic region, and a nucleotide fragment corresponding to or derived from the target genomic region can be bound and pulled down by at least one of the probes. For a given pair of probes comprising an overlapping sequence, the pair may comprise non-overlapping sequences with complementarity to the target genomic region extending from different ends of the overlapping sequence. Other levels of tiling are possible, such as 3* tiling, 4* tiling, etc., wherein each nucleotide in a target region can bind to more than two probes.

[0168] In some embodiments, a single base in a target genomic region is overlapped by exactly two probes. Probes that extend in both directions beyond a target genomic region are useful to pull down cfDNA fragments comprising a portion of the target genomic region and DNA sequences adjacent to the target genomic region. In some instances, even relatively small target regions may be targeted with three probes. A probe set comprising three or more probes is optionally used to capture a larger genomic region. In some embodiments, subsets of probes will collectively extend across an entire target genomic region (e.g., may be complementary to non-converted or converted fragments from the entire genomic region). A tiled probe set optionally comprises probes that collectively include at least two probes that overlap every nucleotide in the target genomic region. This is done to ensure that cfDNAs comprising a small portion of a target genomic region at one end will have a substantial overlap extending into the adjacent non-targeted genomic region with at least one probe, to provide for efficient capture.

[0169] In some embodiments, each target genomic region is targeted by a set of probes. Probe sets can be designed in a tiled fashion such that adjacent probes have overlapping sequences that hybridize to the same portion of a genomic region. As DNA has two strands, a probe set may also include overlapping probes that hybridize to the other strand, for a total of four probes that hybridize to the same portion of a genomic region. In some embodiments, a set of probes configured to hybridize to a target genomic region does not span the entire region - i.e. at least some sequences within the target genomic region do not have a corresponding probe. For example, a sequence within a target genomic region may be similar or identical to many other sequences in the genome and no probe is designed to target this sequence because such a probe would hybridize to a number of off-target regions above a threshold number.

[0170] For example, a 100 bp cfDNA fragment comprising a 30 nt target genomic region will have at least 65 bp overlap with at least one of the overlapping probes. Other levels of tiling are possible. For example, to increase target size and add more probes in a panel, probes can be designed to expand a 30 bp target region by at least 70 bp, 65 bp, 60 bp, 55 bp, or 50 bp. To capture any fragment that overlaps the target region at all (even if by only Ibp), the probes can be designedto extend past the ends of the target region on either side, such as by at least 50 bp, 55 bp, 60 bp, 65 bp 70 bp, 75 bp, 80 bp, or 85 bp. The probes can be designed to extend past the ends of the target region on either side by 75 bp. In some embodiments, the existence of a probe designed to extend past an end of a target genomic region does not increase the size of the target genomic region (e.g., is not included in determining the size of the respective target genomic region or the collective size of a plurality of target genomic regions).

[0171] In some embodiments, the probes targeting genomic regions differentially methylated between general cancerous (pan-cancer) samples and non-cancerous samples, or only in cancerous samples with a specific cancer type (e.g., lung cancer-specific targets). For example, in some embodiments, a cancer assay panel is designed to include differentially methylated genomic regions based on converted (e.g., bisulfite) sequencing data generated from the cfDNA and / or whole genomic DNA of a set of cancer and non-cancer individuals.

[0172] In some embodiments, each of the target genomic regions is differentially methylated in at least one of a plurality of cancer types. In some embodiments, the plurality of cancer types comprises at least 10 cancer types. In some embodiments, the plurality of cancer types include one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0173] In some embodiments, the target genomic regions can be selected to have at least 3, 5, or 7 methylation sites. In some embodiments, each target genomic regions comprises at least five methylation sites. In some embodiments, the number of methylation sites (e.g., at least 5 methylation sites) are methylation sites that are differentially methylated in at least one cancer of a plurality of cancers.

[0174] Each of the probes (or probe pairs) may be designed to target one or more target genomic regions. The target genomic regions can be selected based on several criteria designed to increase selective enriching of informative cfDNA fragments while decreasing noise and non-specific bindings. Various filtering or modeling procedures for determining whether to include a target genomic region are described herein. In some embodiments, the target genomic regions for enrichment by the bait oligonucleotides are identified by a trained model (e.g., a trained model described herein) as differentially methylated in the at least one of a plurality of cancer types relative to non-cancer tissue or relative to cancer of a different type. In some embodiments, two or more of the filtering or modeling procedures described herein are used in combination.

[0175] The bait oligonucleotides can be designed to enrich target genomic regions located in various positions in a genome, including but not limited to promoters, enhancers, exons, introns,intergenic regions, and other parts. In some embodiments, bait oligonucleotides targeting nonhuman genomic regions, such as those targeting viral genomic regions, can be added.

[0176] In some embodiments, each bait oligonucleotide is conjugated to a solid surface (e.g., a chip or a bead, such as a magnetic or paramagnetic bead) or to a non-nucleotide affinity moiety (e.g., a member of a binding pair). In some embodiments, such conjugation is used to facilitate separation of DNA molecules bound to bait oligonucleotides from unbound DNA molecules. In general, “binding pair” refers to a first and a second moiety, wherein the first and the second moiety have a specific binding affinity for each other. Non-limiting examples of binding pairs include antigens / antibodies; biotin / avidin (or biotin / streptavidin); calmodulin binding protein (CBP) / calmodulin; hormone / hormone receptor; lectin / carbohydrate; peptide / cell membrane receptor; enzyme / cof actor; and enzyme / substrate. In some embodiments, the affinity moiety is biotin.

[0177] Non-limiting examples of target genomic regions, and bait oligonucleotides for enriching the same, are described in US20210025011A1, US20210238693A1,US20220119890A1, US20220064737A1, and US20220098672A1, which are incorporated herein by reference.

[0178] After a hybridization step 120, the hybridized target nucleic acids are enriched (e.g., by capturing or otherwise separating from unbound nucleic acids) and may also be amplified using PCR (enrichment 125). For example, the target nucleic acids can be enriched to obtain enriched sequences that can be subsequently sequenced. In general, a variety of methods can be used to isolate, and enrich for, probe-hybridized target nucleic acids. For example, a biotin moiety can be added to the 5'-end of the probes (i.e., biotinylated) to facilitate isolation of target nucleic acids hybridized to probes using a streptavidin-coated surface (e.g., streptavidin-coated beads).

[0179] In step 130, sequence reads are generated from the enriched nucleic acid fragments. Sequencing data may be acquired from the enriched DNA sequences by known means in the art. For example, the method may include next generation sequencing (NGS) techniques including synthesis technology (Illumina), pyrosequencing (454 Life Sciences), ion semiconductor technology (Ion Torrent sequencing), single-molecule real-time sequencing (Pacific Biosciences), sequencing by ligation (SOLiD sequencing), nanopore sequencing (Oxford Nanopore Technologies), or paired-end sequencing. In some embodiments, massively parallel sequencing is performed using sequencing-by-synthesis with reversible dye terminators.

[0180] In some embodiments, the sequence reads may be aligned to a reference genome using various methods 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.

[0181] In various embodiments, a sequence read is comprised of a read pair denoted asand R2. For example, the first readmay be sequenced from a first end of a nucleic acid fragment whereas the second read R2may be sequenced from the second end of the nucleic acid fragment. Therefore, nucleotide base pairs of the first read R1and 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., / ?x) 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 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 alignment map) format may be generated and output for further analysis.

[0182] From the sequence reads, the location and methylation state for each of CpG site may be determined based on alignment to a reference genome. Further, a methylation state vector for each fragment may be generated 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). The methylation state vectors may be stored in temporary or persistent computer memory for later use and processing. Further, duplicate reads or duplicate methylation state vectors from a single subject may be removed. In an additional embodiment, it may be determined that a certain fragment has one or more CpG sites that have an indeterminate methylation status. Such fragments may be excluded from later processing or selectively included where downstream data model accounts for such indeterminate methylation statuses.

[0183] In step 140, methylation state vectors are generated from the sequence reads. To do so, a sequence read is aligned to a reference genome. The reference genome helps provide the context as to what position in a human genome the DNA fragment (e.g., cfDNA) originates from. In a simplified example, the sequence read is aligned such that the three CpG sites correlate to CpG sites 23, 24, and 25 (arbitrary reference identifiers used for convenience of description). After alignment, there is information both on methylation status of all CpG sites on the cfDNA fragment and which position in the human genome the CpG sites map to. With the methylation status and location, a methylation state vector may be generated for the DNA fragment.

[0184] FIG. IB is an illustration of the exemplary process 100 of FIG. 1A of sequencing a cfDNA fragment to obtain a methylation state vector, according to an embodiment. As an example, the analytics system takes a cfDNA fragment 112. In this example, the cfDNA fragment 112 contains three CpG sites. As shown, the first and third CpG sites of the cfDNA fragment 112 are methylated 114. During the treatment step 120, the cfDNA fragment 112 is converted to generate a converted cfDNA fragment 122. During the treatment 120, the second CpG site which was unmethylated has its cytosine converted to uracil. However, the first and third CpG sites are not converted.

[0185] After conversion, a sequencing library 130 is prepared and sequenced 140 generating a sequence read 142. The analytics system aligns 150 the sequence read 142 to a reference genome 144. The reference genome 144 provides the context as to what position in a human genome the fragment cfDNA originates from. In this simplified example, the analytics system aligns 150 the sequence read 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 thus generates information both on methylation status of all CpG sites on the cfDNA fragment 112 and which to position in the human genome the CpG sites map. As shown, the CpG sites on sequence read 142 which were methylated are read as cytosines. In this example, the cytosine’ s appear in the sequence read 142 only in the first and third CpG site which allows one to infer that the first and third CpG sites in the original cfDNA fragment were methylated. The second CpG site is read as a thymine (U is converted to T during the sequencing process), and thus one can infer that the second CpG site was unmethylated in the original cfDNA fragment. With these two pieces of information, the methylation status and location, the analytics system generates 160 a methylation state vector 152 for the fragment cfDNA 112. In this example, the resulting methylation state vector 152 is < M23, U24, M25 >, wherein M corresponds to a methylated CpG site, U corresponds to an unmethylated CpG site, and the subscript numbers correspond to positions of each CpG site in the reference genome.Detection of Polypeptides

[0186] In some embodiments, the methods described herein comprise detecting (e.g., measuring levels of) one or more polypeptides.

[0187] In some embodiments, the polypeptides comprise at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides. In some embodiments, the polypeptides comprise between 5 and 7500, 10 and 5000, 25 and 3000, 50 and 1000, or 100 and 500 different polypeptides. In some embodiments, the polypeptides comprise at least 20 different polypeptides. In some embodiments, the polypeptides comprise at least 100 different polypeptides. In someembodiments, the polypeptides comprise at least 500 different polypeptides. In some embodiments, the polypeptides comprise at least 3000 different polypeptides.

[0188] In some embodiments, each of the polypeptides is differentially expressed in at least one of a plurality of cancer types. In some embodiments, the plurality of cancer types comprises at least 10 cancer types. In some embodiments, the plurality of cancer types include one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0189] Various suitable methods for detecting one or more target polypeptides are available. Non-limiting examples include competitive and non-competitive immunoassays, enzyme immunoassays (EIA), radioimmunoassays (RIA), antigen capture assays, two-antibody sandwich assays, Western blot analysis, enzyme linked immunosorbent assays (ELISA), colorimetric assays, chemiluminescent assays, fluorescence assays, immunohistochemistry assays, chromatography, liquid chromatography, size exclusion chromatography, high performance liquid chromatography (HPLC), gas chromatography, mass spectrometry, tandem mass spectrometry, matrix assisted laser desorption / ionization-time of flight (MALDI-TOF) mass spectrometry, electrospray ionization (ESI) mass spectrometry, surface-enhanced laser desorption / ionization-time of flight (SELDL TOF) mass spectrometry, quadrupole-time of flight (Q-TOF) mass spectrometry, atmospheric pressure photoionization mass spectrometry (APPL MS), Fourier transform mass spectrometry (FTMS), matrix-assisted laser desorption / ionization- Fourier transform-ion cyclotron resonance (MALDI-FT-ICR) mass spectrometry, secondary ion mass spectrometry (SIMS), microscopy, microfluidic chip-based assays, and surface plasmon resonance.

[0190] In some embodiments, one or more polypeptides are detected (and optionally, relative level determined) using a proximity extension assay (PEA). In embodiments, PEA comprises the simultaneous binding of a pair of proximity probes to a biomarker in proximity. Upon binding of the pair of proximity probes to the biomarker, the nucleic acid domains are capable of interacting and forming a nucleic acid duplex, which may enable at least one of the nucleic acid domains to be extended from its 3’ end. This extension product forms a detectable nucleic acid detection product, optionally following amplification, e.g., by PCR. Exemplary PEA methods are described in greater detail in WO 2012 / 104261 and US2015 / 0044674, which are incorporated herein by reference. Target polypeptides may be detected singly, or more preferably multiple target polypeptides may be detected simultaneously in a multiplexed detection format.

[0191] In some embodiments, one or more polypeptides are detected (and optionally, relative level determined) using a Multiple Reaction Monitoring (MRM) assay. A variety of MRMmethods are available. In embodiments, the MRM assay uses triple quadrupole mass spectrometers coupled to liquid chromatography to detect or quantify target polypeptides. In the first quadrupole (QI), a peptide that corresponds to a protein of interest is selected. The peptide is then fragmented in the second quadrupole (Q2) and a filter is applied to allow a particular fragment to enter into the third quadrupole (Q3) where its intensity is measured. Target polypeptides may be detected singly, or more preferably multiple target polypeptides may be detected simultaneously in a multiplexed detection format. Further non-limiting examples of MRM are described in US20190277846 and US20180024108, which are incorporated herein by reference.

[0192] In some embodiments, one or more polypeptides are detected (and optionally, relative level determined) using a quantitation platform integrating nanoparticle (NP) protein coronas with liquid chromatography-mass spectrometry. In embodiments, the platform is a Proteograph platform. In embodiments, a protein corona is a protein layer adsorbed onto NPs upon contact with biological fluids. Varying the physicochemical properties of engineered NPs translates to distinct protein corona patterns enabling differential and reproducible interrogation of biological samples. In embodiments, the Proteograph platform uses a multi-NP protein corona approach and mass spectrometry. In embodiments, this approach includes four steps: (1) NP-biological sample incubation and protein corona formation; (2) NP protein corona purification by a magnet; (3) digestion of corona proteins; and (4) LC-MS / MS analysis. In this context, each biological sample- NP well is a sample, for a total of 96 samples per plate. Target polypeptides may be detected singly, or more preferably multiple target polypeptides may be detected simultaneously in a multiplexed detection format. A non-limiting example of an NP -based protein corona detection is described in W02020096631 A2, which is incorporated herein by reference.

[0193] In some embodiments, one or more polypeptides are detected (and optionally, relative level determined) using an aptamer-based detection assay. In general, an “aptamer” refers to a nucleic acid that has a specific binding affinity for a target molecule. In this context, the “specific binding affinity” of an aptamer for its target means that the aptamer binds to its target generally with a much higher degree of affinity than it binds to other components in a test sample. Methods for the production of aptamer are known in the art, and include, e.g., the SELEX process (see, e.g., U.S. Patent No. 5,475,096). Aptamer assays that permit an aptamer to capture its target in solution and then employ separation steps that are designed to remove specific components of the aptamertarget mixture prior to detection have also been described (see, e.g., US2009 / 0042206). Exemplary solution-based aptamer assays that can be used to detect (and optionally quantify) a protein in a biological sample include those described in US20210215711A1.

[0194] In some embodiments, the one or more polypeptides are polypeptides identifying one or more particular proteins. A protein may be identified by any of a variety of features identifiablyassociated with that protein. For example, a particular protein may be identifiable by a particular epitope, or by a particular sequence (or subsequence) of amino acids that distinguish the protein from other proteins. Accordingly, a polypeptide that identifies a particular protein may be a portion of that protein, which portion sufficiently identifies the particular protein as the source of the portion. In some embodiments, the plurality of different polypeptides comprise polypeptides identifying proteins (e.g., 2, 5, 10, 20, 30, 40, or more proteins) selected from List 1 of Table 1 herein. In some embodiments, the plurality of different polypeptides comprise polypeptides identifying each of the proteins from List 1 of Table 1 herein. In some embodiments, the plurality of different polypeptides comprise polypeptides identifying proteins (e.g., 2, 5, 10, 15, or more proteins) selected from any one of Lists 2-19 of Table 1 herein. In some embodiments, the plurality of different polypeptides comprise polypeptides identifying each of the proteins from any one of Lists 2-19 of Table 1 herein. In some embodiments, the plurality of different polypeptides comprise polypeptides identifying proteins (e.g., 2, 5, 10, 15, or more proteins) selected from List 20 of Table 1 herein. In some embodiments, the plurality of different polypeptides comprise polypeptides identifying each of the proteins from List 20 of Table 1 herein. In some embodiments, the plurality of different polypeptides comprise polypeptides identifying one or more (e.g, 2, 3, 4, 5, or 6) of CHAD, KRT19, MMP12, PTN, SERPINA3, and SPP1.

[0195] In some embodiments, the plurality of different polypeptides comprise polypeptides identifying proteins selected from the following list of proteins (identified by UniProt reference number): P31483, P21964, Q9NRD8, P16860, 060635, 096017, Q9UKL0, Q8NHS0, P58546, 043854, P40225, Q99549, P08319, P25815, Q8TE57, Q04760, Q9BYF1, 014793, Q9NWQ8, Q13444, P34913, P09496, P34947, P55259, P01375, P52789, P09668, 075354, Q9BWV1, P22004, P05231, P46379, P40818, P62736, P51161, P09237, Q 15165, Q92558, 043186, P08670, P07585, Q15831, P19429, Q9UKP3, 095988, P36952, Q16619, P61978, P17676, Q96N03, Q13105, 095684, P21246, P34998, Q6UWL2, Q969D9, P35218, P55082, P17516, 015354, Q12912, P31997, Q9NRV9, Q9Y2B0, 095183, P13807, P20718, Q9H5Y7, Q8NC01, 075356, Q96A56, Q9GZM7, P27352, P12104, Q9NQX5, P12724, Q9UBU3, P35754, P41159, P09382, P40189, Q92692, Q15067, Q16620, P21583, P31431, P09417, Q8WVQ1, Q15846, Q9UKJ0, 000161, Q6WN34, Q92823, P00568, Q13043, P09525, Q05315, Q9UHL4, Q03154, P10644, 094903, P16234, Q9H773, 014917, Q9H7M9, NT -proBNP, P31949, Q9Y4X3, P01222, P21980, P21549, Q9UMF0, Q6GTS8, Q9NY25, Q9HBB8, P16112, P55285, 060664, P08263, P52888, Q969P0, 075340, Q9ULL4, P41218, P48357, Q9Y286, P51693, 095502, 075791, Q06418, Q12864, Q9Y5X1, Q13541, Q9UHD0, Q8WX77, Q8WTU2, P78380, Q99674, Q8NI22, P23526, Q8IW75, P09601, Q9BQR3, Q6PJW8, Q8IZP9, P06858, Q13158, Q9NR28, Q86VZ4, P35247, 095544, Q14956, P18827, P10145, Q53H82, Q9BUD6, Q16820, Q9Y5K6, P41236, Q13275,Q96LA6, P19022, P00797, Q8N1Q1, Q07108, Q9UK05, 095841, Q9UEW3, P02462, P07204, P01241, A6NI73, Q01973, Q16773, P09467, P42830, Q9BQB4, Q76M96, P19971, Q92520, P07711, P04792, Q99523, P20711, 060496, P07911, Q 13361 , P00750, 075326, P23141, P22748, P55058, P01130, P13598, Q8NBP7, P15090, Q76LX8, P08833, P33151, Q16270, P54760, Q96AP7, P32942, P08118, Q06141, P01589, P07858, Q9UBP4, Q86U17, P04066, Q14767, Q9NQ79, 014798, Q5VY43, P48304, P15085, Q07507, P17931, P04275, P55808, Q03167, P14555, Q9Y275, P08581, Q9H2A7, Q9UM47, P07451, P09619, P80370, Q14162, Q99988, P04080, P02144, Q13822, P08236, Q01638, Q13740, P48960, P17813, P31146, P12111, P16581, P15086, Q15828, Q9NNX6, P04054, Q9H1U4, P19021, P48745, P20062, 075023, P18065, 000584, P19961, Q12860, Q13231, P39060, P25445, P23284, 015467, Q13867, Q13332, P19957, P35590, P09093, P46531, P04746, P78324, P04085, Q9HD89, 015031, P24158, P05107, P13987, 075594, Q12884, P05121, P00533, P13686, P02452, P20160, P42574, P10451, Q16769, Q14393, P42785, Q8TDL5, Q16663, Q8N423, P10586, Q9Y4L1, P15907, Q8NHL6,P43121, P00740, P12830, P15529, P13591, P12318, Q9UBR2, P18428, Q12794, P07478,P07359, P98160, P08887, P59665, P24821, P16109, Q14515, Q86VB7, 095998, P20023,Q9NZK5, Q13508, Q15485, P80188, P30530, Q99650, Q15113, 014786, Q96KN2, Q6EMK4,P19320, P00441, 075015, P07339, Q16853, P15144, P08174, 000533, P02786, P05556, P10646, Q9BXJ1, Q9NPY3, P10721, P14543, 095445, Q96H15, P08571, Q99969, A1L4H1, Q07654, P35443, P55774, Q9Y5C1, Q16627, P08709, P41222, P06681, P24592, Q15582, P36222, Q06033, Q9UGM5, P49747, Q92820, P00915, P13501, P05451, Q12805, P03950, P27487, P04070, P05362, P01034, P17936, P01033, P14902, Q14160, P12829, Q9BY49, Q9NZN3, Q96C92, Q5SW79, 075506, Q15477, P04141, P21817, A6BM72, 000291, Q8IZC4, 060701, 014958, E2RYF7, Q9NVZ3, P23634, Q9Y4C8, Q9Y623, P54709, Q07973, P48507, P06753, Q04695, P25391, Q15059, 000567, Q9NZJ5, P35228, Q13503, P08913, P33121, Q9BY32, P30049, P10109, P55011, Q01780, Q6UWF7, Q9Y3B8, A6NCE7, Q08499, P46783, Q96DA2, P49755, Q96HD9, B6SEH8, 043734, 095180, Q9H2M3, P06729, Q96IW2, P55769, Q9Y2W1, 095858, Q9H347, P78524, Q14353, Q15370, P20929, Q9BW61, Q5TA50, 015305, P05026, Q86UW2, P38935, Q14088, Q9Y2Y0, Q8WZ42, P12270, 075521, P05976, P14415, Q9UFP1, Q9BZC7, Q6NZY4, Q9NYX4, P16066, Q99707, Q8N8E3, P37058, Q92935, P21673, 043290, Q96K76, Q13296, Q6P4F2, P05000, P57078, Q9UKX7, Q02127, Q6ZN66, Q9BV94, Q07075, P23511, Q96LB8, Q8NET8, Q9NV35, Q16774, Q16836, P54296, Q9BZL6, Q10587, A6NHS7, Q15018, 000425, Q9UNN8, Q14807, P35606, P20382, Q96PU4, P00966, P48668, 000327, 095670, P50461, Q3SXY8, Q03013, 043896, P59901, Q01484, P19838, P22033, Q12986, Q01581, 094766, Q14781, Q96A35, Q58F21, Q8NFP7, P46926, Q9UBV2, Q5JTV8, Q8ND90, P32241, P35609, 075427, Q93052, Q86VR7, P41227, Q5W0V3, Q86VP3, Q99598, Q13563,075534, A6NDB9, Q5VVQ6, Q96EU7, P55010, Q9Y2L6, P13224, P0C7L1, 015018, P10082, Q7Z7H5, Q16206, P29536, Q14324, Q96ID5, P13929, P20645, P23327, Q9H173, Q9BTK6, P01225, Q8TER0, Q0VD83, 095980, Q13316, P50053, Q14457, Q99942, I3L3R5, Q99807, Q53T59, Q8N668, P55809, 075348, Pl 1532, Q9Y5X3, P05305, Q8WZ75, Q8IVF2, P35914, Q14643, Q9BQI0, P36776, Q9H7C9, 014841, Q8WXC3, 075061, Q8NC42, Q8TAE8, Q5GAN6, P35520, P30084, Q8WUF8, 043423, Q13137, 094979, Q16621, Q9H3K6, P07098, P21754, P07492, P20042, 060476, Q9NYZ4, Q09666, Q92835, P43487, Q6ZRY4, P07355, Q6YN16, Q9UJ70, 095825, Q24JP5, P02458, P09543, P50914, Q7L266, P01189, Q8NFL0, Q96DR5, Q9HB40, Q8IWT1, Q5FWE3, Q6UY14, Q9BV79, Q6UWR7, P07942, Q9NR61, P09681, P58107, Q12841, P0DPI2, P08590, Q86X76, Q96DC8, Q8TCD5, Q7Z7M9, Q00872, Q14914, Q9UBQ7, Q9BVM4, Q12982, P33681, Q6UW49, P51511, Q9BW04, 014933, Q8WWV6, P23919, 075711, Q6UXI7, P29692, P02008, Q9NQR4, Q9BQS7, Q9Y2E5, Q9H3S4, 043405, Q96C24, 060234, Q7Z304, P78539, Q9P2J2, Q8N4F0, P53674, P16035, Q8N436, Q13442, P14854, P23467, Q13428, 075223, 075154, Q6NUS6, Q96EM0, Q96FZ7, Q969H8, P98161, Q9BXD5, P54687, Q9BXN1, P51688, 014960, P23471, P32320, P08138, Q6PI73, Q8NDI1, P08582, P52209, 043681, P15502, Q969X0, Q96MK3, Q8IZF2, Q96AG4, Q7Z7K0, P07093, P62072, P61026, P45954, Q6ZMM2, P05413, Q15388, Q9UBR1, P49593, 000194, P13667, P23560, P30046, Q86TH1, P02730, P13796, Q9Y303, Q6H9L7, P07288, P16410, P40199, Q8N6C8, Q02817, P98095, P02461, Q6UWP8, Q6UVK1, P39059, Q9BYJ0, Q9HCU0, Q96CG8, Q96NZ9, P47972, P02818, Q8N114, Q6IBS0, P30405, P32971, Q9Y2Y8, P35579, P13727, P08575, 043280, Q9NRR1, 075339, Q9H2X3, Q9Y646, P10645, Q04721, 095965, Q9Y251, Q8TDY8, Q15063, P08217, Q9UQP3, P17900, P37837, Q8WWQ8, P55000, P12277, Q13510, Pl 1279, P07602, P17174, P61916, P19878, P40933, Pl 1274, P52564, Q9UN19, P24394, Q6ZUJ8, P01730, Q13241, P35613, P50452, 043915, 000253, P10147, Q92609, Q9GZT9, Q9Y266, Q14242, Q12918, Q3KPI0, Q9NRM6, Q01344, P02745, Q9HBG7, 094992, Q08174, 060449, 015455, P22304, P43234, P14210, Q12866, P51671, P42701, P09874, Q5R372, Q13459, 095760, P14784, Q8NHJ6, P01584, P60568, 076038, 095715, Q8N6P7, P22301, Q9UPV0, P28838, 060934, P57771, Q03426, 014904, Q9Y478, P20809, P05412, 043707, Q96PD4, P05112, P35225, Q96AX2, Q9NYY1, Q96P31, Q9NP70, Q13007, Q9HCU5, Q8WV07, Q9Y2J8, Q9Y3P8, Q8IU57, P30838, 014867, P19801, Q16552, Q7Z739, 060575, P26951, Q8TAD2, Q9P0M4, Q7Z6M3, Q8TCS8, Q5T4W7, Q99748, P48061, Q04759, Q12933, P42768, 095379, Q13219, Q13574, P63241, 043736, 060542, P13693, P09038, Q9Y5A7, Q6UXK5, P01375, Q13651, Q96RJ3, P27540, Q969V3, Q9UHF4, Q06520, Q6UB28, Q0Z7S8, 060880, Q12968, P78362, P01903, P78410, 043521-2, P01583, P01579, Q05084, Q7L8A9, P05113, 043597, Q13261, P12034, Q92844, 095644, P09919, Q9BXJ7, Q13291, P51617,Q12778, Q14435, P30048, P32456, P01591, P55957, Q12765, Q6ZMH5, Q8N8S7, Q9Y6K9, P18564, P58294, Q9HB29, P05231, P12872, Q96DB9, Q96LC7, 075475, P19474, B1AKI9, P13232, P13747, Q9UNK0, P33241, Q8WTT0, P13725, Q8IVG5, Q8TD46, Q9UHC6, P50995, Q6DN72, P23582, Q8NDB2, Q01151, P45984, Q9NRJ3, Q9NZN5, Q9HD26, P28827, P29965, P16455, Q9BT73, Q8N608, P28845, Q9UNE0, P20849, Q9HCM2, P01588, P23229, P80098, 076036, P01374, P42575, P24071, Q9NWZ3, Q6UXB4, P37235, Q9Y258, Q9UKX5, Q9H0P0, P08727, P20340, Q9UIB8, P78310, P32970, Q29983_Q29980, 014788, Q9UDT6, Q9C035, P26022, Q07065, P80162, P20783, Q14773, Q16698, P50591, Q8WXI8, 094856, P49771, Q14005, Q15517, 015169, Q9NQ25, Q9UMR7, 043561, P10145, Q96SB3, P41217, P14317, Q9BZW8, Q16719, 000273, Q13478, 075077, Q9UQV4, P24001, P36959, P30203, P20273, Q6UXB2, P68106, P12544, 095971, P43489, P01137, Q15661, Q04637, P48023, P40259, Q03431, Q9Y6Q6, Q96LA5, Q9BXN2, Q9H4D0, P29460, P42702, Q99616, P00813, P30044, 060884, P15692, 043508, 015444, P10144, P01135, P78556, P03956, P49763, Q9BY76, 095750, 014836, P46109, Q03405, 043598, Q9HCB6, Q9NQ30, Q16651, 000468, P29350, Q07325, Q9UII2, Q9H008, P19876, Q6UWV6, P09341, Q9H3U7, Q92583, Q99538, 000182, Q03403, P53634, Q5ZPR3, P55773, P25116, Q9NZC2, P34896, Q15389, 000626, 075888, P47712, Q15166, Q14118, Q9BZZ2, Q8NFT8, Q99685, 000585, P19256, P09326, Q5KU26, Q6GTX8, P09603, P51888, P16422, P01133, P02778, Q92484, Q7KYR7, 043291, Q9Y6N7, Q8WU39, P35625, 043639, 076096, 000339, 075462, Q9UJU6, Q15109, Q08334, Q9HC38, P21709, P27930, Q9UJA9, Q9NZV1, P15260, P21860, P09238, Q9NR12, Pl 1684, P18510, Q14210, Q14116, P13236, Q99435, P36941, P30613, P55145, 000241, P29279, P0DMV8, 000300, Q8WXD2, 014773, Q96KG7, Q4KMG0, 095866, P56470, 075563, P01127, Q96PL1, 095633, Q9UKU9, Q9BYZ8, P24387, Q99983, Q13232, Q9Y3D6, P19883, P15291, P12532, Q9UHX3, Q9NQ76, Q6UXH1, Q99895, P07148, Q16363, Q92956, P22466, Q8TEU8, Q8IYS5, 000175, P25942, P54317, Q9BU40, Q8N907, Q04609, P62834, Pl 0070, P61328, 000206, P46013, P24864, Q9H832, P85299, P49715, Q9Y4C1, Q9NP95, P06401, Q6UXL0, Q9NP85, P15531, Q6UXM1, Q05329, P09693, P24928, Q09472, Q9HBE5, P43378, Q92185, Q9BY41, P10767, Q01201, P41273, P03372, Q9UPW0, P25490, Q6R327, Q13190, Q6UXZ4, Q9NQI0, Q8WX93, Q16665, P15927, Q99665, 075365, P24530, Q04837, 000401, Q9NZS2, Q9NWV8, Q7Z698, Q9Y2I7, Q9NRR2, Q96EB6, Q12888, Q9NY59, Q8WVV4, Q13114-2, Q6EBC2, Q14511, Q8NHP1, 095429, P36897, Q8IX19, Q6B9Z1, Q9Y3D3, Q8IYW5, Q96MM7, Q9UHN6, Q5TBC7, P09923, Q96D71, Q92574, P98170, P36551, 000148, 043583, P48730, Q9NS62, Q8NI17, Q93062, Q9UIK4, Q8NBK3, P35219, 060447, Q8IV38, P54274, Q9GZN4, 075173, P17643, Q15465, Q7Z5L3, Q96EP0, Q9NQ66, Q5JS54, 043184, Q9UHA7, Q15697, P17050, Q9H0U9, P56645, P16671, P26436, Q96PX8, P29536, P14902, Q14160, Q16718,Q15223, Pl 1487, P48546, Q8WV28, 060500, Q96PL5, Q86UE4, P20701, P52630, 043320, P81534, Q9Y2X7, Q15399, Q9H3T2, P55211, Q99584, Q9UHI8, P23743, Q99062, 075688, Q5T2W1, P49757, Pl 1234, P06213, 095157, 060437, Q9H7Z7, 015400, Q02880, P06730, Q15762, Q9BV40, P49765, P32927, Q5QGZ9, Q9NZH8, Q16643, Q9H171, P09564, 060238, P24666, Q96PU5, 094916, 095835, Q8N556, P17301, Q8NEU8, Q02223, Q5JS37, Q6QNK2, Q8N8U9, P20908, Q9BX67, P22303, Q9BUH6, 076074, Q9BRK3, Q9H7Y0, P01037, P04083, Q8TEA8, Q9ULI3, P30040, Q96QR1, 075347, Q9UGN4, Q8NDA2, P06280, Q6P5S2, Q16378, P61457, Q9UI42, P27348, Q99574, P06132, Q5TDH0, Q9HCU4, P40197, P04406, Q9NS98, P00325, P55083, Q8TDQ7, Q9H939, Q9BUN1, 075190, P04090, 095393, 043399, Q96HD1, Q02952, Q4VCS5, P10912, Q9BXJ0, P30101, P0C862, P30047, Q15276, 014745, Q6FHJ7, Q58EX2, Q9HB71, P14091, 060279, P20155, A6NC86, A2VDF0, P49862, P09529, P25686, Q86SX6, P22455, Q53FA7, 096007, Q8NHV1, Q9NQ48, Q96AJ9, Q6UX06, Q96A49, P01210, Q92619, P21854, P52758, Q9Y4D1, Q9UK23, Q8IUZ5, Q9NRS6, Q5VTT5, P37840, P07311, Q04323, Q6P589, P32321, P56192, 015335, Q9P2T1, P35611, P54764, Q13976, Q8N0X7, Q6GMV3, P50502, P32119, P0DML2, P40925, P10599, Q92686, Q08830, Q13790, P07333, P00390, P0DJD7, P02748, P17927, P12821, Q6UXB8, P22897, P16233, P50552, P04180, 043493 , 000602, Q93091, P12955, Q0ZGT2, P20061, P02652, P04278, P00742, P07307, P08294, Q86YW5, P09172, P27169, P16442, 095497, Q04756, P06276, Q9BXR6, P54108, P06396, P0DN86, P26927, P04040, P34096, P07998, Q01459, P06744, P04114, P01019, P0DUB6 P0DTE7 P0DTE8, P02765, 014791, P07358, P36980, P08185, P05543, P36955, Q96PD5, Q08380, P10909, P61769, P43652, Q92496, P35542, PODOY2, P00748, Pl 1226, P00734, P02649, Q9Y5Y7, P00746, P07225, P03952, Q16610, 075882-2, P05154, 000391, P02775, Q96IY4, P20742, P06727, P09871, P02776, P05452, Q14624, P05546, P02647, Q15848, P10643, P08519, P00736, P05090, P14151, P43251, Q9NZP8, P02763, P19827, P02654, P00751, P01009, P49908, P01031, P02750, P04196, P00747, P27918, P02787, P02751, P02766, P05160, P02774, P08697, P02671, P05155, P01008, P01024, 043866, P29622, P02743, P01011, P04217, P08603, P03951, P05156, A1E959, P06748, Q9NRG1, P58417, Q9H3R2, Q6UX27, Q15043, Q9HA65, 060242, Q06323, Q92765, P16278, Q9H1C3, Q9NR71, 095994, 060609, Q9HD42, Q16762, P61244, P09211, Q6UXK2, P28325, P78423, P29466, P07196, Q9Y2W6, 094985, Q6UWL6, Q9UHV9, P50135, Q9H3S3, 014763, 015496, Q9Y6Y9, 000559, Pl 1464, P15509, P42892, Q96NZ8, P78333, P78560, 014618, P07306, P13500, P02533, Q9NRW1, Q6ZMC9, P53582, Q6ZMJ4, Q8IU54, Q8N2G4, P20936, P48047, Q9BXS1, Q4LE39, P05455, Q6P1M0, Q9NXA8, Q6PIL6, P53985, Q86VW0, P49023, P51531, P08758, P15018, P13385, P15336, 096013, Q96JP9, P45452, P10636, 014579, P09769, P53539, Q9HAW4, 095466, Q96FQ6, 000399, Q8WTV0, P22676, Q9UBB4, Q9H2W6, Q07812, Q16864, Q9BW30, P23276, Q92597,P26440, Q9UM07, Q9UJY5, P14625, Q5JZY3, Q8WUW1, QI 5126, P30519, Q86Z14, Q9H0C8, 014523, 094813, P13647, Q9NS71, Q13308, Q9UKV5, Q9NSK7, P29474, P60484, Q15633, Q9Y5P4, Q9UPY6, P01375, P29475, P51452, P41208, 060240, Q9Y6E0, P30039, Q03393, Q13426, P39905, Q9UK53, Q02643, Q9Y2V2, Q9UJ72, Q8NBI3, P50453, P52798, Q92932, Q02083, Q9H5V8, Q16740, P52943, P36269, Q9Y6D9, P63098, 095256, Q9HAN9, Q96IU4, 043155, P07741, 000220, Q96PP9, Q92917, Q8IUN9, Q96B36, Q92851, 095817, 095630, Q2TAL6, Q9UBT3, 000308, P09466, Q08629, Q6UX15, Q53H47, P49789, P05231, P18031, P29218, P14868, 095727, P01138, Q9BZM5, Q8TCT1, Q2MKA7, Q9NQ88, Q9Y680, Q86WV1, 014944, Q13451, Q9Y4K4, P16444, 076070, Q15814, P08962, 015232, P48740, P38484, Q6ISS4, Q6P1N0, Q07011, Q8WWN9, 015197, Q9NPH6, P10145, Q9P0K1, Q9H477, Q8N474, P40222, P02771, Q02790, Q96GW7, Q9NZQ7, P01258, Q14108, P56279, Q9UL46, Q9BS40, P23588, P08134, Q8NFP4, P22079, 043557, Q6XZF7, P37023, 095407, Q06830, Q8WV92, Q92752, Q99426, P12644, Q9BYC5, 043927, P15151, Q08345, P06733, Q96GP6, P14384, 014594, Q96B86, Q6P4E1, 014625, P35237, P52823, Q08708, P01178, P17405, P50225, P15311, P13611, Q8NBJ7, Q8NBS9, P55291, P22223, 014737, P57087, P21757, Q9NP79, P05060, P78325, 094907, P30533, Q9P126, Q10589, P01236, Q92854, P28906, Q14739, P01732, Q99731, Q9H446, Q9UNZ2, P21217_Q11128, Q9BRF8, P14207, P22894, Q10588, P09104, Q6ZMJ2, Q16288, Q6NW40, Q01469, 000214, Q9HCK4, P13473, P12429, Q96CD2, P28908, P48052, P25774, 094779, 075509, P04216, P04233, Q9NZD4, Q14112, P01215, Q96RD9, Q9NPH3, P16871, Q9P232, Q8N6Q3, P06734, P00749, Q9Y240, Q8TCZ2, P20774, Q96FE7, Q6UXH9, Q969Z4, Q9NQ38, P20333, P14174, Q9UBW5, P12081, P04118, P30740, P23381, Q86T13, Q99972, Q15155, Q15818, Q9UGT4, P05164, P55103, P22692, Pl 1215, P02749, P50895, P19440, P31948, 076076, P30086, Q96F46, P08254, P14778, Q04900, Q02487, 060462, 015389, P05067, Q6YHK3, P04179, P00352, 043278, 076061, Q14126, Q9Y279, Pl 1717, P15289, Q99727, Q02747, 015117, Q9HCN6, Q5T2D2, P22105, P16284, Q9Y624, P30043, P14209, P01833, Q8IWV2, P22749, P14780, P04155, P28799, P00918, Q99497, P17538, Q9UKK9, Q08431, Q9UKJ1, Q16881, Q14696, P35442, Q8TDQ0, Q13093, P08648, P63313, P20827, P00995, Q8N149, P07108, P19438, P23280, Q6UXG3, 043505, 015394, Q5SXM8, Q96Q89, Q9NQW8, P26441, Q96GG9, P48169, P49792, Q03721, Q96L15, Q8N111, Q96P66, P35221, Q13936, 095372, P04234, P25440, 095049, P08908, Q30KQ4, Q13509, Q8IVU1, Q6P5Q4, Q9NZA1, P0C7M8, P63172, Q02641, Q14894, 076039, Q7RTW8, Q5U5Z8, P07954, Q8N8R5, 015350, Q9ULW2, Q9Y3E2, Q9BZ29, P47874, Q8IWB1, P23435, Q99700, Q8WTQ1, Q9P2M7, Q14129, Q13002, Q9H492, Q6QNY1, P26378, Q86T26, Q99653, Q6QEF8, Q12774, Q8WVC0, Q9NXH3, Q13084, P30419, Q8N967, Q8IZJ0, Q15042, Q96PV0, P05549, Q0VDD7, P28222, 015382, Q687X5, Q5T871, Q9UL42, 095196, Q9C0A0, 015013,P12319, P54284, Q96I25, Q01432, Q9Y6X8, Q8WW22, P04439, Q9NWU2, Q5SW96, 015240, Q92859, Q14627, Q9Y3B9, Q96A32, Q9NPB3, Q9ULU8, Q6GQQ9, Q96HC4, Q9BV20, Q14184, P43007, Q68J44, P14649, Q96G03, Q9NY46, 075718, B6A8C7, Q9Y4C0, Q7L311, Q15735, P48431, Q6ZVN8, 043665, Q6Y7W6, Q8WXW3-4, Q9H4X1, Q16625, P78318, Q6ZVM7, Q9H6S1, P52179, Q99250, 075631, Q96J84, Q9UKV0, Q19T08, Q9ULH4, 095295, A8MVW0, Q9BZE9, P61764, P68400, Q13224, 075312, P60880, A6NGG8, P05787, Q99856, Q96PH6, P52788, Q9Y6U3, Q9Y285, Q9HCM4, P26715, Q6P9F5, Q8WXG9, 015020, P55283, P10523, 014503, 043396, Q8IY33, Q2M3V2, P48775, P61366, Q10571, Q8N2Q7, Q12809, 015212, P61266, Q86UP6, P0DKB5, Pl 1229, Q5VT99, Q16650, Q6ZUT3, Q6UWJ8, P63211, P20823, Q16401, 075121, 015234, Q86UW9, Q86WK6, 015083, Q53GD3, Q8NFZ4, P54105, Q93015, Q9UHL0, Q8N6Q1, Q16520, Q96KJ4, Q9UGI9, Q53EL9, P04062, Q7L0J3, P01270, Q8IY31, P10301, P98164, Q92834, 014967, Q9H461, 043474, Q9NY72, P43220, P41732, Q17R60, Q9UKW4, Q9UJC5, Q96BJ3, 060218, P52799, P43146, P49069, P09683, Q7Z3D4, P24386, Q96QH8, Q9UM54, P08651, Q9HCY8, 060662, Q96JB5, A6NFN3, Q9P2M1, Q9HBL6, P10746, Q2UY09, Q9NPD7, Q9BY14, Q9BZJ3, P48436, Q9NUG6, P09455, Q0P6D2, Q8WWM7, Q07617, Q9BX66, Q15398, P35556, Q8IWP9, Q9UNY4, Q9UH03, P53367, P60763, P63010, 014578, 060939, P20916, Q9BX10, Q86TM3, QI 5102, P54277, P50897, Q8TC05, 095202, Q9UQ16, Q9H6Q3, Q15714, Q8IVM0, Q86YD3, P29536, 095954, Q5T848, B2RUY7, Q14151, Q92888, Q96A00, Q15366, 060869, 075167, 095467, P22102, Q07021, Q08378, 075792, 043432, Q9BUJ2, Q96PE7, Q6UWW0, P51687, Q01826, P04808, Q8WWV3, Q9H6H4, 060469, 075592, Q9H777, Q14197, Q5JSP0, 043776, P31751, P29377, P14902, Q99460, Q9BXI9, Q5T5Y3, Q6PKH6, Q13277, P21579, Q53GL0, 060890, 015269, Q9H1P3, Q96RU2, P01350, P78352, P26718, 077932, Q92599, P13995, Q8N163, Q8IWY9, Q86VP1, Q9H251, Q14149, P43320, P51649, Q9NWM8, Q14160, Q9BUP0, 094830, Q8IWQ3, Q9NPG4, Q6UXV0, Q4ZHG4, Q13938, P47813, Q9UKY0, P54315, Q15256, 014530, Q15025, P13861, P10092, Q9ULA0, P59780, Q8WYQ3, Q9NPE2, Q6UX71, Q6XQN6, P46937, Q9BXI3, Q9P2X3, Q99704, Q6P1J6, 060235, Q03252, Q8TAT2, Q5F1R6, Q03014, Q6QNY0, Q9BPX1, P32455, Q9Y3C0, P62760, Q2L4Q9, P55210, P51460, Q5VV43, A0FGR8, Pl 1137, Q9BRJ6, Q7Z4V5, P08579, Q9BQT9, Q9UBC9, P0CG30, Q8IXS6, 075146, Q9NRY6, Q96CN9, Q9NXV2, P07320, P54252, Q8IXM2, A8MVW5, 060220, Q8TF65, P53814, Q07817, Q00722, P40313, Q8N4C8, P54819, Q9NR46, P09110, 014713, Q13145, Q9NX58, 095498, Q07954, Q9BTE6, Q6UWW8, P01229, P05937, P08069, P00519, Q96I82, Q9BS26, Q14241, Q9HAV7, Q9BSL1, Q9C0C4, 000592, Q9UK85, Q7Z5R6, P30041, P82980, P47992, Q9NSA1, Q9NTU7, Q14213_Q8NEV9, P13726, P06756, P61218, Q96NA2, P50579, Q9UBG3, Q14790, P35637, Q13490, Q9UQB8, Q6EIG7, P80075, 000292, Q9BSG5, Q99075, Q9Y5W5, P42658, Q99717,043699, Q86SJ6, P35318, P35813, Q7L5Y9, P01375, Q9Y265, P42331, P06850, Q8IUK5, Q9BSW2, 095388, Q2VWP7, Q00796, 095786, Q9UHF1, P14136, P31994, P55789, P55273, Q9Y243, P22307, P43628, P31350, P39748, 014964, Q9NRA1, Q05516, P48643, P46060, 075569, Q6UX82, Q99683, P01242, Q08AG7, Q96DU3, P43629, 043752, 060828, P35070, Q8IWL2, Q7Z7D3, P34130, Q9UKR0, Q6NXT1, P54727, Q6BAA4, Q92982, Q8NBZ7, P41586, 075787, Q15797, Q96NB1, Q07960, P50749, Q6PGN9, P06731, Q8IWL1, 014662, Q7Z6M1, Q9UQQ2, P25786, Q9H4P4, 075493, Q9NS15, A4D1B5, P49788, P21810, Q7LG56, Q9P0J1, Q9Y5V3, Q8N5S9, Q7Z434, P07332, 015116, P43490, 075380, 060907, Q01543, Q9UKS7, Q06787, P04637, Q8WUX2, Q9Y6A5, P34949, Q8WYN0, Q96PQ0, P15121, P36888, Q9Y662, Q8TE58, Q9BYE9, P05231, Q7L5N7, P55008, P40198, Q9Y223, Q9Y5L3, P05783, Q8TD06, Q9Y2Z0, Q9P0V8, P51580, 043524, 075695, 000233, Q9GZY6, Q5VIR6, Q9UJ71, Q86WD7, Q15427, P10606, P51692, P0CG37, Q9H4A9, P08473, Q9NUY8, P17948, P10747, Q16772, Q9BUE0, 000186, Q3B7J2, Q6P2H3, 000221, Q9BQ51, 094760, Q9UHD8, P30260, Q9Y639, 095831, Q6UXD5, 075054, Q9Y570, P07947, P15848, QI 1201, P55039, Q8IX05, Q12846, Q96RT1, 015357, P23515, P28907, 060911, Q7Z5A7, Pl 6870, 060760, Q96EK5, Q8N9I9, 060825, Q9UBM4, 060763, P07949, Q8N386, Q8NEZ2, P15514, P18627, Q86SF2, 000622, 075144, Q13576, 000748, P58499, P26010, Q9UKR3, P49441, 043570, P37108, P38936, Q13561, 014828, P07948, Q9NZT2, P01275, P50583, Q9Y653, Q8N129, Q49AH0, P29317, Q9Y5K8, 000451, P29459 P29460, Q99795, Q99536, Q9GZV9, Q9H156, P98073, Q9P0G3, 043715, Q9ULX7, Q86SR1, Q9C005, Q13421, Q15116, Q9UJM8, P05187, P25685, Q8WXI7, P10145, 043827, P39900, P09105, P13521, P50120, P09960, Q9HAV5, P05089, Q9H4F8, Q02742, 014558, Q14203, Q9Y336, P01303, Q9H6B4, P47929, Q6UWN8, P40121, Q16595, Q8IXJ6, P80511, Q86SJ2, P98082, Q9BXY4, P06127, P80303, Q9NS68, Q9H6S3, 015263, Q9Y5K2, Q9P1Z2, Q16653, P08397, Q7Z5L0, Q96JA1, Q16790, P09758, 060243, Q9NPH0, Q96I15, P16562, P27695, Q02246, Q9BZR6, P62166, Q10471, Q8WWY7, Q6PCB0, P51858, Q16775, Q8TDQ1, P02760, Q9H3G5, Q496F6, P35052, P56159, P35475, P32926, Q96D42, P20472, 015123, P29017, Q14508, 043895, Q9UBX1, P07237, Q6FI81, P41439, Q5JTD0, 014974, P37173, Q15303, Q92832, Q96NY8, Q96J42, Q9H8J5, P21741, Q9BYH1, Q16543, Q9NZ53, P20851, P41271, P35916, P20138, Q96SM3, P35968, Q02763, P21589, 095721, P09486, Q9UP79, P32004, 043464, Q7Z4W1, Q9HAT2, Q8NCC3, P21802, Q14512, Q9NP84, 000244, Q96PD2, P78552, P01298, P13688, P26447, 075629, P09958, P48307, P18084, P15328, 060259, Q9UJ68, P26842, P06870, P12931, 043240, 095274, 000548, P49767, P04626, QI 6674, Q9UBX7, Q92876, Q6NT46, Q7Z460, Q7Z4W2, Q9UPY8, 000337, P23771, Q5VSG8, P43630, A7E2Y1, P30304, Q9P2D8, Q14123, P12004, 060941, 043504, P19075, 060502, Q13443, Q93033, Q8NHZ8, Q92499, 075781, Q6UWK7, Q9UMS0, Q96KB5,Q8ND71, 075330, Q14677, Q9GZT3, Q00994, Q86SQ0, Q9NPI5, Q9H9E1, 043889, P48165, Q9UQE7, 015264, P06493, Q15652, P00167, Q8N130, Q6PUV4, Q9H741, P78395, Q3MIW9, Q9UJZ1, Q12836, A6NGN9, A8MVZ5, Q99259, Q8N4E4, Q8WZ55, P18848, 000534, P31689, P31371, Q8IYV9, Q13017, Q12899, Q15014, Q9H867, Q14674, P17980, Q9NS37, 043739, Q9H293, Q587J8, Q4VC05, Q9NZQ9, 075794, Q8NDC4, Q8TE77, Q13127, 094986, Q9GZP4, Q8TDX7, A8MTB9, Q9Y6I3, Q5VT06, P01100, P21781, P29353, P08700, P34910, Q6P996, Q9UN42, Q6PH85, Q9UI15, 075409, Q969P6, Q9Y3C4, Q12849, A8MYV0, P49756, Q7Z6A9, 043422, Q86TS9, P07992, Q96SD1, Q13087, Q9H0R8, Q9ULR5, Q13972, Q9NQP4, Q06609, Q9UBU8, Q14554, P27701, Q03111, 095793, Q14055, Q9UNP9, Q00653, Q9UHJ6, P42681, Q68DV7, 060603, P01112, Q03518, 015078, Q99487, Q9UHL9, 043903, P47928, 014879, Q9H2G2, P49137, B0FP48, P43166, Q13445, Pl 1310, P49662, Q6N021, Q8TCU4, P59282, Q8WWU5, Q9UK41, A6NLU5, 075665, 015164, 095777, 043247, P04183, Q16181, 015294, Q9HCM3, Q96F10, Q2M296, 060237, P51815, 095696, P07766, Q8TEW0, P19526, P10398, P78358, Q8WUY3, P22528, 015211, P15248, 075293, Q6UY09, 000422, Q5VUJ9, Q9Y4G8, Q14204, P25092, Q32MZ4, Q7Z6I6, P50454, Q96NB3, Q9UMX5, Q9ULD2, P43357, P78317, Q9Y2J4, Q495A1, 014717, P30279, P26639, Q9UJ99, Q8IV48, P33764, Q9NYJ8, P53420, Q6UXC1, Q15796, Q92973, Q6ZVL6, Q9HC77, 043768, Q8WWF5, P51948, P49366, Q92817, P78540, Q6P995, Q9NQ84, P09914, 095997, Q7Z6P3, Q9BW66, 043663, P18754, P20702, Q99963, Q96AT9, P49454, Q6P5Z2, 075843, Q9UHY7, P21695, 015213, Q9P000, Q14641, 075460, Q8IYS2, P52848, Q96IQ7, Q6B8I1, Q9GZZ8, Q9HC57, Q16891, P56851, P63146, Q5VX71, Q9BZW2, Q92485, P49354, P05091, Q9UKM9, 095433, Q13287, Q06643, Q6PKG0, Q9Y5E8, P35249, Q8NEB7, Q9BU02, 060749, 075351, Q6UW88, P20807, Q8N5J2, Q96RE7, Q86TE4, Q8WUD1, Q9NUW8, A1KZ92, Q49A26, Q6UW15, P01266, Q9GZX6, Q8N0Z9, Q13410, P31785, Q8IY22, Q9BW85, P36873, Q14160, P29536, Q765P7, Q7Z692, 095166, P19525, Q14BN4, A6NM11, Q9Y3L3, 075528, Q9H8Y8, Q96T91, P61812, P21912, Q96AQ6, Q8TBM8, Q9HD43, P84022, P43632, P43627, Q86UX2, Q8NG06, Q9UHP3, P13051, Q9H2R5, Q9BRQ6, Q92783, P54652, Q6WCQ1, Q86SQ7, Q9H6E4, Q9ULC4, P51808, Q15599, Q05193, P33316, 060245, Q16549, P10415, 043805, Q96K21, Q6UW56, 060232, Pl 1387, Q96A25, Q9BQE9, Q9Y5Q6, Q03169, P53384, Q7Z4H3, P14902, Q16819, Q9P013, 000203, Q8IXQ3, 075940, Q68D85, Q9UH65, Q14011, P17181, Q676U5, Q96RF0, Q15172, Q86UU1, 043312, P20700, Q02750, Q99733, 043653, Q6NUJ1, P54577, Q2WEN9, P42081, Q8WXX5, Q8IV16, Q63HQ2, Q9UIM3, P21128, 075830, Q14246, Q96DE0, Q9Y5S2, 075071, Q8TF64, Q15262, Q14258, P13284, Q674X7, Q92890, Q6PL24, Q7Z569, Q8N6M0, P53990, 094988, Q17RW2, P49223, Q99447, Q96BQ1, Q9H910, P17568, Q9H2K0, Q9HC56, Q9NPJ3, Q6BCY4, P62330, Q8IWZ8, P48060, Q96R05, 015182.Samples

[0196] In various embodiments, the present disclosure involves obtaining a test sample, e.g., a biological test sample, such as a body fluid sample, from a subject for purposes of analyzing a plurality of target molecules (e.g., a plurality of polypeptides and / or cfDNA molecules) therein. Samples in accordance with embodiments of the invention can be collected in any clinically- acceptable manner. Any sample suspected of containing a plurality of target molecules can be used in conjunction with the methods of the present invention. In some embodiments, a sample can comprise a body fluid. In some embodiments, a biological sample is collected from a healthy subject. In some embodiments, a biological sample is collected from a subject who is known to have a particular disease or disorder (e.g., a particular cancer or tumor). In some embodiments, a biological sample is collected from a subject who is suspected of having a particular disease or disorder.

[0197] As used herein, the terms “body fluid” and “biological fluid” refer to a liquid material derived from a subject, e.g., a human or non-human mammal. Non-limiting examples of body fluids that are commonly used in conjunction with the present methods include mucous, blood, plasma, serum, serum derivatives, synovial fluid, lymphatic fluid, bile, phlegm, saliva, sweat, tears, sputum, amniotic fluid, menstrual fluid, vaginal fluid, semen, urine, cerebrospinal fluid (CSF), such as lumbar or ventricular CSF, gastric fluid, a liquid sample comprising one or more material(s) derived from a nasal, throat, or buccal swab, a liquid sample comprising one or more materials derived from a lavage procedure, such as a peritoneal, gastric, thoracic, or ductal lavage procedure, and the like.

[0198] In some embodiments, the biological fluid comprises fluid blood, plasma, serum, urine, saliva, pleural fluid, pericardial fluid, cerebrospinal fluid (CSF), peritoneal fluid, or any combination thereof. In some embodiments, the biological fluid comprises blood, a blood fraction, plasma, or serum. In some embodiments, the biological fluid is plasma.

[0199] In some embodiments, a sample can comprise a fine needle aspirate or biopsied tissue. In some embodiments, a sample can comprise media containing cells or biological material. In some embodiments, a sample can comprise a blood clot, for example, a blood clot that has been obtained from whole blood after the serum has been removed. In some embodiments, a sample can comprise stool. In one embodiment, a sample is drawn whole blood. In one aspect, only a portion of a whole blood sample is used, such as plasma, red blood cells, white blood cells, and platelets. In some embodiments, a sample is separated into two or more component parts in conjunction with the present methods. For example, in some embodiments, a whole blood sample is separated into plasma, red blood cell, white blood cell, and platelet components.

[0200] In some embodiments, a sample includes a plurality of polypeptides and / or nucleic acids (e.g., cfDNA) not only from the subject from which the sample was taken, but also from one or more other organisms, such as viral DNA / RNA that is present within the subject at the time of sampling.

[0201] Nucleic acid and / or polypeptides can be extracted from a sample according to any suitable methods known in the art, and the extracted nucleic acid can be utilized in conjunction with the methods described herein. In some embodiments, polypeptides are purified from a sample. In some embodiments, cell free nucleic acid (e.g., cfDNA) is extracted from a sample. In some embodiments, polypeptides are detected (and optionally quantified) without a protein extraction step. For example, polypeptides may be detected by contact detection reagents directly to a biological sample (e.g., a sample of biological fluid, such as serum or plasma).

[0202] In embodiments, the sample is a “matched” or “paired” sample. In general, the terms “matched sample” and “paired sample” refer to a pair of samples collected from the same subject, preferably at about the same time (e.g., as part of a single procedure or office visit, or on the same day). In embodiments, a pair of samples comprises two samples of biological fluid, which may be the same or different. In embodiments, a pair of samples comprises two aliquots separated from a single original sample (e.g., two aliquots of plasma from a blood sample). The terms may also be used to refer to polypeptides and / or polynucleotides derived from the matched sample, or sequencing reads thereof. In embodiments, a plurality of paired samples are analyzed. The plurality of paired samples may be from the same individual collected at different times (e.g., as in a paired sample from an early stage of cancer, and a paired sample from a later stage of cancer), from different individuals at the same or different times, or a combination of these. In embodiments, the matched samples are from different subjects. In embodiments, the matched samples in a plurality are from subjects with the same cancer type, and optionally the same cancer stage.Cancer Types

[0203] Methods in accordance with embodiments of the disclosure can be used for detecting the presence or absence of cancer. In some embodiments, the cancer stage is stage I cancer, stage II cancer, stage III cancer, or stage IV cancer. In some embodiments, the cancer stage is a stage 0 cancer (e.g., carcinoma in situ).

[0204] In some embodiments, the methods involve detecting the presence or absence of, determining the stage of, monitoring the progression of, and / or classifying a cancer selected from breast cancer, uterine cancer, cervical cancer, ovarian cancer, bladder cancer, urothelial cancer of renal pelvis, renal cancer other than urothelial, prostate cancer, anorectal cancer, anal cancer,colorectal cancer, hepatobiliary cancer arising from hepatocytes, hepatobiliary cancer arising from cells other than hepatocytes, liver / bile-duct cancer, esophageal cancer, pancreatic cancer, stomach cancer, squamous cell cancer of the upper gastrointestinal tract, upper gastrointestinal cancer other than squamous, head and neck cancer, lung cancer, lung adenocarcinoma, small cell lung cancer, squamous cell lung cancer and cancer other than adenocarcinoma or small cell lung cancer, neuroendocrine cancer, melanoma, thyroid cancer, sarcoma, plasma cell neoplasm, multiple myeloma, myeloid neoplasm, lymphoma, and leukemia. In some embodiments, the cancer is selected from anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0205] In some embodiments, the same assay is applied to detect any of a plurality of cancer conditions (e.g., cancer type, and / or cancer stages disclosed herein). For example, an assay in accordance with an embodiment can be used to detect the presence (and optionally stage) of a breast cancer in a sample from first subject, and repeated to detect the presence (and optionally stage) of a lung cancer in a sample from a second subject, based on evaluating biomarkers for each condition in both samples. In embodiments, the same assay is repeated across multiple samples to identify presence of at least 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 75, 100, or more cancer conditions. In embodiments, the same assay is repeated across multiple samples to identify presence of at least 10 cancer conditions. In embodiments, the same assay is repeated across multiple samples to identify presence of at least 20 cancer conditions. In embodiments, the same assay is repeated across multiple samples to identify presence of at least 30 cancer conditions. In embodiments, the same assay is repeated across multiple samples to identify presence of at least 50 cancer conditions.Detection and Treatment

[0206] In still another embodiment, information obtained from any method described herein (e.g., an aggregate probability score) 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 aggregate probability score exceeds a threshold, a physician can prescribe an appropriate treatment (e.g., a resection surgery, radiation therapy, chemotherapy, and / or immunotherapy). In some embodiments, information such as a likelihood or probability score can be provided as a readout to a physician or subject.

[0207] In one aspect, the method comprises selecting a subject having or being at increased risk of developing a cancer type, and administering a treatment to the subject effective to treat thecancer type, wherein the selecting comprises (a) measuring levels of first target molecules from a first sample of the subject, wherein the first target molecules comprise cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer types; (b) measuring levels of second target molecules from a second sample of the subject, wherein the second target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types; (c) applying a trained classifier to the measured levels of the first and second target molecules to assign an aggregate probability score for the cancer; wherein applying the trained classifier comprises: (i) applying a first trained model to the measured levels of the first target molecules to assign a first probability score for the cancer; (ii) applying a second trained model to the measured levels of the second target molecules to assign a second probability score for the cancer; and (iii) aggregating the first probability score and the second probability score; and (d) detecting the cancer by identifying that the aggregate probability score is above a threshold for presence of the cancer; and the treatment comprises surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

[0208] In some embodiments of any of the various methods described herein, measuring and analyzing one of the first target molecules and second target molecules is performed in parallel, or sequentially in any order. In some embodiments, analysis proceeds sequentially. In some embodiments, the first target molecules are measured and the first trained model is applied thereto before measuring levels of the second target molecules. In some embodiments, the second target molecules are measured and the second trained model is applied thereto before measuring levels of the first target molecules. In some embodiments in which measuring and analyzing is performed sequentially, the second measuring and analyzing steps, and the step of aggregating the first and second probability scores are only performed when confidence in the result of the first measuring and analyzing step is below a specified threshold. For example, results of analyzing a first type of analyte (e.g., the plurality of different polypeptides) may produce a probability score for the presence of cancer that is sufficiently low (so as to have a high confidence in the absence of a cancer), or sufficiently high (so as to have a high confidence in the presence of a cancer) that resorting to analysis of a second type of analyte (e.g., cfDNA) is unnecessary to detect the presence or absence of cancer. In such cases, analysis of the second type of analyte (e.g., cfDNA) proceeds for those samples where the probability score from the single type of analyte is between thresholds for confidently calling the presence or absence of cancer. In some embodiments, measuring both of the first and second target molecules and applying trained models thereto proceeds sequentially or in parallel regardless of the results of either.

[0209] In some embodiments, analysis proceeds sequentially, beginning with analysis of one or more (e.g., a plurality) of polypeptides, and only proceeds to analysis of cfDNA when a probability score for the presence of cancer based on the polypeptide analysis is above a threshold. For example, when the probability score for the presence of cancer based on polypeptide analysis is below the threshold, the sample may be scored as non-cancer; whereas if the probability score is at or above the threshold, analysis of cfDNA is performed to determine whether a probability score based on cfDNA (and / or an aggregate score based on analysis of both protein and cfDNA) is above another threshold. In this way, analysis of one or more polypeptides may be used as a screen to reduce the number of samples subjected to cfDNA analysis. In some embodiments, the probability scores based on polypeptide and cfDNA analysis are determined using trained models as described herein. In some embodiments, the specificity for the detection of cancer by the polypeptide-based model is lower than the specificity for the detection of cancer by the cfDNA-based (or aggregated) model. Whereas lower specificity in polypeptide-only assays would give rise to higher false- positives, pairing such lower-specificity polypeptide assay with a higher-specificity cfDNA assay allows for the second assay to reduce false-positives while also reducing the number of samples subjected to what could be more costly and resource-intensive sample processing and analysis steps. In some embodiments, polypeptide analysis uses a first trained model having a defined specificity for cancer detection that is 0.500 or higher (e.g., at least 0.600, 0.700, 0.800, 0.900, 0.950, or higher). In some embodiments, cfDNA analysis uses a second trained model having a defined specificity for cancer detection that is 0.900 or higher (e.g., at least 0.950, 0.975, 0.980, 0.985, 0.990, 0.995, or higher). In some embodiments, the second trained model aggregates results for both protein and cfDNA to assign a probability score for the presence of cancer, with cancer being detected when the probability score is above a threshold.

[0210] Models and classifiers (as described herein) can be used to determine a probability score (e.g., an aggregate probability score, or a second probability score) that a sample is from a subject that has cancer. In one embodiment, an appropriate treatment (e.g., resection surgery or therapeutic) is prescribed when the probability score exceeds a threshold. For example, in one embodiment, if the probability score is greater than or equal to 60, one or more appropriate treatments are prescribed. In another embodiments, if the aggregate probability score 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, a cancer log-odds ratio can indicate the effectiveness of a cancer treatment. For example, an increase in the cancer log-odds ratio over time (e.g., at a second time, after treatment) can indicate that the treatment was not effective. Similarly, a decrease in the cancer log-odds ratio over time (e.g., at a second time, aftertreatment) can indicate successful treatment. In another embodiment, if the cancer log-odds ratio is greater than 1, greater than 1.5, greater than 2, greater than 2.5, greater than 3, greater than 3.5, or greater than 4, one or more appropriate treatments are prescribed.

[0211] 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, anthracyclines, 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 (HDAC) 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, antiandrogens, and GnRH agonists or analogs. In 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.Cancer and Treatment Monitoring

[0212] In certain embodiments, a first time point is before a cancer treatment (e.g., before a resection surgery or a therapeutic intervention), and a second time point is after a cancer treatment (e.g., after a resection surgery or therapeutic intervention), and the method is utilized to monitor the effectiveness of the treatment. For example, if the an aggregate probability score at the second time point is decreased compared to the an aggregate probability score at the first time point, then the treatment may be considered to have been effective. However, if the second aggregate probability score increases compared to the first aggregate probability score, then the treatment may be considered to have not been effective. In other embodiments, both the first and secondtime 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., before a resection surgery or a therapeutic intervention) and the method is used to monitor the effectiveness of the treatment or loss of effectiveness of the treatment. In still other embodiments, cfDNA and polypeptide 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.

[0213] 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 aboutI, 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 30 days, or such as about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10,I I, 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 3 months, at least once every 6 months, at least once a year, at least once every 2 years, at least once every 3 years, at least once every 4 years, or at least once every 5 years.Computer Systems and Devices

[0214] In one aspect, the present disclosure provides a computer system for implementing one or more steps of a method disclosed herein. In another aspect, the present disclosure provides a non-transitory, computer-readable medium, having stored thereon computer-readable instructions for implementing one or more steps of a method disclosed herein.

[0215] Methods of the disclosure can be performed using software, hardware, firmware, hardwiring, or combinations of any of these. Features implementing functions can also be physically located at various positions, including being distributed such that portions of functions are implemented at different physical locations (e.g., imaging apparatus in one room and host workstation in another, or in separate buildings, for example, with wireless or wired connections).

[0216] Processors suitable for the execution of computer programs include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Generally, a processor will receive instructions and data from a read-only memory or a random-access memory, or both. The essential elements of a computer are aprocessor for executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks. Information carriers suitable for embodying computer program instructions and data include all forms of non-volatile memory, including, by way of example, semiconductor memory devices, (e.g., EPROM, EEPROM, solid state drive (SSD), and flash memory devices); magnetic disks, (e.g., internal hard disks or removable disks); magnetooptical disks; and optical disks (e.g., CD and DVD disks). The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0217] To provide for interaction with a user, the subject matter described herein can be implemented on a computer having an VO device, e.g., a CRT, LCD, LED, or projection device for displaying information to the user and an input or output device such as a keyboard and a pointing device, (e.g., a mouse or a trackball), by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well. For example, feedback provided to the user can be any form of sensory feedback, (e.g., visual feedback, auditory feedback, or tactile feedback), and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0218] The subject matter described herein can be implemented in a computing system that includes a back-end component (e.g., a data server), a middleware component (e.g., an application server), or a front-end component (e.g., a client computer having a graphical user interface or a web browser through which a user can interact with an implementation of the subject matter described herein), or any combination of such back-end, middleware, and front-end components. The components of the system can be interconnected through a network by any form or medium of digital data communication, e.g., a communication network. For example, a reference set of data may be stored at a remote location and a computer can communicate across a network to access the reference data set for comparison purposes. In other embodiments, however, a reference data set can be stored locally within the computer, and the computer accesses the reference data set within the CPU for comparison purposes. Examples of communication networks include, but are not limited to, cell networks (e.g., 3G or 4G), a local area network (LAN), and a wide area network (WAN), e.g., the Internet.

[0219] The subject matter described herein can be implemented as one or more computer program products, such as one or more computer programs tangibly embodied in an information carrier (e.g., in a non-transitory computer-readable medium) for execution by, or to control the operation of, a data processing apparatus (e.g., a programmable processor, a computer, or multiple computers). A computer program (also known as a program, software, software application, app,macro, or code) can be written in any form of programming language, including compiled or interpreted languages (e.g., C, C++, Perl), and it can be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. Systems and methods of the disclosure can include instructions written in any suitable programming language known in the art, including, without limitation, C, C++, Perl, Java, ActiveX, HTML5, Visual Basic, or JavaScript.

[0220] A computer program does not necessarily correspond to a file. A program can be stored in a file or a portion of a file that holds other programs or data, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code). A computer program can be deployed to be executed on one computer or on multiple computers at one site or distributed across multiple sites and interconnected by a communication network.

[0221] A file can be a digital file, for example, stored on a hard drive, SSD, CD, or other tangible, non-transitory medium. A file can be sent from one device to another over a network (e.g., as packets being sent from a server to a client, for example, through a Network Interface Card, modem, wireless card, or similar).

[0222] Writing a file according to the disclosure involves transforming a tangible, non-transitory computer-readable medium, for example, by adding, removing, or rearranging particles (e.g., with a net charge or dipole moment into patterns of magnetization by read / write heads), the patterns then representing new collocations of information about objective physical phenomena desired by, and useful to, the user. In some embodiments, writing involves a physical transformation of material in tangible, non-transitory computer readable media (e.g., with certain optical properties so that optical read / write devices can then read the new and useful collocation of information, e.g., burning a CD-ROM). In some embodiments, writing a file includes transforming a physical flash memory apparatus such as NAND flash memory device and storing information by transforming physical elements in an array of memory cells made from floating-gate transistors. Methods of writing a file are well-known in the art and, for example, can be invoked manually or automatically by a program or by a save command from software or a write command from a programming language.

[0223] Suitable computing devices typically include mass memory, at least one graphical user interface, at least one display device, and typically include communication between devices. The mass memory illustrates a type of computer-readable media, namely computer storage media. Computer storage media may include volatile, nonvolatile, removable, and non-removable media implemented in any method or technology for storage of information, such as computer readable instructions, data structures, program modules, or other data. Examples of computer storage mediainclude RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, Radiofrequency Identification (RFID) tags or chips, or any other medium that can be used to store the desired information, and which can be accessed by a computing device.

[0224] Functions described herein can be implemented using software, hardware, firmware, hardwiring, or combinations of any of these. Any of the software can be physically located at various positions, including being distributed such that portions of the functions are implemented at different physical locations.

[0225] As one skilled in the art would recognize as necessary or best suited for performance of the methods of the disclosure, a computer system for implementing some or all of the described inventive methods can include one or more processors (e.g., a central processing unit (CPU) a graphics processing unit (GPU), or both), main memory and static memory, which communicate with each other via a bus.

[0226] A processor will generally include a chip, such as a single core or multi-core chip, to provide a central processing unit (CPU). A process may be provided by a chip from Intel or AMD.

[0227] Memory can include one or more machine-readable devices on which is stored one or more sets of instructions (e.g., software) which, when executed by the processor(s) of any one of the disclosed computers can accomplish some or all of the methodologies or functions described herein. The software may also reside, completely or at least partially, within the main memory and / or within the processor during execution thereof by the computer system. Preferably, each computer includes a non-transitory memory such as a solid-state drive, flash drive, disk drive, hard drive, etc.

[0228] While the machine-readable devices can in an exemplary embodiment be a single medium, the term “machine-readable device” should be taken to include a single medium or multiple media (e.g., a centralized or distributed database, and / or associated caches and servers) that store the one or more sets of instructions and / or data. These terms shall also be taken to include any medium or media that are capable of storing, encoding, or holding a set of instructions for execution by the machine and that cause the machine to perform any one or more of the methodologies of the present disclosure. These terms shall accordingly be taken to include, but not be limited to, one or more solid-state memories (e.g., subscriber identity module (SIM) card, secure digital card (SD card), micro SD card, or solid-state drive (SSD)), optical and magnetic media, and / or any other tangible storage medium or media.

[0229] A computer of the disclosure will generally include one or more I / O device such as, for example, one or more of a video display unit (e.g., a liquid crystal display (LCD) or a cathode raytube (CRT)), an alphanumeric input device (e.g., a keyboard), a cursor control device (e.g., a mouse), a disk drive unit, a signal generation device (e.g., a speaker), a touchscreen, an accelerometer, a microphone, a cellular radio frequency antenna, and a network interface device, which can be, for example, a network interface card (NIC), Wi-Fi card, or cellular modem.

[0230] Any of the software can be physically located at various positions, including being distributed such that portions of the functions are implemented at different physical locations.

[0231] Additionally, systems of the disclosure can be provided to include reference data. Any suitable genomic data may be stored for use within the system. Examples include, but are not limited to: comprehensive, multi-dimensional maps of the key genomic changes in major types and subtypes of cancer from The Cancer Genome Atlas (TCGA); a catalog of genomic abnormalities from The International Cancer Genome Consortium (ICGC); a catalog of somatic mutations in cancer from COSMIC; the latest builds of the human genome and other popular model organisms; up-to-date reference SNPs from dbSNP; gold standard indels from the 1000 Genomes Project and the Broad Institute; exome capture kit annotations from Illumina, Agilent, Nimblegen, and Ion Torrent; transcript annotations; small test data for experimenting with pipelines (e.g., for new users).

[0232] In some embodiments, data is made available within the context of a database included in a system. Any suitable database structure may be used including relational databases, object- oriented databases, and others. In some embodiments, reference data is stored in a relational database such as a “not-only SQL” (NoSQL) database. In various embodiments, a graph database is included within systems of the disclosure. It is also to be understood that the term “database” as used herein is not limited to one single database; rather, multiple databases can be included in a system. For example, a database can include two, three, four, five, six, seven, eight, nine, ten, fifteen, twenty, or more individual databases, including any integer of databases therein, in accordance with embodiments of the disclosure. For example, one database can contain public reference data, a second database can contain test data from a patient, a third database can contain data from healthy subjects, and a fourth database can contain data from sick subjects with a known condition or disorder. It is to be understood that any other configuration of databases with respect to the data contained therein is also contemplated by the methods described herein.ILLUSTRATIVE EMBODIMENTS

[0233] The present disclosure provides the following illustrative embodiments.

[0234] Embodiment 1 : A method of detecting cancer in a subject, the method comprising:(a) measuring levels of first target molecules from a first sample of the subject, wherein the first target molecules comprise cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer types;(b) measuring levels of second target molecules from a second sample of the subject, wherein the second target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types;(c) applying a trained classifier to the measured levels of the first and second target molecules to assign an aggregate probability score for the cancer; wherein applying the trained classifier comprises: (i) applying a first trained model to the measured levels of the first target molecules to assign a first probability score for the cancer; (ii) applying a second trained model to the measured levels of the second target molecules to assign a second probability score for the cancer; and (iii) aggregating the first probability score and the second probability score; and(d) detecting the cancer by identifying that the aggregate probability score is above a threshold for presence of the cancer.

[0235] Embodiment 2: The method of embodiment 1, wherein the trained classifier was trained using reference first probability scores from the first trained model, reference second probability scores from the second trained model, and reference aggregated probability scores aggregating the reference first probability scores and reference second probability scores, for reference samples from (1) reference subjects having known cancers, and (2) reference subjects without cancer.

[0236] Embodiment 3 : The method of embodiment 1 or 2, wherein the trained classifier assigns an aggregate probability score for each of a plurality of different cancer types, and detecting the cancer comprises identifying the cancer type as the cancer type with the highest aggregate probability score.

[0237] Embodiment 4: The method of any one of embodiments 1-3, wherein aggregating the first cancer probability score and the second cancer probability score comprises combining the first and second probability scores for the cancer in a linear model.

[0238] Embodiment 5: The method of any one of embodiments 1-4, wherein the first sample and the second sample are the same.

[0239] Embodiment 6: The method of any one of embodiments 1-5, wherein the plurality of different target genomic regions comprises at least 1000, 5000, 10000, 20000, or 30000 target genomic regions.

[0240] Embodiment 7: The method of any one of embodiments 1-6, wherein the plurality of target genomic regions comprises a total collective length of at least 50 kb, 100 kb, 500 kb, or 1000 kb.

[0241] Embodiment 8: The method of any one of embodiments 1-7, wherein each of the plurality of different target genomic regions comprises at least five methylation sites.

[0242] Embodiment 9: The method of any one of embodiments 1-8, wherein measuring the first target molecules comprises sequencing converted cfDNA from the plurality of different target genomic regions, or amplification products thereof, wherein the converted cfDNA comprises cfDNA treated with a deaminating agent.

[0243] Embodiment 10: The method of embodiment 9, further comprising treating the cfDNA with the deaminating agent, optionally wherein the deaminating agent is a cytosine deaminase or bisulfite.

[0244] Embodiment 11 : The method of embodiment 9 or 10, wherein the sequencing produces at least 100,000 sequencing reads.

[0245] Embodiment 12: The method of any one of embodiments 8-11, wherein measuring the first target molecules comprises enriching for the converted cfDNA or amplification products thereof, to produce an enriched sample of polynucleotides.

[0246] Embodiment 13: The method of embodiment 12, wherein the enriching comprises capturing the converted cfDNA or amplification products thereof with a plurality of corresponding bait oligonucleotides.

[0247] Embodiment 14: The method of embodiment 13, wherein the plurality of different target genomic regions for enrichment by the bait oligonucleotides are genomic regions identified by the first trained model as differentially methylated in the at least one of a plurality of cancer types relative to non-cancer tissue or relative to cancer of a different type.

[0248] Embodiment 15: The method of any one of embodiments 1-14, wherein the plurality of different polypeptides comprise at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides.

[0249] Embodiment 16: The method of embodiment 15, wherein the plurality of different polypeptides comprise: (a) polypeptides identifying proteins selected from List 1; (b) polypeptides identifying proteins selected from any one of Lists 2-19; (c) polypeptides identifying proteins selected from List 20; or (d) polypeptides identifying one or more of CHAD, KRT19, MMP12, PTN, SERPINA3, and SPP1.

[0250] Embodiment 17: The method of any one of embodiments 1-16, wherein the trained classifier discriminates a subject with cancer from a subject without cancer with a defined specificity for each of the plurality of cancer types.

[0251] Embodiment 18: The method of embodiment any one of embodiments 1-17, wherein the trained classifier has a higher sensitivity for cancer detection than each of the first trained model and the second trained model; optionally wherein the trained classifier has a specificity for cancerdetection that is equal to or greater than each of the first trained model and the second trained model.

[0252] Embodiment 19: The method of any one of embodiments 1-18, wherein the trained classifier is a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier.

[0253] Embodiment 20: The method of any one of embodiments 1-19, wherein the first trained model and / or the second trained model is a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier.

[0254] Embodiment 21 : The method of any one of embodiments 1-19, wherein the first trained model binarizes measured levels of the first target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected.

[0255] Embodiment 22: The method of any one of embodiments 1-19, wherein the second trained model log-transforms measured levels of the second target molecules normalized against control protein present in a known amount.

[0256] Embodiment 23: The method of any one of embodiments 1-22, wherein (a) the first trained model is trained using measured levels of the first target molecules for first reference samples, (b) the second trained model is trained using measured levels of the second target molecules for second reference samples, and (c) the first and second reference samples comprise samples from reference subjects having known cancers, and reference subjects without cancer.

[0257] Embodiment 24: The method of any one of embodiments 1-23, wherein the first sample and / or second sample comprises a biological fluid; optionally where the biological fluid comprises blood, plasma, serum, urine, saliva, pleural fluid, pericardial fluid, cerebrospinal fluid (CSF), peritoneal fluid, or any combination thereof.

[0258] Embodiment 25: The method of embodiment 24, wherein the biological fluid comprises blood, a blood fraction, plasma, or serum.

[0259] Embodiment 26: The method of embodiment 25, wherein the first sample and / or second sample is a plasma sample.

[0260] Embodiment 27: The method of any one of embodiments 1-26, wherein the plurality of cancer types comprise at least 10 cancer types.

[0261] Embodiment 28: The method of any one of embodiments 1-27, wherein the plurality of cancer types comprises one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.-n-

[0262] Embodiment 29: The method of any one of embodiments 1-28, further comprising treating the subject for the cancer type.

[0263] Embodiment 30: The method of embodiment 29, wherein the treating comprises surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

[0264] Embodiment 31 : A method of treating cancer in a subject, the method comprising selecting a subject based on the results of a detection assay, and treating the subject for the cancer, wherein:(a) the detection assay comprises the method of any one of embodiments 1-28; and(b) the treating comprises surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

[0265] Embodiment 32: A method of training a classifier for detecting target molecules from a cancer, the method comprising:(a) receiving first measured levels of first target molecules for first samples of reference subjects, wherein (i) the first target molecules comprise cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer types, and (ii) the reference subjects comprise first subjects having known cancer types, and second subjects without cancer;(b) training a first model to generate a first probability score for the presence of cancer in a subject by applying a first machine learning algorithm to the first measured levels;(c) receiving second measured levels of second target molecules for second samples of the reference subjects, wherein the second target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types;(d) training a second model to generate a second probability score for the presence of cancer in a subject by applying a second machine learning algorithm to the second measured levels;(e) generating reference first cancer probability scores for the first samples using the trained first model;(f) generating reference second cancer probability scores for the second samples using the trained second model;(g) generating reference aggregated cancer probability scores for a plurality of the reference subjects by aggregating the reference first cancer probability score and reference second cancer probability score for each respective reference subject; and(h) training a classifier to generate an aggregate cancer probability score for a subject by applying a third machine learning algorithm to the reference first cancer probability scores, reference second cancer probability scores, and reference aggregated cancer probability scores.

[0266] Embodiment 33: The method of embodiment 32, wherein aggregating the first cancer probability score and the second cancer probability score comprises combining the first and second probability scores for the cancer in a linear model.

[0267] Embodiment 34: The method of embodiment 32 or 33, wherein the first machine learning algorithm, second machine learning algorithm, and / or third machine learning algorithm is an Ll- regularized logistic regression, an L2-regularized logistic regression, a generalized linear model (GLM), a random forest, a multinomial logistic regression, a multilayer perceptron, a support vector machine, or a neural network.

[0268] Embodiment 35: The method of any one of embodiments 32-34, wherein the first trained model binarizes measured levels of the first target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected.

[0269] Embodiment 36: The method of any one of embodiments 32-35, wherein the second trained model log-transforms measured levels of the second target molecules normalized against control protein present in a known amount.

[0270] Embodiment 37: The method of any one of embodiments 32-36, wherein (a) the first trained model is trained using measured levels of the first target molecules for first reference samples, (b) the second trained model is trained using measured levels of the second target molecules for second reference samples, and (c) the first and second reference samples comprise samples from reference subjects having known cancers, and reference subjects without cancer.

[0271] Embodiment 38: The method of any one of embodiments 32-37, wherein the third machine learning algorithm is a logistic regression.

[0272] Embodiment 39: The method of any one of embodiments 32-38, wherein the plurality of different target genomic regions comprises at least 1000, 5000, 10000, 20000, or 30000 target genomic regions.

[0273] Embodiment 40: The method of any one of embodiments 32-39, wherein the plurality of target genomic regions comprises a total collective length of at least 50 kb, 100 kb, 500 kb, or 1000 kb.

[0274] Embodiment 41 : The method of any one of embodiments 32-40, wherein each of the plurality of different target genomic regions comprises at least five methylation sites.

[0275] Embodiment 42: The method of any one of embodiments 32-41, wherein the first measured levels comprise sequencing results for the cfDNA or amplicons thereof.

[0276] Embodiment 43: The method of embodiment 42, wherein the sequencing results comprise at least 100,000 reads for each of the first samples.

[0277] Embodiment 44: The method of any one of embodiments 32-43, wherein the plurality of different polypeptides comprise at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides.

[0278] Embodiment 45: The method of embodiment 44, wherein the plurality of different polypeptides comprise: (a) polypeptides identifying proteins selected from List 1; (b) polypeptides identifying proteins selected from any one of Lists 2-19; (c) polypeptides identifying proteins selected from List 20; or (d) polypeptides identifying one or more of CHAD, KRT19, MMP12, PTN, SERPINA3, and SPPL

[0279] Embodiment 46: The method of any one of embodiments 32-45, wherein the plurality of cancer types comprise at least 10 cancer types.

[0280] Embodiment 47: The method of any one of embodiments 32-46, wherein the plurality of cancer types comprises one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0281] Embodiment 48: A method of screening for cancer in a subject, the method comprising:(a) measuring levels of first target molecules from a first sample of the subject, wherein the first target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of a plurality of cancer types;(b) applying a first trained model to the measured levels of the first target molecules to assign a first probability score for each of the plurality of cancer types; wherein (i) the first trained model has a first specificity for cancer detection, and (ii) the first probability score for at least one of the cancer types is above a first threshold for the presence of cancer;(c) measuring levels of second target molecules from a second sample of the subject, wherein the second target molecules comprise cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of the plurality of cancer types;(d) applying a second trained model to the measured levels of the second target molecules to assign a second probability score for the cancer; wherein (ii) the second trained model has a second specificity for cancer detection, and (ii) the second specificity is higher than the first specificity; and(e) detecting the cancer by identifying that the second probability score is above a threshold for the presence of cancer.

[0282] Embodiment 49: The method of embodiment 48, wherein (a) the first trained model is trained using measured levels of the first target molecules for first reference samples, (b) thesecond trained model is trained using measured levels of the second target molecules for second reference samples, and (c) the first and second reference samples comprise samples from reference subjects having known cancers, and reference subjects without cancer.

[0283] Embodiment 50: The method of embodiment 48 or 49, wherein the first sample and the second sample are the same.

[0284] Embodiment 51 : The method of any one of embodiments 48-50, wherein the plurality of different target genomic regions comprises at least 1000, 5000, 10000, 20000, or 30000 target genomic regions.

[0285] Embodiment 52: The method of any one of embodiments 48-51, wherein the plurality of target genomic regions comprises a total collective length of at least 50 kb, 100 kb, 500 kb, or 1000 kb.

[0286] Embodiment 53: The method of any one of embodiments 48-52, wherein each of the plurality of different target genomic regions comprises at least five methylation sites.

[0287] Embodiment 54: The method of any one of embodiments 48-53, wherein measuring the second target molecules comprises sequencing converted cfDNA from the plurality of different target genomic regions, or amplification products thereof, wherein the converted cfDNA comprises cfDNA treated with a deaminating agent.

[0288] Embodiment 55: The method of embodiment 54, wherein the sequencing produces at least 100,000 sequencing reads.

[0289] Embodiment 56: The method of any one of embodiments 54-55, wherein measuring the first target molecules comprises enriching for the converted cfDNA or amplification products thereof, to produce an enriched sample of polynucleotides.

[0290] Embodiment 57: The method of embodiment 56, wherein the enriching comprises capturing the converted cfDNA or amplification products thereof with a plurality of corresponding bait oligonucleotides.

[0291] Embodiment 58: The method of embodiment 57, wherein the plurality of different target genomic regions for enrichment by the bait oligonucleotides are genomic regions identified by the second trained model as differentially methylated in the at least one of a plurality of cancer types relative to non-cancer tissue or relative to cancer of a different type.

[0292] Embodiment 59: The method of any one of embodiments 48-58, wherein the plurality of different polypeptides comprise at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides.

[0293] Embodiment 60: The method of embodiment 59, wherein the plurality of different polypeptides comprise: (a) polypeptides identifying proteins selected from List 1; (b) polypeptides identifying proteins selected from any one of Lists 2-19; (c) polypeptides identifying proteinsselected from List 20; or (d) polypeptides identifying one or more of CHAD, KRT19, MMP12, PTN, SERPINA3, and SPP1.

[0294] Embodiment 61 : The method of any one of embodiments 48-60, wherein the first trained model and / or the second trained model is a neural network classifier, a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier.

[0295] Embodiment 62: The method of any one of embodiments 48-61, wherein the second trained model binarizes measured levels of the second target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected.

[0296] Embodiment 63 : The method of any one of embodiments 48-62, wherein the first trained model log-transforms measured levels of the first target molecules normalized against a control polypeptide present in a known amount.

[0297] Embodiment 64: The method of any one of embodiments 48-63, wherein (a) the first sample and / or second sample comprises a biological fluid; optionally where the biological fluid comprises blood, plasma, serum, urine, saliva, pleural fluid, pericardial fluid, cerebrospinal fluid (CSF), peritoneal fluid, or any combination thereof; (b) the first sample and / or second sample is a plasma sample.

[0298] Embodiment 65: The method of any one of embodiments 48-64, wherein (a) the plurality of cancer types comprise at least 10 cancer types; and / or (b) the plurality of cancer types comprises one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

[0299] Embodiment 66: The method of any one of embodiments 48-65, further comprising treating the subject for the cancer type; optionally wherein the treating comprises surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

[0300] Embodiment 67: A method of treating a cancer in a subject, the method comprising selecting a subject based on the results of a screening assay, and treating the subject for the cancer, wherein:(a) the screening assay comprises the method of any one of embodiments 48-65; and(b) the treating comprises surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

[0301] Embodiment 68: A non-transitory computer-readable medium with instructions stored thereon, that when executed by one or more processors, perform one or more steps in the method of any one of embodiments 1-28 or 48-65.

[0302] Embodiment 69: A non-transitory computer-readable medium with instructions stored thereon, that when executed by one or more processors, perform the method of any one of embodiments 32-47.EXAMPLES

[0303] The following examples are put forth so as to provide those of ordinary skill in the art with a complete disclosure and description of how to make and use the present description, and are not intended to limit the scope of what the inventors regard as their description, nor are they intended to represent that the experiments below are all or the only experiments performed. Efforts have been made to ensure accuracy with respect to numbers used (e.g., amounts, temperature, etc.) but some experimental errors and deviations should be accounted for.EXAMPLE 1 — Combined analyte analysis for detecting cancer

[0304] To test how the performance of combined analysis of cfDNA methylation and protein in detecting cancer, 1,538 cancer and 1,485 non-cancer Circulating Cell-free Genome Atlas (CCGA) samples were analyzed. The cancer samples were selected based on potential for mortality benefit improvement and included pre-specified high mortality cancers such as anus, bladder, colon, rectum, esophagus, head and neck, liver, bile-duct, ling, ovary, pancreas and stomach cancers, as well as additional solid tumor cancers with large number of samples available, such as breast, prostate and kidney cancer. The non-cancer samples were age-matched to the cancer cohort and were selected to power a 90% probability of observing specificity at higher than 99% in a holdout set. Samples were analyzed using a panel of about 30,000 methylation markers and about 3,000 protein markers. The proteins were selected from the OLINK EXPLORE panel of proteins (see, e.g., Eldjarn et al., Nature, 2023 Oct;622(7982):348-358).

[0305] Data on cfDNA methylation and protein levels was then analyzed using a model for cancer detection based on single analyte analysis (FIG. 2). False positives for cancer detection based on cfDNA methylation or protein levels were found to be orthogonal (FIG. 3), with distinct false positives suggestive of potential for complementary analysis of both analytes. In addition, considering both cfDNA methylation and protein levels resulted in detection of more samples as cancer samples without a loss of specificity (FIG. 3).

[0306] In view of these results, it was determined that cfDNA methylation and protein levels could be combined for detection of cancer at higher sensitivity without decreasing specificity. An integrated multiomics classifier model was developed (FIG. 4). Two constituent models were trained, a cfDNA model and a Protein model. For feature preprocessing, cfDNA methylationfeatures were binarized and Protein features were log-transformed counts normalized by control counts. The model architecture for both constituent models was the same. The processed features were used as input into an ensemble of 8 logistic regressions and the ensemble member outputs were averaged to yield a single probability of cancer. The fusion model has 3 features: the cfDNA model’s probability of cancer, the Protein model’s probability of cancer, and the product of these probabilities. This fusion model was a logistic regression and outputs a single probability of cancer which was used for binary cancer classification. The integrated multi omics classifier model utilizes the cancer score from a single-analyte model for each of cfDNA methylation and protein level as the input to a combined cancer scoring model. Analysis of the CCGA sample with the integrated multiomics classifier model increased multi-cancer detection, with higher sensitivity and specificity observed for cancer detection by combined analysis of cfDNA methylation and protein level as compared to analysis of cfDNA methylation alone (FIG. 5A). Combined analysis of cfDNA methylation and protein level also led to a higher sensitivity at 99.4% specificity when compared to analysis of cfDNA methylation alone (FIG. 5B).

[0307] In sum, analysis of both methylated cfDNA and protein levels resulted in higher sensitivity at high specificity for detecting cancer in samples, as compared to analysis of either the protein or methylated cfDNA alone. The combined analysis of methylated cfDNA and protein levels would permit for more accurate identification of true positive samples during cancer screening.EXAMPLE 2 - Biomarker Selection and Model Training

[0308] A biomarker selection and model training algorithm was generated for cancer detection as described below.

[0309] From an initial set of 2,907 biomarker proteins (OLINK EXPLORE panel), markers with small variance were removed, and all markers were screened using a fast univariate feature selection metric. An iterative greedy forward selection algorithm was used to select a specified number (K) of proteins. Briefly, the protein which best predicted cancer status within the model training data was identified, as measured by area under the curve (AUC). In each subsequent iteration, the performance gained by adding each candidate protein marker into an existing model (the previous iteration) was evaluated, again measured by AUC. The protein marker that achieved the best performance was added to the model for the next iteration. The iterative loop stopped when the number of markers selected reached the pre-specified K. Next, a logistic regression was fitted on the K selected markers, and the model was subsequently fitted to predict cancer status.

[0310] Cross-validated performance of the models was then evaluated. A training dataset was divided into 6 different splits of data. Biomarker selection was performed using a biomarker selection and model training algorithm (described below) on 5 / 6th of the splits, and the model generated from this process was used to predict cancer status on the remaining l / 6th of the training data. This process was repeated 6 times, leaving out a different l / 6th each time. This resulted in 6 different models. Two additional cross-validated runs were performed, repeating the first step two more times, and randomly shuffling the training set each time into a different mix of 6 splits each time.

[0311] The cross-validated performance of the model resulted in performance estimates for 18 different models (6 splits x 3 random shuffles), for each set of specified number of proteins (K). Table 1 includes exemplary lists of protein biomarkers for cancer detection determined based on the biomarker selection and model training algorithm, including models for a set of 50 proteins (List 1), and models for sets of 20 proteins (Lists 2-20). Six biomarkers were observed to be present in at least 75% of the 20-marker models generated by the different iterations: CHAD, KRT19, MMP12, PTN, SERPINA3, and SPP1.

[0312] Aggregate probability of cancer was determined using a linear model for combining probabilities from the cfDNA-based model and the protein-based model. The aggregate probability (y) was calculated according to the following formula:In the above formula: y = final probability of cancer output given DNA and Protein p_DNA = cancer prediction from DNA model p_prot = cancer prediction from Protein model Po = Bias term, constant / intercept Pi = Learned coefficient for DNA probability P2 = Learned coefficient for Protein probabilityP3 = Learned coefficient for DNA probability and Protein probability product

[0313] FIG. 6 shows the mean sensitivity at 99.4% specificity for cancer detection performance by analysis using the models based on cfDNA methylation markers in combination with analysis of different numbers of protein markers. As few as 5 protein markers provided substantial improvement over cfDNA methylation markers alone, which was improved further by the use of 10 protein markers, and further still by the use of 20 markers. The results alsoshow that models using as few as 20 markers (or 30, 40, or 50 markers) performed approximately as well as models based on the full initial set of 2,907 proteins. The results therefore show that combination models can be both variable and substantially simplified while retaining similar performance over cfDNA methylation analysis alone.Table 1: Exemplary Protein Biomarkers.

[0314] While various embodiments of the present disclosure have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only. Numerous variations, changes, and substitutions will now occur to those skilled in the art without departing from the disclosure. It should be understood that various alternatives to the embodiments of the disclosure described herein may be employed in practicing the disclosure. It is intended that the following claims define the scope of the disclosure and that methods and structures within the scope of these claims and their equivalents be covered thereby.

Claims

CLAIMSWHAT IS CLAIMED IS:

1. A method of detecting cancer in a subject, the method comprising:(a) measuring levels of first target molecules from a first sample of the subject, wherein the first target molecules comprise cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer types;(b) measuring levels of second target molecules from a second sample of the subject, wherein the second target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types;(c) applying a trained classifier to the measured levels of the first and second target molecules to assign an aggregate probability score for the cancer; wherein applying the trained classifier comprises: (i) applying a first trained model to the measured levels of the first target molecules to assign a first probability score for the cancer; (ii) applying a second trained model to the measured levels of the second target molecules to assign a second probability score for the cancer; and (iii) aggregating the first probability score and the second probability score; and(d) detecting the cancer by identifying that the aggregate probability score is above a threshold for presence of the cancer.

2. The method of claim 1, wherein the trained classifier was trained using reference first probability scores from the first trained model, reference second probability scores from the second trained model, and reference aggregated probability scores aggregating the reference first probability scores and reference second probability scores, for reference samples from (1) reference subjects having known cancers, and (2) reference subjects without cancer.

3. The method of claim 1, wherein the trained classifier assigns an aggregate probability score for each of a plurality of different cancer types, and detecting the cancer comprises identifying the cancer type as the cancer type with the highest aggregate probability score.

4. The method of claim 1, wherein aggregating the first probability score and the second probability score comprises calculating a product of the first and second probability scores for the cancer.

5. The method of claim 1, wherein the first sample and the second sample are the same.

6. The method of claim 1, wherein the plurality of different target genomic regions comprises at least 1000, 5000, 10000, 20000, or 30000 target genomic regions.

7. The method of claim 1, wherein the plurality of target genomic regions comprises a total collective length of at least 50 kb, 100 kb, 500 kb, or 1000 kb.

8. The method of claim 1, wherein each of the plurality of different target genomic regions comprises at least five methylation sites.

9. The method of claim 1, wherein measuring the first target molecules comprises sequencing converted cfDNA from the plurality of different target genomic regions, or amplification products thereof, wherein the converted cfDNA comprises cfDNA treated with a deaminating agent.

10. The method of claim 9, further comprising treating the cfDNA with the deaminating agent, optionally wherein the deaminating agent is a cytosine deaminase or bisulfite.

11. The method of claim 9, wherein the sequencing produces at least 100,000 sequencing reads.

12. The method of claim 8, wherein measuring the first target molecules comprises enriching for the converted cfDNA or amplification products thereof, to produce an enriched sample of polynucleotides.

13. The method of claim 12, wherein the enriching comprises capturing the converted cfDNA or amplification products thereof with a plurality of corresponding bait oligonucleotides.

14. The method of claim 13, wherein the plurality of different target genomic regions for enrichment by the bait oligonucleotides are genomic regions identified by the first trained model as differentially methylated in the at least one of a plurality of cancer types relative to non-cancer tissue or relative to cancer of a different type.

15. The method of claim 1, wherein the plurality of different polypeptides comprise at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides.

16. The method of claim 15, wherein the plurality of different polypeptides comprise: (a) polypeptides identifying proteins selected from List 1; (b) polypeptides identifying proteins selected from any one of Lists 2-19; (c) polypeptides identifying proteins selected from List 20; or (d) polypeptides identifying one or more of CHAD, KRT19, MMP12, PTN, SERPINA3, and17. The method of claim 1, wherein the trained classifier discriminates a subject with cancer from a subject without cancer with a defined specificity for each of the plurality of cancer types.

18. The method of claim 1, wherein the trained classifier has a higher sensitivity for cancer detection than each of the first trained model and the second trained model; optionally wherein the trained classifier has a specificity for cancer detection that is equal to or greater than each of the first trained model and the second trained model.

19. The method of claim 1, wherein the trained classifier is a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier.

20. The method of claim 1, wherein the first trained model and / or the second trained model is a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier.

21. The method of claim 1, wherein the first trained model binarizes measured levels of the first target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected.

22. The method of claim 1, wherein the second trained model log-transforms measured levels of the second target molecules normalized against control protein present in a known amount.

23. The method of claim 1, wherein (a) the first trained model is trained using measured levels of the first target molecules for first reference samples, (b) the second trained model is trained using measured levels of the second target molecules for second reference samples, and (c) the first and second reference samples comprise samples from reference subjects having known cancers, and reference subjects without cancer.

24. The method of claim 1, wherein the first sample and / or second sample comprises a biological fluid; optionally where the biological fluid comprises blood, plasma, serum, urine, saliva, pleural fluid, pericardial fluid, cerebrospinal fluid (CSF), peritoneal fluid, or any combination thereof.

25. The method of claim 24, wherein the biological fluid comprises blood, a blood fraction, plasma, or serum.

26. The method of claim 25, wherein the first sample and / or second sample is a plasma sample.

27. The method of claim 1, wherein the plurality of cancer types comprise at least 10 cancer types.

28. The method of claim 1, wherein the plurality of cancer types comprises one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

29. The method of any one of claims 1-28, further comprising treating the subject for the cancer type.

30. The method of claim 29, wherein the treating comprises surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

31. A method of treating cancer in a subject, the method comprising selecting a subject based on the results of a detection assay, and treating the subject for the cancer, wherein:(a) the detection assay comprises the method of any one of claims 1-28; and(b) the treating comprises surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

32. A method of training a classifier for detecting target molecules from a cancer, the method comprising:(a) receiving first measured levels of first target molecules for first samples of reference subjects, wherein (i) the first target molecules comprise cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of a plurality of cancer types, and (ii) the reference subjects comprise first subjects having known cancer types, and second subjects without cancer;(b) training a first model to generate a first probability score for the presence of cancer in a subject by applying a first machine learning algorithm to the first measured levels;(c) receiving second measured levels of second target molecules for second samples of the reference subjects, wherein the second target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of the plurality of cancer types;(d) training a second model to generate a second probability score for the presence of cancer in a subject by applying a second machine learning algorithm to the second measured levels;(e) generating reference first cancer probability scores for the first samples using thetrained first model;(f) generating reference second cancer probability scores for the second samples using the trained second model;(g) generating reference aggregated cancer probability scores for a plurality of the reference subjects by aggregating the reference first cancer probability score and reference second cancer probability score for each respective reference subject; and(h) training a classifier to generate an aggregate cancer probability score for a subject by applying a third machine learning algorithm to the reference first cancer probability scores, reference second cancer probability scores, and reference aggregated cancer probability scores.

33. The method of claim 32, wherein aggregating the first cancer probability score and the second cancer probability score comprises calculating a product of the first and second probability scores for the cancer.

34. The method of claim 32, wherein the first machine learning algorithm, second machine learning algorithm, and / or third machine learning algorithm is an LI -regularized logistic regression, an L2-regularized logistic regression, a generalized linear model (GLM), a random forest, a multinomial logistic regression, a multilayer perceptron, a support vector machine, or a neural network.

35. The method of claim 32, wherein the first trained model binarizes measured levels of the first target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected.

36. The method of claim 32, wherein the second trained model log-transforms measured levels of the second target molecules normalized against control protein present in a known amount.

37. The method of claim 32, wherein (a) the first trained model is trained using measured levels of the first target molecules for first reference samples, (b) the second trained model is trained using measured levels of the second target molecules for second reference samples, and (c) the first and second reference samples comprise samples from reference subjects having known cancers, and reference subjects without cancer.

38. The method of claim 32, wherein the third machine learning algorithm is a logistic regression.

39. The method of claim 32, wherein the plurality of different target genomic regions comprises at least 1000, 5000, 10000, 20000, or 30000 target genomic regions.

40. The method of claim 32, wherein the plurality of target genomic regions comprises a total collective length of at least 50 kb, 100 kb, 500 kb, or 1000 kb.

41. The method of claim 32, wherein each of the plurality of different target genomic regions comprises at least five methylation sites.

42. The method of claim 32, wherein the first measured levels comprise sequencing results for the cfDNA or amplicons thereof.

43. The method of claim 42, wherein the sequencing results comprise at least 100,000 reads for each of the first samples.

44. The method of claim 32, wherein the plurality of different polypeptides comprise at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides.

45. The method of claim 44, wherein the plurality of different polypeptides comprise: (a) polypeptides identifying proteins selected from List 1; (b) polypeptides identifying proteins selected from any one of Lists 2-19; (c) polypeptides identifying proteins selected from List 20; or (d) polypeptides identifying one or more of CHAD, KRT19, MMP12, PTN, SERPINA3, and SPP1.

46. The method of claim 32, wherein the plurality of cancer types comprise at least 10 cancer types.

47. The method of claim 32, wherein the plurality of cancer types comprises one or more of anorectal cancer, bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

48. A method of screening for cancer in a subject, the method comprising:(a) measuring levels of first target molecules from a first sample of the subject, wherein the first target molecules comprise a plurality of different polypeptides that are differentially expressed in at least one of a plurality of cancer types;(b) applying a first trained model to the measured levels of the first target molecules to assign a first probability score for each of the plurality of cancer types; wherein (i) the firsttrained model has a first specificity for cancer detection, and (ii) the first probability score for at least one of the cancer types is above a first threshold for the presence of cancer;(c) measuring levels of second target molecules from a second sample of the subject, wherein the second target molecules comprise cell-free DNA (cfDNA) from a plurality of different target genomic regions that are differentially methylated in at least one of the plurality of cancer types;(d) applying a second trained model to the measured levels of the second target molecules to assign a second probability score for the cancer; wherein (ii) the second trained model has a second specificity for cancer detection, and (ii) the second specificity is higher than the first specificity; and(e) detecting the cancer by identifying that the second probability score is above a threshold for the presence of cancer.

49. The method of claim 48, wherein (a) the first trained model is trained using measured levels of the first target molecules for first reference samples, (b) the second trained model is trained using measured levels of the second target molecules for second reference samples, and (c) the first and second reference samples comprise samples from reference subjects having known cancers, and reference subjects without cancer.

50. The method of claim 48, wherein the first sample and the second sample are the same.

51. The method of claim 48, wherein the plurality of different target genomic regions comprises at least 1000, 5000, 10000, 20000, or 30000 target genomic regions.

52. The method of claim 48, wherein the plurality of target genomic regions comprises a total collective length of at least 50 kb, 100 kb, 500 kb, or 1000 kb.

53. The method of claim 48, wherein each of the plurality of different target genomic regions comprises at least five methylation sites.

54. The method of claim 48, wherein measuring the second target molecules comprises sequencing converted cfDNA from the plurality of different target genomic regions, or amplification products thereof, wherein the converted cfDNA comprises cfDNA treated with a deaminating agent.

55. The method of claim 54, wherein the sequencing produces at least 100,000 sequencing reads.

56. The method of claim 54, wherein measuring the first target molecules comprises enriching for the converted cfDNA or amplification products thereof, to produce an enriched sample of polynucleotides.

57. The method of claim 56, wherein the enriching comprises capturing the converted cfDNA or amplification products thereof with a plurality of corresponding bait oligonucleotides.

58. The method of claim 57, wherein the plurality of different target genomic regions for enrichment by the bait oligonucleotides are genomic regions identified by the second trained model as differentially methylated in the at least one of a plurality of cancer types relative to non-cancer tissue or relative to cancer of a different type.

59. The method of claim 48, wherein the plurality of different polypeptides comprise at least 5, 10, 25, 50, 100, 200, 500, 1000, 2000, 3000, 5000, or 7500 different polypeptides.

60. The method of clam 59, wherein the plurality of different polypeptides comprise: (a) polypeptides identifying proteins selected from List 1; (b) polypeptides identifying proteins selected from any one of Lists 2-19; (c) polypeptides identifying proteins selected from List 20; or (d) polypeptides identifying one or more of CHAD, KRT19, MMP12, PTN, SERPINA3, and SPP1.

61. The method of claim 48, wherein the first trained model and / or the second trained model is a neural network classifier, a binary classifier, a mixture model classifier, a multilayer perceptron model classifier, or a logistic regression classifier.

62. The method of claim 48, wherein the second trained model binarizes measured levels of the second target molecules by assigning a first value if a target genomic region is detected, and a second value if a target genomic region is not detected.

63. The method of claim 48, wherein the first trained model log-transforms measured levels of the first target molecules normalized against a control polypeptide present in a known amount.

64. The method of claim 48, wherein (a) the first sample and / or second sample comprises a biological fluid; optionally where the biological fluid comprises blood, plasma, serum, urine, saliva, pleural fluid, pericardial fluid, cerebrospinal fluid (CSF), peritoneal fluid, or any combination thereof; (b) the first sample and / or second sample is a plasma sample.

65. The method of claim 48, wherein (a) the plurality of cancer types comprise at least 10 cancer types; and / or (b) the plurality of cancer types comprises one or more of anorectal cancer,bladder cancer, colorectal cancer, esophageal cancer, head and neck cancer, liver cancer, bile duct cancer, lung cancer, ovarian cancer, pancreatic cancer, stomach cancer, breast cancer, prostate cancer, kidney cancer, cervical cancer, endometrial cancer, and hematological cancer.

66. The method of any one of claims 48-65, further comprising treating the subject for the cancer type; optionally wherein the treating comprises surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

67. A method of treating a cancer in a subject, the method comprising selecting a subject based on the results of a screening assay, and treating the subject for the cancer, wherein:(a) the screening assay comprises the method of any one of claims 48-65; and(b) the treating comprises surgical resection, radiation therapy, chemotherapy, and / or immunotherapy.

68. A non-transitory computer-readable medium with instructions stored thereon, that when executed by one or more processors, perform one or more steps in the method of any one of claims 1-28 or 48-65.

69. A non-transitory computer-readable medium with instructions stored thereon, that when executed by one or more processors, perform the method of any one of claims 32-47.

Images