Conditional return of tissue of origin for localization accuracy

Abstract

Disclosed herein are systems and methods for localization of a disease state (e.g., tissue of origin) using a nucleic acid sample. In one embodiment, the method includes receiving a plurality of cancer signals of a sample, each cancer signal indicating a probability that the sample is associated with a different disease state of the plurality of disease states. The method determines a first cancer signal having a maximum probability from among the plurality of cancer signals. In response to determining that the first cancer signal meets the criterion, the method associates the sample with the first disease state. In response to determining that the first cancer signal does not meet the criterion, the method determines that a second cancer signal has a second maximum probability from among the plurality of cancer signals and associates the sample with the first disease state and the second disease state.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority to U.S. Provisional Application No. 63 / 171,355, filed April 6, 2021, which is incorporated by reference in its entirety for all purposes. Background technology [Technical field]

[0002] This disclosure relates generally to conditional return of tissue of origin determination for localization of a disease state. [Background technology]

[0003] Models can be trained to predict the tissue of origin of suspected cancers. However, due to biological ambiguity, there may be multiple plausible tissues of origin predictions. For example, biological samples with different tissues of origin may have similar characteristics. It is difficult for a physician or other health care provider to analyze ambiguous or complex cancer signals to determine a diagnosis for an individual. Samples with low tumor shedding (e.g., early cancer) also have low informative fragments, making localization diagnosis difficult. Summary of the Invention

[0004] Disclosed herein are methods for localizing a disease state (e.g., the presence or absence of cancer, the type of cancer, and / or the tissue of origin of the cancer (also referred to herein as the "cancer signal origin")) using a nucleic acid sample. The embodiments disclosed herein provide improvements to existing techniques in the field of cancer diagnostics and early detection using non-invasive methods. In one aspect, the disclosure provides 1. A method for cancer diagnosis comprising: receiving a first plurality of cancer signals of a first sample of a first individual, each of the first plurality of cancer signals indicative of a probability that the first sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a maximum probability among the first plurality of cancer signals; in response to determining that the first cancer signal satisfies a criterion, associating the first sample with a disease state corresponding to the first cancer signal; providing the disease status corresponding to the first cancer signal associated with the first sample for display on a client device to determine a first diagnosis for the first individual; receiving a second plurality of cancer signals of a second sample of a second individual, each of the second plurality of cancer signals indicative of a probability that the second sample is associated with a different one of the plurality of disease states; determining a second cancer signal having a maximum probability among the second plurality of cancer signals; in response to determining that the second cancer signal does not meet the criterion, associating the second sample with a subset of the plurality of disease states that corresponds to a subset of the second plurality of cancer signals that includes at least the second cancer signal; providing the subset of the plurality of disease states corresponding to the subset of the second plurality of cancer signals associated with the second sample for display on the client device to determine a second diagnosis for the second individual; A method is disclosed that includes:

[0005] In some embodiments, the disclosed method, system, or non-transitory computer readable medium comprises: determining a third cancer signal having a second highest probability among the second plurality of cancer signals, the subset of the second plurality of cancer signals further including the third cancer signal.

[0006] In some embodiments, the criterion is a probability threshold: Determining that the first cancer signal meets the criteria includes determining that the maximum probability of the first cancer signal is greater than the probability threshold, hi some embodiments, the probability threshold is at least 88%, 89%, 90%, 91%, or 92%.

[0007] In some embodiments, the disclosed method, system, or non-transitory computer readable medium further comprises determining a criterion based on the accuracy and false positives of the cancer signal probability.

[0008] In some embodiments, the disclosed method, system, or non-transitory computer readable medium further includes determining a criteria based on a current residual risk of cancer associated with the sample.

[0009] In some embodiments, the disclosed method, system, or non-transitory computer readable medium further comprises: determining a subset of n cancer signals of the first plurality of cancer signals having n highest probabilities among the first plurality of cancer signals; in response to determining that a threshold number of at least one of the subset of the first plurality of cancer signals is associated with a classification of disease states, associating the first sample with each disease state of the classification of disease states; Includes.

[0010] In some embodiments, The category of disease states is Human Papillomavirus (HPV) cancer. Said classification of disease states includes gastric cancer and intestinal cancer.

[0011] In some embodiments, the multiple disease conditions include non-cancerous conditions.

[0012] In some embodiments, the plurality of disease conditions includes: The cancer types include one or more types of cancer selected from the group consisting of anal cancer, breast cancer, uterine cancer, cervical cancer, ovarian cancer, bladder cancer, urothelial carcinoma of the renal pelvis and ureter, renal cancer other than urothelial carcinoma, prostate cancer, anorectal cancer, colorectal cancer, squamous cell carcinoma of the esophagus, esophageal cancer other than squamous cell carcinoma, gastric cancer, hepatobiliary cancer arising from hepatocytes, hepatobiliary cancer arising from cells other than hepatocytes, pancreatic cancer, human papillomavirus-associated head and neck cancer, head and neck cancer not associated with human papillomavirus, lung adenocarcinoma, small cell lung cancer, lung squamous cell carcinoma, lung cancer other than adenocarcinoma or small cell lung cancer, neuroendocrine cancer, melanoma, thyroid cancer, sarcoma, multiple myeloma, lymphoma, leukemia, kidney cancer, liver cancer, bile duct cancer, plasma cell neoplasm cancer, upper gastrointestinal tract cancer, vulvar cancer, lung neuroendocrine tumors, and other high-grade neuroendocrine tumors.

[0013] In some embodiments, the method, system, or non-transitory computer readable medium of the present disclosure further includes providing a graphical comparison of each disease state corresponding to the subset of the plurality of disease states associated with a second sample for display on a client device to determine a diagnosis, in some embodiments, the graphical comparison is a bar plot based on the probabilities of the second plurality of cancer signals.

[0014] In another aspect, the present disclosure provides a method for producing a method for manufacturing a pharmaceutical composition comprising: 1. A system including a computer processor and a memory, the memory storing computer program instructions that, when executed by the computer processor, cause the processor to perform steps including: receiving a first plurality of cancer signals of a first sample of a first individual, each of the first plurality of cancer signals indicative of a probability that the first sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a maximum probability among the first plurality of cancer signals; in response to determining that the first cancer signal satisfies a criterion, associating the first sample with a disease state corresponding to the first cancer signal; providing the disease status corresponding to the first cancer signal associated with the first sample for display on a client device to determine a first diagnosis for the first individual; receiving a second plurality of cancer signals of a second sample of a second individual, each of the second plurality of cancer signals indicative of a probability that the second sample is associated with a different one of the plurality of disease states; determining a second cancer signal having a maximum probability among the second plurality of cancer signals; in response to determining that the second cancer signal does not meet the criterion, associating the second sample with a subset of the plurality of disease states that corresponds to a subset of the second plurality of cancer signals that includes at least the second cancer signal; providing the subset of the plurality of disease states corresponding to the subset of the second plurality of cancer signals associated with the second sample for display on the client device to determine a second diagnosis for the second individual; The present invention provides a system including:

[0015] In another aspect, the present disclosure provides: A non-transitory computer-readable medium containing instructions that, when executed by one or more processors, cause the one or more processors to perform steps, the steps including: receiving a first plurality of cancer signals of a first sample of a first individual, each of the first plurality of cancer signals indicative of a probability that the first sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a maximum probability among the first plurality of cancer signals; in response to determining that the first cancer signal satisfies a criterion, associating the first sample with a disease state corresponding to the first cancer signal; providing the disease status corresponding to the first cancer signal associated with the first sample for display on a client device to determine a first diagnosis for the first individual; receiving a second plurality of cancer signals of a second sample of a second individual, each of the second plurality of cancer signals indicative of a probability that the second sample is associated with a different one of the plurality of disease states; determining a second cancer signal having a maximum probability among the second plurality of cancer signals; in response to determining that the second cancer signal does not meet the criterion, associating the second sample with a subset of the plurality of disease states that corresponds to a subset of the second plurality of cancer signals that includes at least the second cancer signal; providing the subset of the plurality of disease states corresponding to the subset of the second plurality of cancer signals associated with the second sample for display on the client device to determine a second diagnosis for the second individual; The present invention provides a system including:

[0016] In another aspect, the present disclosure provides: receiving a plurality of cancer signals of a sample, each of the plurality of cancer signals indicative of a probability that the sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a highest probability among the plurality of cancer signals; In response to determining that the first cancer signal satisfies a criterion, associating the sample with a first disease state corresponding to the first cancer signal; responsive to determining that the first cancer signal does not meet the criterion, determining a second cancer signal having a second greatest probability among the plurality of cancer signals; Associating the sample with the first disease state corresponding to the first cancer signal and with a second disease state corresponding to the second cancer signal; The present invention provides a method for localizing a cancer signal, comprising:

[0017] In some embodiments, the disclosed method, system, or non-transitory computer readable medium comprises: responsive to said determining that the first cancer signal satisfies the criterion, providing said first cancer signal as an input to a machine learning model to determine a prediction of cancer in the sample; responsive to the determination that the first cancer signal does not meet the criterion, providing the first cancer signal and the second cancer signal as inputs to the machine learning model to determine the prediction of cancer in the sample; Includes.

[0018] In some embodiments, the disclosed method, system, or non-transitory computer readable medium comprises: In response to the determination that the first cancer signal meets the criterion, creating a first training set including associations between the samples and the first disease state corresponding to the first cancer signal to train a machine learning model for cancer signal localization; In response to the determination that the first cancer signal does not meet the criterion, creating a second training set including associations of the samples with the first disease state corresponding to the first cancer signal and the second disease state corresponding to the second cancer signal to train the machine learning model;

[0019] In another aspect, the present disclosure provides a method for producing a method for manufacturing a pharmaceutical composition comprising: receiving a plurality of cancer signals of a sample, each of the plurality of cancer signals indicative of a probability that the sample is associated with a different one of a plurality of disease states; determining a first conditional probability that a first cancer signal of the plurality of cancer signals is a true positive given that remaining cancer signals of the plurality of cancer signals are inaccurate; in response to determining that the first conditional probability satisfies a criterion, associating the sample with at least one disease state corresponding to the first cancer signal; determining a subset of the plurality of cancer signals excluding the first cancer signal; determining a second conditional probability that a second cancer signal of the subset of the plurality of cancer signals is a true positive given that a remaining cancer signal of the subset of the plurality of cancer signals is inaccurate; in response to determining that the second conditional probability satisfies the criterion, associating the sample with at least one disease state corresponding to the second cancer signal; Includes.

[0020] In various embodiments, a system includes a computer processor and a memory, the memory including computer program instructions that, when executed by a computer processor, cause the processor to perform any of the methods described herein. In various embodiments, a non-transitory computer readable medium stores one or more programs including instructions that, when executed by an electronic device including a processor, cause the device to perform any of the methods described herein. [Brief description of the drawings]

[0021] [Figure 1A] FIG. 1A is a flowchart of a method for cancer signal localization according to various embodiments. [Figure 1B] FIG. 1B is a flowchart of another method for cancer signal localization, according to various embodiments. [Diagram 2] 2A shows a system for sequencing a nucleic acid sample according to various embodiments, and FIG 2B is a block diagram of a system for analyzing cancer signal localization according to various embodiments. [Diagram 3] FIG. 3 is a flow chart describing the process of sequencing a nucleic acid, according to various embodiments. [Figure 4] FIG. 4 shows experimental results of true positives and false positives during cancer signal localization, according to one embodiment. [Diagram 5] FIG. 5 is a flowchart of a method for conditional probability based cancer signal localization, according to various embodiments. [Figure 6]FIG. 6 shows experimental results of cancer signal localization according to one embodiment. [Figure 7] FIG. 7 shows experimental results of conditional return-based cancer signal localization, according to one embodiment. [Figure 8] FIG. 8 shows experimental results of cancer signal localization from a developed cancer sample, according to one embodiment. [Figure 9] FIG. 9 is a plot illustrating subsampling of samples according to one embodiment. [Figure 10A] 10A and 10B show detected cancer samples that have been subsampled to match expected screening cancer signal intensities, according to one embodiment. [Figure 10B] 10A and 10B show detected cancer samples that have been subsampled to match expected screening cancer signal intensities, according to one embodiment. [Figure 11A] 11A and 11B show cancer signal intensity by cancer type before and after subsampling according to some embodiments. [Figure 11B] 11A and 11B show cancer signal intensity by cancer type before and after subsampling according to some embodiments. [Figure 12] FIG. 12 shows cancer signal intensity by cancer type and stage, before and after subsampling, according to some embodiments. [Figure 13A] 13A and 13B include bar graphs of distributions of CSL call probabilities, such as the percentage of CSL signals captured by the first, second, third, and fourth CSL calls, according to some embodiments. [Figure 13B] 13A and 13B include bar graphs of distributions of CSL call probabilities, such as the percentage of CSL signals captured by the first, second, third, and fourth CSL calls, according to some embodiments. [Figure 14A] 14A and 14B include bar graphs of distribution of CSL call probabilities, such as the percentage of CSL signals captured by the first, second, third, and fourth CSL calls, by actual cancer type, according to some embodiments. [Figure 14B] 14A and 14B include bar graphs of distribution of CSL call probabilities, such as the percentage of CSL signals captured by the first, second, third, and fourth CSL calls, by actual cancer type, according to some embodiments. [Figure 15A] 15A, 15B, and 15C include bar graphs of median cancer scores divided into false positives and true positives, according to some embodiments. [Figure 15B] 15A, 15B, and 15C include bar graphs of median cancer scores divided into false positives and true positives, according to some embodiments. [Figure 15C] 15A, 15B, and 15C include bar graphs of median cancer scores divided into false positives and true positives, according to some embodiments. [Figure 16] FIG. 16 illustrates cumulative probability scores, according to some embodiments. [Figure 17A] 17A and 17B show conditional accuracy of cancer signal localization according to some embodiments. [Figure 17B] 17A and 17B show conditional accuracy of cancer signal localization according to some embodiments. [Figure 18A] 18A and 18B show the conditional accuracy of cancer signal localization for solid and liquid sample types, according to some embodiments. [Figure 18B] 18A and 18B show the conditional accuracy of cancer signal localization for solid and liquid sample types, according to some embodiments. [Figure 19A] 19A and 19B show conditional accuracy of cancer signal localization based on stage of cancer, according to some embodiments. [Figure 19B] 19A and 19B show conditional accuracy of cancer signal localization based on stage of cancer, according to some embodiments. [Figure 20A] 20A and 20B show cumulative accuracy of cancer signal localization according to some embodiments. [Figure 20B] 20A and 20B show cumulative accuracy of cancer signal localization according to some embodiments. [Figure 21A] 21A and 21B show false positive cancer signal localization according to some embodiments. [Figure 21B] 21A and 21B show false positive cancer signal localization according to some embodiments. [Figure 22A] 22A and 22B show false positive cancer signal localization based on cancer type, according to some embodiments. [Figure 22B] 22A and 22B show false positive cancer signal localization based on cancer type, according to some embodiments. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] Reference will now be made in detail to several embodiments, examples of which are illustrated in the accompanying drawings. It should be noted that, wherever feasible, like or similar reference numbers may be used in the figures and may indicate like or similar functionality. It should also be noted that the contents of all published materials (patent applications, patents, papers, conference proceedings, etc.) referenced herein are hereby incorporated by reference in their entirety.

[0023] I. Definition Unless otherwise defined, all technical and scientific terms used herein have the meaning commonly understood by one of ordinary skill in the art to which this specification belongs. As used herein, the following terms have the meanings set forth below.

[0024] The term "individual" refers to a human individual. The term "healthy individual" means an individual who is presumed to be cancer or disease free.

[0025] The term "subject" refers to an individual whose DNA is analyzed. A subject may be a subject whose DNA is evaluated using whole genome sequencing or a targeted panel as described herein to evaluate whether the person has a disease state (e.g., cancer, a type of cancer, or a tissue of origin of a cancer). A subject may also be part of a control group known to not have cancer or another disease. A subject may also be part of a cancer or other disease group known to have cancer or another disease. Control and cancer / disease groups may be used to aid in the design or validation of a targeted panel.

[0026] A "reference sample" refers to a sample obtained from a subject with a known disease state.

[0027] The term "training samples" refers to samples obtained from known disease states that can be used to generate sequence reads. The training samples can be applied to a model of probability to generate features that can be used for disease state classification.

[0028] A "test sample" refers to a sample that may have an unknown disease state.

[0029] The term "sequence read" refers to a base sequence read from a sample obtained from an individual. A sequence read may be generated from a nucleic acid fragment in a sample. A sequence read may be a collapsed sequence read generated from multiple sequence reads derived from multiple amplicons from a single origin nucleic acid molecule. In some embodiments, a sequence read may be a de-duplicated sequence read. A sequence read may be obtained through various methods known in the art.

[0030] The term "disease state" refers to the presence or absence of an infection, the type of infection, and / or the tissue of origin of the infection. For example, in one embodiment, the present disclosure provides methods, systems, and non-transitory computer-readable media that detect risk (i.e., the presence or absence of risk), the type of risk, or the tissue of origin of risk.

[0031] "Tissue of origin," or "TOO," refers to an organ, group of organs, body area, or cell type in which a disease state originates or may originate. For example, identification of a tissue of origin or cancer cell type typically allows for identification of appropriate next steps to determine further diagnosis, staging, and treatment.

[0032] The term "methylation" as used herein refers to the chemical process by which a base is added to a DNA molecule. Two of the four bases in DNA, cytosine ("C") and adenine ("A"), can undergo metabolism. For example, the hydrogen atom on the pyrimidine ring of the cytosine base can be converted to a methyl group to form 5-methylcytosine. Methylation tends to occur at the dinucleotides of cytosine and guanine, referred to herein as "CpG sites." In other instances, methylation can occur at a cytosine that is not part of a CpG site, or at another nucleotide that is not a cytosine. However, these occur rarely. In this disclosure, for clarity, methylation is discussed with reference to CpG sites. However, the principles described herein are equally applicable to the detection of methylation in non-CpG contexts (including non-cytosine methylation). For example, adenine methylation has been observed in bacterial, plant and mammalian DNA, but has received less attention.

[0033] In such embodiments, the wet lab assays used to detect methylation may vary from those described herein, as is well known in the art. Additionally, the methylation state vector may include a component that is a general vector of methylated or unmethylated sites (even if those sites are not specifically CpG sites). With that substitution, the remaining processes described herein remain the same, and thus the inventive concepts described herein are applicable to those other forms of methylation.

[0034] The term "CpG site" refers to a region of nucleotide molecules that contains a cytosine followed by a guanine in the linear sequence of nucleotides along the 5' to 3' direction. "CpG" is an abbreviation for 5'-C-phosphate-G-3', where the cytosine and guanine are separated by only one phosphate group. The phosphate joins the two nucleotides together in DNA. The cytosine in a CpG dinucleotide is methylated to form 5-methylcytosine.

[0035] The terms "cell-free deoxyribonucleic acid," "cell-free DNA," or "cfDNA" refer to deoxyribonucleic acid fragments that circulate in bodily fluids, such as blood, sweat, urine, or saliva, and that are derived from one or more healthy cells and / or one or more cancer cells.

[0036] The term "circulating tumor DNA" or "ctDNA" refers to deoxyribonucleic acid fragments derived from tumor cells or other types of cancer cells. It can be released into individual body fluids such as blood, sweat, urine, or saliva as a result of biological processes such as apoptosis or necrosis of dying cells, or actively released by viable tumor cells.

[0037] II. Overview of localization methods FIG 1A is a flow chart of a method 100 for cancer signal localization according to various embodiments. FIG 2B is a block diagram of an analysis system 200 for cancer signal localization according to various embodiments. In the embodiment shown in FIG 2B, the analysis system 200 includes a sequence processor 210, a machine learning engine 220, a probabilistic model 230, a classifier 240, and a localization engine 250. In various embodiments, the analysis system 200 performs any of the methods described herein. The method 100 includes, but is not limited to, the following steps:

[0038] In step 110, the localization engine 250 receives a first set of cancer signals for a first sample. The cancer signals are also referred to as "probability scores" or "cancer scores." Each cancer signal in the first set of cancer signals indicates a probability that the first sample is associated with a different disease state in the set of disease states. Each (probability) cancer signal can be specified on a scale from 0% to 100%, 0 to 100, or 0 to 1. The cancer signals in the first set can sum to 100%, 100, or 1.

[0039] The cancer signal may be generated by one or more classifiers 240. In various embodiments, the classifier 240 generates the cancer signal by processing sequence reads of the sample. The sequence processor 210 may generate the sequence reads of the sample. In some embodiments, the signal is associated with a disease state other than cancer. For example, the disease state may include a medical or physiological condition, a general disorder, a health-related measurement, and other types of infections.

[0040] In various embodiments, the classifier 240 generates a set of 22 cancer signals, including cancer signals for 21 different cancer types and one non-cancerous signal. The 21 cancer types include: anal; bladder and urothelium; breast; cervix; colon and rectum; head and neck; kidney; liver and bile duct; lung; neuroendocrine cells of lung or other organs; lymphoid; melanocytic; myeloid; ovarian; pancreas and gallbladder; plasma cell; prostate; bone and soft tissue; thyroid and esophagus; and uterus. In other embodiments, the classifier generates a set that includes a different number of cancer signals or a set that includes different types of disease states than the above list.

[0041] In step 120, localization engine 250 determines a first cancer signal that has a maximum probability among the first set of cancer signals. In step 130, in response to determining that the first cancer signal meets the criteria, localization engine 240 associates the first sample with at least one disease state corresponding to the first cancer signal. For example, localization engine 250 can report a prediction that the first sample is associated with a cancer having an origin tissue indicated by the disease state. In some embodiments, localization engine 250 reports only the disease state corresponding to the first cancer signal; i.e., localization engine 250 does not report predictions of disease states corresponding to other cancer signals of the first set of cancer signals. Reporting only one disease state when the criteria are met can help reduce the complexity of the output provided by analysis system 200 that may aid in a physician's practice.

[0042] In various embodiments, the criterion is a probability threshold of 90% positive cancer score. That is, the localization engine 250 determines whether the classifier 240 assigns 90% of the mass of the cancer signal origin score mass to the first cancer signal (which corresponds to the disease state). In some embodiments where the set of cancer signals includes 22 cancer types as described above, the probability threshold does not take into account one non-cancer signal. That is, the localization engine 250 determines whether the classifier 240 assigns 90% of the cancer signal organization of the origin score mass among the 21 cancer signals to the first cancer signal. In other embodiments, the probability threshold accounts for one non-cancer signal in addition to the cancer signal indicating the presence of cancer. In other embodiments, the criterion may be another predefined probability threshold, such as, for example, 88%, 89%, 91%, 92%, etc.

[0043] In various embodiments, the localization engine 250 determines a criterion based on the cancer signal probability and the accuracy of false positives. Selecting a criterion probability threshold that increases the proportion of correctly detected true positives also increases the number of false positives; i.e., falsely predicting the presence of cancer in healthy samples where the presence of cancer is not actually found. This trade-off relationship is illustrated in the plot 400 shown in FIG. 4. At lower probability thresholds, the marginal benefit to detecting true positives is higher. At probability thresholds above 90%, the marginal benefit of detecting true positives decreases because the proportion of false positives increases. In embodiments, the localization engine 250 determines the probability threshold by determining an inflection point of the curve on the plot 400 of true positive versus false positive detection. Based on the inflection point, the localization engine 250 determines that a probability threshold, for example 90%, is optimal. This is because using a probability threshold to determine a cancer prediction improves the accuracy of true positive detection while reducing the risk of false positive detection. The probability threshold improves over conventional methods that do not consider the risk of false positives when making true positive predictions. Conventional methods with high false positive rates reduce the overall accuracy of the predictions. Thus, the present probability threshold is advantageous for practical applications in determining cancer predictions, especially in non-invasive procedures, e.g., using blood samples instead of tissue biopsies that require surgery.

[0044] In step 140, the localization engine 250 receives a second set of cancer signals of a second sample. The first sample and the second sample may be from two different patients or the same patient. The samples may include any cell-free nuclear material sample (e.g., cfDNA), solid samples, and / or other types of biological samples. Each cancer signal of the second set of cancer signals indicates a probability that the second sample is associated with a different disease state of disease states (e.g., of the same set for the first set of cancer signals).

[0045] At step 150, the localization engine 250 determines a second cancer signal having a maximum probability among the second set of cancer signals. At step 160, in response to determining that the second cancer signal does not meet the criteria, the localization engine 250 associates the second sample with a subset of the set of disease states that corresponds to the subset of the set of second cancer signals. In some embodiments, the subset of the second set of cancer signals can include the cancer signals having the two largest probabilities among the second set of cancer signals. In other embodiments, the subset of the second set of cancer signals can include a different number of cancer signals, such as, for example, 3, 4, 5, or more cancer signals.

[0046] In some embodiments, the localization engine 250 determines a first set of cancer signals having n highest probabilities among the first set of subset n cancer signals. In response to determining that at least a threshold number of the subset cancer signals are associated with a disease state of a classification, the localization engine 250 associates the first sample with each disease state of the disease state classification. For example, the disease state classification is human papillomavirus (HPV) cancer. In another example, the disease state classification includes gastric cancer and intestinal cancer. In other embodiments, the disease state classification can include one or more other types of cancer.

[0047] In some embodiments, the localization engine 250 can determine a baseline based on the current conference residual risk associated with the sample (the risk that an individual will be diagnosed with cancer). For example, the localization engine 250 may determine to report additional cancer signals based on the conditional probability of cancer given an inaccurate tissue of origin prediction, where v is a ranked sorting vector of calibrated tissue of origin probabilities.

number

[0048] The localization engine 250 can determine the probability that an individual has cancer after a positive cancer test where cancer is not detected in a first tissue of origin. Cancer may be detected in a second or third tissue of origin.

[0049] The localization engine 250 can present a disease state determination (e.g., localization of the tissue of origin of the cancer) to a user, such as a physician, doctor, or physician, among other types of health care providers. For example, the localization engine 250 provides the user with a disease state corresponding to a first cancer signal associated with a first sample for display on a client device. The localization engine 250 can provide a graphical comparison for each disease state, which corresponds to a subset of the set of graphical comparisons associated with a second disease state. In various embodiments, the graphical comparison is a bar plot based on the probability of the second set of cancer signals. By presenting a visual depiction of the probability, the user can intuitively interpret the information output by the localization engine 250. For example, the graphical comparison can suggest that the user place more weight on the tissue of origin having a greater probability of being a true positive tissue of origin of the detected cancer.

[0050] FIG. 1B is a flow chart of another method 170 for cancer signal localization, according to various embodiments. The method 170 includes, but is not limited to, the following steps:

[0051] In step 172, the localization engine 250 receives a set of sample cancer signals. Each cancer signal in the set of cancer signals indicates a probability that the sample is associated with a different disease state of the set of disease states. In step 174, the localization engine 250 determines a first cancer signal in the set of cancer signals that has the greatest probability.

[0052] In step 176, following a determination that the first cancer signal meets a criterion (such as any of the criteria described above), the localization engine 250 associates the sample with a first disease state that corresponds to the first cancer signal.

[0053] In step 178, in accordance with a determination that the first cancer signal does not meet the criterion, the localization engine 250 determines a second cancer signal having a second highest probability among the set of cancer signals. In step 180, the localization engine 250 associates the sample with a disease state corresponding to the first cancer signal and a second disease state corresponding to the second cancer signal. In other words, the localization engine 250 associates the sample with the cancer signal having the highest two probabilities among the second set of cancer signals.

[0054] 5 is a flowchart of a method 500 for cancer signal localization based on conditional probability, according to various embodiments. Instead of using a predetermined probability threshold, the localization engine 250 can determine the threshold based on the conditional probability of the nth cancer signal being accurate when the previous n-1 cancer signals are inaccurate. In this case, the localization engine 250 can continue to return cancer signals as long as P(nth cancer signal is accurate and n-1 cancer signal is inaccurate) meets a criterion such as exceeding a threshold probability. The method 500 includes, but is not limited to, the following steps:

[0055] In step 510, the localization engine 250 receives a set of sample cancer signals. Each cancer signal indicates a probability that the sample is associated with a different disease state of a set of disease states.

[0056] At step 520, the localization engine 250 determines a first conditional probability that a first cancer signal in the set is a true positive if the remaining cancer signals in the set are inaccurate. At step 530, in response to determining that the first conditional probability meets a criterion, the localization engine 250 associates the sample with at least one disease state that corresponds to the first cancer signal.

[0057] At step 540, the localization engine determines a subset of the plurality of cancer signals excluding the first cancer signal. At step 550, the localization engine determines a second conditional probability that a second cancer signal of the subset of cancer signals is a true positive given that the remaining cancer signals of the subset of cancer signals are inaccurate. At step 560, in response to determining that the second conditional probability meets the criterion, the localization engine 250 associates the sample with at least one disease state corresponding to the second cancer signal.

[0058] II.A. Testing Protocol 3 is a flow chart describing a process 300 for sequencing nucleic acids, according to one embodiment. In an embodiment, process 300 is performed to generate sequence reads that are used by analysis system 200 to perform any of the methods for cancer signal localization described herein.

[0059] At step 310, a nucleic acid sample (e.g., DNA or RNA) is extracted from the subject. In this disclosure, DNA and RNA may be used interchangeably unless otherwise noted. That is, the embodiments described herein are applicable to both DNA and RNA types of nucleic acid sequences. However, the examples described herein may focus on DNA for clarity and illustration. The sample may include nucleic acid molecules derived from any subset of the human genome, including the entire genome. The sample may include blood, plasma, serum, urine, feces, saliva, other types of bodily fluids, or any combination thereof. In some embodiments, methods for drawing blood samples (e.g., syringe or finger prick) can be less invasive than procedures for obtaining tissue biopsies, which may require surgery. The extracted sample may include cfDNA and / or ctDNA. If the subject has a disease condition, such as cancer, the cell-free nucleic acid (e.g., cfDNA) in the sample extracted from the subject generally includes a detectable concentration of nucleic acid that can be used to assess the disease condition.

[0060] In step 315, the extracted nucleic acid (e.g., including cfDNA fragments) is treated to convert unmethylated cytosines to uracil. In some embodiments, the method 300 uses bisulfite treatment of the sample to convert unmethylated cytosines to uracil without converting them to methylated cytosines. For example, a commercially available kit such as EZ DNA Methylation®-Gold, EZ DNA Methylation®-Direct, or EZ DNA Methylation®-Lightning Kit (available from Zymo Research Corp, Irvine, CA) is used for bisulfite conversion. In another embodiment, the conversion of unmethylated cytosines to uracil is achieved using an enzymatic reaction. For example, this conversion can use a commercially available kit to convert unmetabolized cytosines to uracil, such as APOBEC-Seq (NEBiolabs, Ipswich, MA).

[0061] In step 320, a sequence library is prepared. In some embodiments, the preparation includes at least two steps. In the first step, ssDNA adapters are added to the 3'-OH ends of the bi-sulfite-converted ssDNA molecules using ssDNA ligation. In some embodiments, the ssDNA ligation uses CircLigase II (Epicentre) to ligate ssDNA adapters to the 3'-OH ends of the bi-sulfite-converted ssDNA molecules, where the 5' ends of the adapters are lysed and the bi-sulfite-converted ssDNA is dephosphorylated (i.e., the 3' ends have hydrolytic groups). In another embodiment, the ssDNA ligation uses Thermostable 5 AppDNA / RNA Ligers (available from New England BioLabs, Ipswich, Mass.) to ligate ssDNA adapters to the 3'-OH ends of the bi-sulfite-converted ssDNA molecules. In this example, the initial UMI adapters are adenylylated at the 5' end and blocked at the 3' end. In another embodiment, the ssDNA ligation reaction uses T4 RNA ligators (available from New England BioLabs) to ligate ssDNA adapters to the 3'-OH ends of bi-sulfite converted ssDNA molecules.

[0062] In the second step, the second strand DNA is synthesized in an extension reaction.For example, the extension primer that hybridizes to the primer sequence contained in the ssDNA adaptor is used in a primer extension reaction to form a double-stranded disulfite-converted DNA molecule.Optionally, in some embodiments, the extension reaction uses an enzyme that can read the uracil residue in the bisulfite-converted template strand.

[0063] Optionally, in a third step, dsDNA adapters are added to the double-stranded bisulfite-converted DNA molecules. The double-stranded bisulfite-converted DNA can then be amplified to add sequencing adapters. For example, a PCR amplification using a forward primer containing a P5 sequence and a reverse primer containing a P7 sequence is used to add P5 and P7 sequences to the bisulfite-converted DNA. Optionally, unique molecular identifiers (UMIs) can be added to the nucleic acid molecules (e.g., DNA molecules) via adapter ligation during library creation. UMIs are short nucleic acid sequences (e.g., 4-10 base pairs) that are added to the ends of DNA fragments during adapter ligation. In some embodiments, UMIs are degenerate base pairs that function as unique tags that can be used to identify sequence reads originating from a particular DNA fragment. During PCR amplification after adapter ligation, the UMI is replicated along with the ligated DNA fragment. This provides a way to identify sequence reads that came from the same original fragment in downstream analyses.

[0064] In optional step 325, the nucleic acids (e.g., fragments) can be hybridized. Hybridization probes (also referred to herein as "probes") can be used to target and pull down nucleic acid fragments that contain information useful for a disease state. For a given workflow, the probes can be designed to anneal (or hybridize) to a target (complementary) strand of DNA or RNA. The target strand can be the "positive" strand (e.g., the strand that is transcribed into mRNA and subsequently translated into protein) or the complementary "negative" strand. The probes can be either 10s, 100s, or 1000s in length. Additionally, the probes can cover overlapping portions of the target region.

[0065] In optional step 330, the hybridized nucleic acid fragments are captured and can be amplified, for example, using PCR. In some embodiments, the target DNA sequence can be enriched from the library. This is used, for example, when a panel assay of targets is being performed on the sample. For example, the target sequence can be enriched to obtain an abundant sequence that can be sequenced later. Generally, any method known in the art can be used to isolate and concentrate the probe-hybridized target nucleic acid. For example, as is known in the art, a biotin moiety can be added to the 5' end of the probe (i.e., biotinylation) and a streptavidin-coated surface (e.g., streptavidin-coated beads) can be used to facilitate isolation of the target nucleic acid hybridized to the probe.

[0066] In step 335, sequence reads are generated from the nucleic acid sample, e.g., the enriched sequences. Sequence data can be obtained from the enriched DNA sequences by methods known in the art. For example, methods can include next generation sequencing (NGS) techniques such as synthesis technology (Illumina), Pillows sequencing (454 Life Science), on-semiconductor technology (Ion Torrent sequencing), single molecule real-time sequencing (Pacific Biosciences), sequencing by ligature (SOLiD sequencing), Nanopore sequencing (Oxford Nanopore Technologies), paired-end sequencing, etc. In some embodiments, massively parallel sequencing is performed using sequencing by synthesis with reversible dye terminators.

[0067] II.B. Exemplary Sequencer and Analysis Systems 2A shows a system for sequencing a nucleic acid sample, according to various embodiments. The illustration includes devices such as a sequencer 270 and an analysis system 200. The sequencer 270 and the analysis system 200 may operate in parallel (in tandem) to perform one or more steps in the processes described herein.

[0068] In various embodiments, the sequencer 270 receives the enriched nucleic acid sample 260. As shown in FIG. 2A, the sequencer 270 can include a graphic user interface 275 that allows user interaction with specific tasks, and one or more loading stations 280 for loading a sequencing cartridge containing the enriched fragment sample and / or loading buffers required to perform a sequencing assay. Thus, once a user of the sequencer 270 provides the necessary reagents and sequencing cartridges to the loading station 280 of the sequencer 270, the user can initiate sequencing by interacting with the graphic user interface 275 of the sequencer 270. Once initiated, the sequencer 270 performs sequencing and outputs sequence reads of the enriched fragments from the nucleic acid sample 260.

[0069] In an embodiment, the sequencer 270 is communicatively coupled to the analysis system 200. The analysis system 200 includes several computing devices used to process sequence reads for various applications, such as evaluation of methylation status at one or more CpG sites, variant calling, or quality control. The sequencer 270 may provide the analysis system 200 with sequence reads in a BAM file format. The analysis system 200 can be communicatively coupled to the sequencer 270 via wireless, wired, or a combination of wireless and wired communication technologies. In general, the analysis system 200 is comprised of a processor and a non-transitory computer-readable storage medium that stores computer instructions that, when executed by the processor, cause the processor to process sequence reads or perform one or more steps of any method or process disclosed herein.

[0070] In some embodiments, sequence reads may be aligned to a reference genome using methods known in the art to determine alignment position information. The alignment position may generally describe the start and end positions of a region in the reference genome that corresponds to a beginning client base and an ending client base based on a particular sequence read. Corresponding to a methylation sequence, the alignment position information may be generalized to indicate the first and last CpG sites included in the sequence read according to the alignment to the reference genome. The alignment position information may further indicate the methylation state and position of all CpG sites in a given sequence read. A region in the reference genome may be associated with one or more sequences. For example, the analysis system 200 may label the sequence reads associated with one or more sequences associated with the sequence read. In one embodiment, the fragment length (or size) is determined from the start and end positions.

[0071] In various embodiments, for example, when a paired-end sequence process is used, the sequence reads are composed of a read pair denoted as R_1 and R_2. For example, the first read R_1 is sequenced from the first end of a double-stranded DNA (dsDNA) molecule, while the second read R_2 is sequenced from the second end of the double-stranded DNA (dsDNA). Thus, the base pairs of the first read R_1 and the second read R_2 can be consistently (e.g., in reverse) aligned with the bases of the reference genome. The aligned position information derived from the read pair R_1 and R_2 can include a start position in the reference genome corresponding to the end of the first read (e.g., R_1) and an end position in the reference genome corresponding to the end of the second read (e.g., R_2). In other words, the start and end positions of the reference genome represent likely positions in the reference genome to which the nucleic acid fragments correspond. In one embodiment, the read pair R_1 and R_2 can be assembled into fragments, which are used for subsequent analysis and / or classification. An output file having a SAM (sequence alignment map) format or a BAM (binary) format can be generated and output for further analysis.

[0072] Referring now to FIG. 2B, an analysis system 200 embodies one or more computing devices and / or one or more processors for use in analyzing DNA samples, sequence reads, or other information.

[0073] In some embodiments, the sequence processor 210 generates a methylation state vector of fragments from a sample. For each CpG site on a fragment, the sequence processor 210 generates a methylation state vector for each fragment that specifies the location of the fragment in the reference genome, a number of CpG sites in the fragment, and the methylation state of each CpG site in the fragment (whether methylated, unmethylated, or uncertain). The sequence processor 210 can store the methylation state vectors of fragments in a sequence database 215. The data in the sequence database 215 may be organized such that the methylation state vectors from the samples are related to each other.

[0074] Additionally, multiple different models 230 may be stored in model database 225 or retrieved for use with test samples. In one example, a model is a cancer classifier 240 trained to determine a gamma prediction for a test sample using feature vectors derived from abnormal fragments. The training and use of cancer classifiers is discussed elsewhere herein. Analysis system 200 may train one or more models 230 and / or one or more classifiers 240 and store various trained parameters in parameter database 235. Analysis system 200 stores models 230 and / or classifiers 240 in model database 225 along with functions.

[0075] During inference, the machine learning engine 220 uses one or more models 230 and / or classifiers 240 to return outputs. The machine learning engine accesses the models 230 and / or classifiers 240 in the model database 225 along with trained parameters from the parameter database 235. According to each model, the machine learning engine 220 receives appropriate inputs for the model and calculates outputs based on the function of each model related to the received inputs, parameters, and inputs / outputs. In some use cases, the machine learning engine 220 further calculates metrics that correlate to the confidence of the outputs calculated from the models. In other use cases, the machine learning engine 220 calculates other intermediate values ​​for use within the models.

[0076] III. Model-Based Feature Engineering and Classification III. A. Model-Based Feature Engineering In accordance with one embodiment, the present disclosure is directed to model-based feature engineering to derive features useful for classifying disease states. As described elsewhere herein, a disease state is the presence or absence of infection, the type of infection, and / or the infected tissue or origin. For example, as described herein, a disease state may be the presence or absence of cancer, the type of cancer, and / or the cancer origin tissue of origin. The type of cancer and / or tissue of origin may be selected from the group comprising breast cancer, uterine cancer, cervical cancer, ovarian cancer, bladder cancer, urothelial carcinoma of the renal pelvis, kidney cancer other than urothelial carcinoma, prostate cancer, anorectal cancer, colorectal cancer, esophageal cancer, gastric cancer, hepatobiliary cancer, hepatobiliary cancer arising from liver cells, hepatocellular carcinoma, pancreatic cancer, squamous cell carcinoma of the upper gastrointestinal tract, upper gastrointestinal tract cancer other than squamous cell carcinoma, head and neck cancer, lung cancer, lung cancer, e.g. lung adenocarcinoma, small cell lung cancer, squamous cell lung carcinoma and adenocarcinoma or cancer other than small cell lung cancer, neuroendocrine carcinoma, melanoma, thyroid cancer, multiple myeloma, lymphoma, and leukemia, among other cancer types.

[0077] The process generates a first plurality of sequence reads from a first sample having a first disease state and a second plurality of sequence reads from a second reference sample having a second disease state, as described elsewhere herein. The first plurality of sequence reads and / or the second plurality of sequence reads may be 10,000 or more, 50,000 or more, 100,000 or more, 200,000 or more, 500,000 or more, 1,000,000 or more, 2,000,000 or more, 5,000,000 or more, 10,000,000 or more sequence reads. As used herein, a "reference sample" is a sample obtained from a subject having a known disease state. In some embodiments, one or more reference samples having one or more known disease states can be used to train one or more probabilistic models that can be used to derive features for classifying the disease state of an unknown test sample. The samples can be genomic DNA (gDNA) samples or cell-free DNA (cfDNA) samples. The reference sample may be blood, plasma, serum, urine, feces, and saliva samples. Alternatively, the reference sample may be whole blood, blood fraction, tissue biopsy, pleural fluid, pericardial fluid, cerebrospinal fluid, and ascites. In some embodiments, the first reference sample is obtained from a subject known to have cancer, and the second reference sample is obtained from a normal or non-cancer subject. In some embodiments, the first reference sample is obtained from a subject known to have a first type of image (e.g., an infected cause), and the second reference sample is obtained from a subject known to have a second type of image (e.g., an infected case). In yet other embodiments, the first reference sample is obtained from a subject known to have a first source of infected tissue (e.g., an infection of the respiratory tract), and the second reference sample is obtained from a second source of diseased tissue (e.g., an infected tissue of the library).

[0078] Continuing with this process, the machine learning engine 220 trains a first probabilistic model 230 and a second probabilistic model 230 from the first and second plurality of sequence reads, each of which is associated with a different disease state of one or more possible disease states. As previously described, the disease state may be the presence or absence of cancer, the type of cancer, and / or the cancer tissue of origin. In various embodiments, the training data is split into K subsets (folds) for K-fold cross-validation. The folds may be balanced for cancer / non-cancer status, tissue of origin, stage of cancer, age (e.g., grouped in 10-year buckets), gender, ethnicity, and smoking status. Data from K1 of the folds may be used as training data for the probabilistic models, and the retained folds may be used as test data.

[0079] The machine learning engine 220 trains the first and second probabilistic models 230 for the first and second disease states, respectively, by fitting each of the probabilistic models 230 to the first and second plurality of sequence reads. For example, in one embodiment, the first probabilistic model is fitted using a first plurality of sequence reads derived from one or more samples from a subject known to have cancer, and the second probabilistic model is fitted using a second plurality of sequence reads derived from one or more samples from a normal or non-cancer subject. In other embodiments, the first probabilistic model may be trained for a first type of cancer or a first tissue of origin, and the second probabilistic model may be trained for a second type of cancer or a second tissue of origin. As will be appreciated by those skilled in the art, some disease state probabilistic models may be trained utilizing sequence reads derived from one or more samples taken from subjects having any one of a number of possible disease states. For example, in some embodiments, additional cancer-specific probabilistic models (i.e., for further types of cancer and tissue of origin) may be trained for a third, fourth, fifth, sixth, seventh, eighth, ninth, tenth, etc. (e.g., 20, 30, or more) specific type of cancer, as described elsewhere herein, and used to determine sequence reads from the training set or unknown cancer types, that are likely to originate from other cancer types (or cancer-derived tissues).

[0080] As used herein, a "probability model" is any mathematical model that can assign probabilities to sequence reads based on the methylation state at one or more sites on the read. During training, the machine learning engine 220 can be used to match sequence reads derived from one or more samples from subjects with known conditions and determine the sequence reads to a probability of indicating a disease state utilizing structure determination information or methylation state vectors. In particular, in one embodiment, the machine learning engine 220 determines the observed methylation rate for each CpG site in the sequence read. The methylation rate represents the fraction or rate of base pairs that become metabolites within the CpG site. The trained probability model 230 can be parameterized by the product of the methylation rate. Typically, any known probability model for assigning probabilities to sequence reads from a sample can be used. For example, the probability 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 it can be an independent site model, in which methylation of each CpG is specified by a separate methylation probability with methylation at a site that is assumed to be independent of methylation at one or more other sites on the nucleic acid fragment.

[0081] III.B. Disease state organization of origin classification According to various embodiments, the machine learning engine 220 trains the probabilistic models 230 associated with each of the various disease states of the collection of disease states. As previously mentioned, in various embodiments, a disease state may be the presence or absence of cancer, the type of cancer, and / or the tissue of origin of the cancer. Additionally, a disease state may be associated with another type of disease (not necessarily associated with cancer) or a health condition (absence of cancer or disease).

[0082] The machine learning engine 220 trains the probabilistic model 230 using one or more sets of sequence reads, each of which is generated from a different disease state of the set of multiple disease states. The disease states may include any number of types of cancer or tissue selected from the group of breast cancer, uterine cancer, cervical cancer, ovarian cancer, bladder cancer, urothelial carcinoma of the renal pelvis, kidney cancer other than urothelial carcinoma, urothelial carcinoma, prostate cancer, anorectal cancer, colon cancer, esophageal cancer, gastric cancer, hepatobiliary cancer, hepatobiliary cancer arising from liver cells, pancreatic cancer, squamous cell carcinoma of the upper digestive tract, squamous cell carcinoma of the upper digestive tract other than squamous cell carcinoma, head and neck cancer, lung cancer, e.g., lung adenocarcinoma, small cell lung cancer, squamous cell lung cancer and adenocarcinoma or cancer other than small cell lung cancer, neuroendocrine cancer, melanoma, thyroid cancer, multiple myeloma, lymphoma, and leukemia.

[0083] The machine learning engine 220 trains the probabilistic model 230 for each of the multiple disease states by fitting the probabilistic model 230 to sequence reads derived from each sample corresponding to each of the disease states. For example, in some embodiments, the probabilistic model may be trained for a specific type of cancer. In accordance with this embodiment, a cancer-specific probabilistic model may be trained for a first, second, third, etc. specific type of cancer and used to assess the cancer type (e.g., of an unknown test sample). For example, a lung cancer-specific probabilistic model is fitted using a set of sequence reads derived from one or more samples associated with lung cancer. As another example, a lung cancer-specific probabilistic model is fitted using a set of sequence reads derived from one or more samples associated with breast cancer. In some embodiments, a tissue-specific probabilistic model may be trained for a first, second, third, etc. tissue type and used to assess the tissue of origin of the disease state. For example, a first set of sequence reads derived from a first tissue type (e.g., a biopsy) can be fitted, and a second set of sequence reads derived from a second tissue type (e.g., a liver biopsy) can be used to fit a second tissue sample of a second origin probability model. Alternatively, in some embodiments, a cancer probability model is fitted using a set of sequence reads derived from one or more samples from subjects known to have cancer, and a non-cancer specific probability model is fitted using a set of sequence reads derived from one or more samples from normal or non-cancer subjects. As will be appreciated by those skilled in the art, some disease state probability models can be trained utilizing sequence reads derived from one or more samples taken from subjects having any one of a number of possible disease states.For example, in some embodiments, multiple sequence reads can be generated from 3, 4, 5, 6, 7, 8, 9, 10 or more reference samples, each from one or more subjects with different disease states (e.g., different types of conferences), and can be used to train 3, 4, 5, 6, 7, 8, 9, 10 or more probability models.

[0084] During training, the machine learning engine 220 may be trained on sequence reads indicative of disease states utilizing methylation information or methylation state vectors. In particular, the machine learning engine 220 determines an observed methylation fraction for each CpG site in the sequence read. The methylation fraction represents the fraction or fraction of base pairs that metabolize within the CpG site. The trained probability model 230 may be parameterized by the product of the methylation fraction. As previously described, any known probabilistic model for assigning probabilities to sequence reads from a sample may be used. For example, the probability model may be a binomial model, in which every site (e.g., CpG site) on a nucleic acid fragment is assigned a probability of methylation, or may be an independent site model, in which the methylation of each CpG is specified by a separate methylation probability with methylation at one site assumed to be independent of methylation at one or more other sites on the nucleic acid fragment.

[0085] In some embodiments, there is a Markov model, where the probability of methylation at each CpG site depends on the methylation state at the preceding CpG sites in the sequence read, or on the nuclear reaction molecule from which the sequence read is derived. See U.S. Patent Application No. 16 / 352,602 (entitled "Anomalous Fragment Detection and Classification," filed March 13, 2019).

[0086] In some embodiments, the probabilistic model 230 is a "mixture model" that is fitted using a mixture of components from the underlying models. For example, in some embodiments, the mixture components can be determined using a multiple independent site model, where the methylation (e.g., methylation rate) at each CpG site is assumed to be independent of the methylation at other CpG sites. Utilizing the independent site model, the probability assigned to a sequence read, or the molecule of nuclear material from which it is derived, is the product of the methylation probability at each CpG site at which the sequence read is methylated, minus one from the methylation probability at each CpG site at which the sequence read is unmethylated. According to this embodiment, the machine learning engine 220 determines the methylation rate of each of the mixture components. The mixture model is parameterized by the sum of the mixture components, each associated with a product of methylation rates. A probabilistic model Pr for n mixture components can be expressed as follows:

number

[0087] In some embodiments, the machine learning engine 220 configures the probabilistic model 230 with a parameter set that maximizes the log probability of all fragments derived from the disease state, subject to a regularization penalty applied to each methylation fraction with regularization strength r, using maximum likelihood estimation. We apply it to identify JPEG2024513563000007.jpg819. The maximum number of N total fragments can be expressed as:

number

[0088] The analysis system 200 applies the probability model 230 to calculate a value for each sequence read in the second set of sequence reads. These numerical values ​​are calculated based on at least the probability that the sequence read (and corresponding fragment) originated from a sample associated with the disease state in the probability model 230. The analysis system 200 can repeat this step for each of the different probability models 230. In some embodiments, the analysis system 200 calculates a value using a log-likelihood ratio R using the fitted probability model associated with a particular disease state. Specifically, the log-likelihood ratio can be calculated by looking at the pattern on the fragment with the probability Pr on the fragment to see if the sample is associated with a disease state to a healthy sample:

number

[0089] III.C. Classification In various embodiments, the analysis system 200 uses the features to generate a classifier 240. The classifier 240 is trained to predict the tissue of origin associated with a disease state for an input sequence read from a test sample of a test subject. The analysis system 200 can select a predetermined number (e.g., 1024) of top-ranked features for each disease state pair for training the classifier, for example, based on a mutual information operation or another computational measure. The predetermined number may be treated as a hyperparameter selected based on performance in cross-validation. The analysis system 200 can also select features from regions of the reference genome that are determined to be more informative in distinguishing the disease state pair. In various embodiments, the analysis system 200 retains the best performing tier for each region and each cancer type pair (including non-cancer as the negative type).

[0090] In one embodiment, analysis system 200 trains classifier 240 by inputting a set of training samples with feature vectors to classifier 240 and adjusting classification parameters such that the function of classifier 240 accurately associates training feature vectors with their corresponding labels. Analysis system 200 can group training samples into one or more sets of training samples for repeated batch training of classifier 240. After inputting all sets of training samples with training feature vectors and adjusting classification parameters, classifier 240 can be sufficiently trained to label test samples according to their feature vectors within a certain margin of error. Analysis system 200 can train classifier 240 according to several methods, for example, according to any one of LI-regularized logistic regression or L2-regularized logistic regression (e.g., with a logarithmic loss function), generalized linear model (GLM), random forest, multinomial logistic regression, multilayer perceptron, support vector machine, neural net, or other suitable machine learning techniques.

[0091] In various embodiments, analysis system 200 trains a multinomial logistic regression classifier on the training data of the folds and generates predictions for the retained data. Analysis system 200 trains one logistic regression for each combination of hyperparameters for each of the K folds. An example of a hyperparameter is the L2 penalty, a form of regularization applied to the weights of the logistic regression. Another exemplary hyperparameter is topK, the number of highly ranked regions to retain for each tissue type pair (including non-cancer). For example, if topK = 16, analysis system 200 retains the top 16 regions for each tissue type pair, as ranked by the mutual information procedure described herein. By following this procedure, analysis system 200 can generate predictions for each sample in the training set while ensuring that the classifier is not trained on the data on which the predictions are generated.

[0092] In various embodiments, for each set of hyperparameters, analysis system 200 evaluates performance on the cross-validated predictions of the full training set, and analysis system 200 selects the set of hyperparameters with the best performance to retrain on the full training set. Performance may be determined based on a log-loss metric. Analysis system 200 may calculate log-loss by taking the negative logarithm of the predictions for the correct labels for each sample, then summing across the samples. For example, if perfect prediction of the correct labels is 1.0, then log-loss will be 0 (lower is more accurate). To generate predictions for new samples, analysis system 200 may compute features using the methods described above, but limited to selected features (region / positive class combinations) under the selected topK values. Analysis system 200 may use the generated features to create predictions using the trained logistic regression model.

[0093] In various embodiments, analysis system 200 applies classifier 240 to predict the tissue of origin of a test sample, which tissue of origin is associated with one of the disease states. In some embodiments, classifier 240 can return predictions or probabilities for multiple disease states or tissues of origin. For example, classifier 240 can return a prediction that the test sample has a 65% chance of having breast cancer tissue of origin, a 25% chance of having lung cancer tissue of origin, and a 10% chance of having normal tissue of origin. Analysis system 200 can further process the predictions to generate a single disease state determination.

[0094] IV. Examples 6 illustrates experimental results for cancer signal localization ("CSL") according to one embodiment. The experimental results show the rate of cancer detection when analysis system 200 reports one cancer signal (i.e., the cancer signal with the highest probability score), two cancer signals (i.e., the cancer signals with the two highest probability scores), and three cancer signals (i.e., the cancer signals with the three highest probability scores). For many types of cancer included in the results, the rate of detection is higher when reporting two cancer signals instead of one.

[0095] The experimental results are based on a set of 450 samples. These samples were selected to reflect the expected distribution of cancer signal intensity in incident cases. Occult cancers are undiagnosed, preclinical cancers. Note that the subsample sizes for some of the cancer types above (anal and bladder & urinary tract) were small compared to the subsample sizes for other cancer types. Figure 6 further shows that when the first two CSLs are inaccurate, the third CSL provides little detectable benefit in that 5% of cases.

[0096] FIG. 7 shows experimental results of cancer signal localization based on conditional return, according to one embodiment. Here, analysis system 200 returns one cancer signal (top score cancer signal) if the cancer signal has a probability score of 90% or more of the positive cancer signal mass. Otherwise, analysis system 200 returns the top two cancer signals associated with a probability score of at most two. The bar graph shows the percentage of samples in each cancer type where one and two cancer signals were returned. For example, 70% of breast cancer samples returned one cancer signal and 30% returned two cancer signals. As another example, 50% of ovarian cancer samples returned one cancer signal and 50% returned two cancer signals.

[0097] In summary, the experimental results showed that the top CSL is accurate in about 90% of cases, while the second CSL is accurate half the time when the top CSL is inaccurate. The third CSL is in error about 80% of the time when the top two are inaccurate, and although it is better than chance, if reported, it may not help to facilitate a doctor or other healthcare provider to make a valid decision. Thus, in some embodiments, at most two localization attempts are provided before other methods of diagnosis / analysis are initiated (e.g., whole body imaging). In particular, the results show that lymphatic and myeloid CSLs are very reliably localized, with the majority of cancers localized to the first two CSLs.

[0098] Reporting top cancer signals with a determined probability threshold provides an improvement to the existing cancer diagnostic process, since the healthcare provider is presented with a filtered subset of one or more cancer signals. The healthcare provider can more accurately and quickly determine a diagnosis by not having to analyze a larger set of signals that may contain cancer signal localizations that may be inaccurate (e.g., false positives) or unreliable. As previously mentioned, tumor dropouts (e.g., early cancers) are difficult to localize because there are fewer informative fragments. Thus, conventional methods for non-invasive cancer prediction have difficulty dealing with false positive or unreliable cancer signals. Reducing this noise from the cancer signals reduces the complexity of the diagnostic process. Improved accuracy of cancer signal localization also reduces unnecessary treatment for individuals with false positive cancer diagnoses.

[0099] In various embodiments, filtering the filtered cancer signals with a probability threshold also improves computer performance since the method for cancer diagnosis is used in subsequent processing steps. For example, analysis system 200 uses the filtered (e.g., a subset of) cancer signals as input to a machine learning model that outputs a cancer prediction. As another example, analysis system 200 uses the filtered cancer signals as training data to train a machine learning model to, for example, determine the tissue of origin when the presence of risk is detected in a sample. In these examples, using the filtered cancer signals reduces the computational resources or processing time required by a computer implementing the machine learning model. The computer saves computation time by processing the top cancer signals (e.g., one or two signals of the subset determined by filtering with a probability threshold) instead of the unfiltered set of cancer signals. The unfiltered set of cancer signals may include more than 10 cancer signals, as evidenced by the various cancer types shown in FIG. 7. Furthermore, the unfiltered set of cancer signals grows as additional cancer signals are identified over time. In various embodiments, analysis system 200 processes cancer signals for many individuals. On a larger scale, improvements in computational capabilities are amplified due to the magnitude of data that the analysis system 200 must process to determine a cancer prognosis. More efficient and rapid determination of a cancer diagnosis can allow for early detection and treatment of cancer, which is crucial to an individual's health and prognosis. Achieving efficient and accurate prediction of cancer using non-invasive methods is even more beneficial, as these methods can make cancer diagnosis accessible to a larger population of individuals.

[0100] FIG. 8 shows experimental results of cancer signal localization from generated samples, according to one embodiment. The X-axis represents the first tissue of origin probability and the Y-axis represents the second tissue of origin probability. The occult cancer samples were not diagnosed with cancer at the time of blood collection from the individual, but the individual was later diagnosed with cancer. Thus, the strength of the cancer signal from the occult cancer samples is weaker compared to the signal from the already diagnosed cancer samples. The cancer signal intensity from the occult cancer samples also has a greater uncertainty regarding the accuracy of the tissue of origin localization.

[0101] 9 is a plot illustrating subsampling of cancer samples, according to one embodiment. The probability of true positive cancer detection for potential cancer samples 900 is low compared to the probability of true positive cancer detection for a set of cancer diagnosis samples 910. To more closely reflect the expected screening cancer signal strength of incident cancer samples 900, a set of diagnosed cancer samples 910 (e.g., 1876 samples) was downsampled into a subset of diagnosed cancer samples 920 (e.g., 450 samples). The subsampled true positives are:

number

number

number

[0102] 10A and 10B show detected cancer samples (true positives) that have been subsampled to match expected screening cancer signal intensity. Subsampling selects less stage iv and more stage i and ii cancer. Additionally, FIG. 10A and FIG. 10B show cancer signal intensity based on image stage, and as image stage progresses from i to iv, the proportion of detected true positives generally increases. However, in a two-person comparison, a sample from a first individual associated with stage i cancer may have a greater cancer signal intensity than a sample from a second individual associated with stage iv cancer.

[0103] 11A and 11B show cancer signal intensity by cancer type before and after subsampling according to some embodiments. For some cancer types (such as lung, colon, pancreatic and gallbladder cancer), the rate of true positive detection decreased after subsampling. For other cancer types (e.g., lymphatic neoplasm, breast, uterine and prostate), the rate of true positive detection increased after subsampling.

[0104] 12 illustrates cancer signal intensity before and after subsampling by cancer type and stage, according to some embodiments. As shown in FIG. 12, the most significant changes are decreases in stage iv lung, pancreatic gallbladder, and colorectal, and increases in stage i breast and stage i uterus.

[0105] 13A and 13B include bar graphs of distributions of CSL call probabilities, such as the percentage of CSL signals captured by the first, second, third, and fourth CSL calls, according to some embodiments. Specifically, FIG. 13A shows an overall graph of the distribution of cumulative and marginal cancer scores across the top four cancer signals. The cumulative bars reflect the sum of the cancer scores of the top 1, 2, 3, and / or 4 cancer signals. The bars are median, with lower and upper error limits of 10% and 90%.

[0106] Figure 13B shows a graph of the distribution of cumulative and marginal cancer scores across various cancer stages. The error bars on the bar graph indicate the 10th and 90th percentile cancer scores. As shown in Figures 13A to 13B, approximately 50-95% of the signal is captured in the upper CSL, with a median of approximately 90%, slightly smaller in earlier stages.

[0107] 14A and 14B include bar graphs of distribution of CSL call probabilities, such as the percentage of CSL signals captured by the first, second, third, and fourth CSL calls, by actual cancer type, according to some embodiments. As shown by the experimental results, samples of HPV-attributed cancers, such as anal and vulvar, have lower cancer scores compared to the cancer scores of other cancer types.

[0108] In some embodiments, the localization gene 250 returns multiple cancer tissue origins from a classification (e.g., HPV-driven cancer) even if the top cancer scores of individual types of cancer within the classification itself do not meet the criteria. For example, the top cancer signal for anal samples has a cancer score of 45% and the top cancer score for vulvar samples has a cancer score of 60%. Although neither cancer score meets the 90% probability threshold, the localization engine 250 can determine to return the anal and vulvar cancer signals if they are within the set of cancer signals with the greatest signal strength (e.g., the top three cancer signals). The localization engine 250 can condition the return of cancer signals based on other classifications, including multiple types of cancer (e.g., gastric and intestinal cancer).

[0109] Figures 15A, 15B, and 15C include bar graphs of median cancer scores divided into false positives and true positives, according to some embodiments. The magnitude of the false positive cancer scores shown in Figure 15A is lower than the magnitude of the true positive cancer scores shown in Figure 15B. Thus, the localization engine 250 more frequently returns two or more cancer signals for a false positive because the top cancer signal is less likely to meet the probability threshold (e.g., 90%).

[0110] FIG. 16 illustrates cumulative probability scores, according to some embodiments. The plot in FIG. 16 shows the number of cancer signals that would need to be returned by the localization engine 250 for their cumulative probability scores to reach the threshold probability. For example, close to 75% of the true positive samples require fewer than three cancer signals (i.e., one or two cancer signals are returned) to accumulate a threshold probability of 90%. In contrast, fewer than 50% of the false positive samples would require fewer than three cancer signals to be returned to accumulate a threshold probability of 90%. These results are consistent with the results shown in FIG. 15A-C, since the false positive cancer scores tend to be lower in magnitude than the true positive cancer scores.

[0111] 17A and 17B show the conditional accuracy of cancer signal localization according to some embodiments. As shown in FIG. 17B, the top cancer signal (i.e., the first label with the highest probability score) is accurate in about 90% of the samples. The second cancer signal (i.e., the second label) is accurate in about 50% of the samples when the top cancer signal is inaccurate. The third cancer signal (i.e., the third label) is accurate in about 20% of the samples when the top two cancer signals are inaccurate.

[0112] 18A and 18B show the conditional accuracy of cancer signal localization for solid and liquid sample types according to some embodiments. FIG. 19A and FIG. 19B show the conditional accuracy of cancer signal localization based on cancer stage according to some embodiments. The results in FIG. 18A show that the cancer signal localization of liquid samples is more accurate than that of solid samples. Compared to the solid samples, for more liquid samples, the localization engine 250 returned the top cancer signal (i.e., the 1st label) which is an accurate localization of the tissue of origin. In contrast, accurate localization of solid samples required more cancer signals to be returned (2nd, 3rd, 4th, 5th+ labels).

[0113] 20A and 20B show the cumulative accuracy of cancer signal localization according to some embodiments. The top cancer signal is accurate localization of the tissue of origin in about 90% of the samples. The cumulative accuracy increases to about 94%, 95%, and 96% for the second, third, and fourth cancer signal localizations, respectively.

[0114] 21A and 21B show false positive cancer signal localization according to some embodiments. Figures 22A and 22B show false positive cancer signal localization based on cancer type according to some embodiments. The results shown in Figures 21A-B indicate whether false positive tissues of origin localization are predicted to have mathematical (Brad) origin or solid (Routh) origin. False positives are predicted primarily for solid localization.

[0115] V. Cancer applications In some embodiments, the methods, analysis systems and / or classifiers of the present disclosure can be used to detect the presence (or absence) of cancer, monitor the progression or recurrence of cancer, monitor therapeutic response or effectiveness, determine the presence or presence or monitoring of minimal residual infection (MRD), or any combination thereof. In some embodiments, the analysis systems and / or classifiers can be used to identify the tissue of origin of the cancer. For example, the systems and / or classifiers can be used to identify the cancer as any of the following cancer types: head and neck cancer, liver / bile duct cancer, upper gastrointestinal cancer, pancreatic / gallbladder cancer; colorectal cancer, ovarian cancer, lung cancer, multiple myeloma, lymphoid neoplasm, melanoma, sarcoma, breast cancer, and uterine cancer. For example, as described herein, the classifiers can be used to generate a likelihood or probability score (e.g., 0% to 100%, or 0 to 100) that a sample feature vector is from a subject with cancer.

[0116] In some embodiments, the probability score is compared to a threshold probability to determine whether the subject has cancer. In other embodiments, the likelihood or probability score can be evaluated at various time points (e.g., before and after treatment) to monitor the progress of treatment or to monitor the effectiveness of treatment (e.g., the effectiveness of treatment). In yet another embodiment, the likelihood or probability score can be used to make or influence medical decisions (e.g., selecting a treatment, evaluating the effectiveness of treatment, etc.). For example, in one embodiment, if the likelihood or probability score exceeds a threshold, a physician can prescribe appropriate treatment. In some embodiments, a test report can be generated to provide patient test results, including, for example, a probability score that the patient has a disease state (e.g., cancer), a type of infection (e.g., a type of cancer), and / or an infected tissue of origin (e.g., a cancer tissue of origin).

[0117] VA Early Cancer Detection In some embodiments, the methods and / or classifiers of the present disclosure are used to detect the presence or absence of cancer in a subject suspected of having cancer. For example, a classifier (as described herein) can be used to determine the likelihood or probability score that a sample feature vector is from a subject with cancer.

[0118] In one embodiment, a probability score of 60 or greater may indicate that the subject has cancer. In yet other embodiments, a probability score of 65 or greater, 70 or greater, 75 or greater, 80 or greater, 85 or greater, 90 or greater, 95 or greater may indicate that the subject has cancer. In other embodiments, the probability score may indicate the severity of the infection. For example, a probability score of 80 may indicate a more severe form, or a later stage, of cancer, compared to a score below 80 (e.g., a score of 70). Similarly, an increase in the probability value over time (e.g., at a second, later time point) may indicate progression of the disease or a decrease in the probability value over time (e.g., at a second, later time point) may indicate successful treatment.

[0119] In another embodiment, a log ratio can be calculated for a test subject by taking the logarithm of the ratio of the probability of having cancer to the probability of not having cancer (i.e., 1 minus the probability of having cancer), as described herein. According to this embodiment, a cancer log-odds ratio greater than 1 can indicate that the subject has cancer. In yet other embodiments, a cancer log-odds ratio greater than 1.2, greater than 1.3, greater than 1.4, greater than 1.5, greater than 1.7, greater than 2, greater than 2.5, greater than 3, greater than 3.5, or greater than 4 indicated that the subject has cancer. In other embodiments, the cancer log-odds ratio can indicate the severity of infection. For example, a cancer log-odds ratio greater than 2 can indicate a more severe form, or later stage, compared to a score lower than 2 (e.g., a score of 1). Similarly, an increase in the log odds ratio of cancer over time (e.g., at a second, later time point) can indicate disease progression or a decrease in the log odds ratio of cancer over time (e.g., at a second, later time point) can indicate successful treatment.

[0120] According to aspects of the disclosure, the methods and systems of the present disclosure can be trained to detect or classify multiple cancer indications. For example, the methods, systems, and classifiers of the present disclosure can be used to detect the presence of one or more, two or more, three or more, five or more, or ten or more different types of cancer.

[0121] VB Cancer and Treatment Monitoring In certain embodiments, the first time point is pre-treatment (e.g., before treatment) and the second time point is post-treatment (e.g., after treatment to monitor the effectiveness of the treatment), and the method is utilized to monitor the effectiveness of the treatment. For example, if the second likelihood or probability score is decreased compared to the first likelihood or probability score, the treatment is deemed successful. However, if the second likelihood or probability score is increased compared to the first likelihood or probability score, the treatment is deemed unsuccessful. In other embodiments, both the first and second time points are pre-treatment (e.g., before an amputation operation or therapeutic intervention). In yet other embodiments, both the first and second time points are post-treatment (e.g., before an amputation operation or therapeutic intervention), and the method is utilized to monitor the effectiveness of the treatment or loss of effectiveness of the treatment. In yet other embodiments, cfDNA samples can be obtained and analyzed from a patient at first and second time points, e.g., to monitor the progression of the patient, to determine whether the patient is regenerating (e.g., post-treatment), or to monitor or detect persistence or recurrence of infection, or to monitor the effectiveness of a treatment (e.g., therapy).

[0122] One of ordinary skill in the art will readily appreciate that test samples can be obtained from a patient over any desired series of time points and analyzed according to the disclosed methods to monitor the patient's condition within the patient. In some embodiments, the first and second time points are separated by an amount of time ranging from 15 minutes to about 30 years, e.g., about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, or 24 hours, e.g., about 1, 2, 3, 4, 5, 10, 15, 20, 25 or 30 hours, e.g., about 1, 2, 3, 4, 5, 10, 15, 20, 25 or 30 days, e.g., about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 or 12 months, e.g., 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, In other embodiments, test samples are obtained from the patient at least once every three months, at least once every six months, at least once a year, at least once every two years, at least once every three years, at least once every four years, or at least once every five years or more.

[0123] VC treatment In yet another embodiment, information obtained from any of the methods described herein (e.g., likelihood or probability scores) can be used to make or influence medical decisions (e.g., medical diagnoses, selection of treatments, evaluation of treatment efficacy, etc.). For example, in one embodiment, if the likelihood or probability score exceeds a threshold, a physician can prescribe appropriate treatment (e.g., treatment, irradiation, therapy, and / or therapy). In some embodiments, information such as the likelihood or probability score can be provided as a readout to a physician or to a subject.

[0124] A classifier, as described herein, can be used to determine a likelihood or probability score that a sample feature vector is from a subject with cancer. In one embodiment, if the likelihood or probability exceeds a threshold, an appropriate treatment (e.g., amputation surgery or treatment) is prescribed. For example, in one embodiment, if the likelihood or probability score is 60 or greater, one or more appropriate treatments are prescribed. In another embodiment, if the likelihood or probability score is 65 or greater, 70 or greater, 75 or greater, 80 or greater, 85 or greater, 90 or greater, 95 or greater, one or more appropriate treatments are prescribed. In other embodiments, the log odds ratio of cancer can indicate the effectiveness of a treatment. For example, an increase in the log odds ratio of cancer over time (e.g., after treatment) can indicate that the treatment was ineffective. Similarly, a decrease in the log odds ratio of cancer over time (e.g., after 1 second of treatment) can indicate successful treatment. In another embodiment, if the log odds ratio of cancer 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.

[0125] In some embodiments, the treatment is one or more therapeutic agents selected from the group including chemotherapeutic agents, targeted cancer therapeutic agents, differential therapy agents, hormonal therapy agents, and immunotherapy agents. For example, the treatment can be one or more chemotherapeutic agents selected from the group including alkylating agents, antimetabolites, anthracyclines, antitumor antibiotics, cytoskeleton disrupting agents (taxanes), 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 therapeutic agents selected from the group including transduction inhibitors (e.g., tyrosine kinase and growth factor receptor inhibitors), histone deactivator (HDAC) inhibitors, retinoreceptor agonists, proteosome inhibitors, angiogenesis inhibitors, and monoclonals. In some embodiments, the treatment is one or more differentiated therapeutic agents, including retinoids, such as tretinoin, alitretinoin, and bexarotene. In some embodiments, the treatment is one or more hormonal therapy agents selected from the group including antiestrogens, 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 including rituximab (RITUXAN) and alemtuzumab (CAMPATH), BCG, non-specific immunotherapeutics and adjuvants such as loking-in-2 (IL-2), and feron alpha, immunomodulatory agents such as thalidomide and lenalidomide (REVLIMID). It is within the capabilities of a skilled physician or oncologist to select the appropriate cancer treatment agent based on characteristics such as tumor type, stage of cancer, previous exposure to cancer treatments or therapeutic agents, and other characteristics of the cancer.

[0126] VI. Additional Considerations The foregoing description of the disclosed embodiments has been presented for purposes of illustration and is not intended to be exhaustive or to limit the invention to the precise form disclosed. Those skilled in the relevant art will recognize that many modifications and variations are possible in light of the above disclosure.

[0127] Some portions of this description describe embodiments of the disclosure in terms of algorithms and symbolic representations of operations on information. These algorithmic descriptions and representations are typically used by those skilled in the data processing arts to effectively convey the substance of their work to others skilled in the art. These operations, while described functionally, computationally, or logically, will be understood to be implemented by computer programs or equivalent electrical circuits, microcode, or the like. Moreover, without loss of generality, it has proven convenient at times to refer to arrangements of these operations as modules. The described operations and their associated modules may be implemented in software, firmware, hardware, or any combination thereof.

[0128] Any of the steps, operations, or processes described herein may be performed or implemented in one or more hardware or software units, alone or in combination with other devices. In some embodiments, the software units are embodied in a computer program product that includes a computer-readable, non-transitory medium containing computer program code, which may be executed by a computer processor to perform any or all of the described steps, operations, or processes.

[0129] Embodiments may also relate to products produced by the computational processes described herein, such products may include information stored on a non-transitory, tangible, computer-readable storage medium, and may include the results of computational processing, which may include any of the embodiments of the computer program products or other data combinations described herein.

[0130] Finally, the language used herein has been selected primarily for ease of reading and instructional purposes, and not to describe or circumscribe the subject matter of the invention. Accordingly, it is intended that the scope of the invention be limited not by this detailed description, but rather by the claims that issue in an application based thereon. Accordingly, the disclosure of embodiments herein is intended to be illustrative, but not limiting, of the scope of the invention, which is set forth in the following claims.

Claims

1. 1. A method for cancer diagnosis comprising: receiving a first plurality of cancer signals of a first sample of a first individual, each of the first plurality of cancer signals indicative of a probability that the first sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a maximum probability among the first plurality of cancer signals; in response to determining that the first cancer signal satisfies a criterion, associating the first sample with a disease state corresponding to the first cancer signal; providing the disease status corresponding to the first cancer signal associated with the first sample for display on a client device to determine a first diagnosis for the first individual; receiving a second plurality of cancer signals of a second sample of a second individual, each of the second plurality of cancer signals indicative of a probability that the second sample is associated with a different one of the plurality of disease states; determining a second cancer signal having a maximum probability among the second plurality of cancer signals; in response to determining that the second cancer signal does not meet the criterion, associating the second sample with a subset of the plurality of disease states that corresponds to a subset of the second plurality of cancer signals that includes at least the second cancer signal; providing the subset of the plurality of disease states corresponding to the subset of the second plurality of cancer signals associated with the second sample for display on the client device to determine a second diagnosis for the second individual; The method includes:

2. 10. The method of claim 1 , determining a third cancer signal having a second highest probability among the second plurality of cancer signals, wherein the subset of the second plurality of cancer signals further includes the third cancer signal.

3. 10. The method of claim 1 , the criterion is a probability threshold, The method, wherein determining that the first cancer signal meets the criteria includes determining that the maximum probability of the first cancer signal is greater than the probability threshold.

4. The method of claim 3 , wherein the probability threshold is at least 90%.

5. 10. The method of claim 1 , determining said criterion based on accuracy and false positives of cancer signal probability.

6. 10. The method of claim 1 , determining said criterion based on a current residual risk of cancer associated with the sample.

7. 10. The method of claim 1 , determining a subset of n cancer signals of the first plurality of cancer signals having n highest probabilities among the first plurality of cancer signals; in response to determining that at least a threshold number of the subset of the first plurality of cancer signals is associated with a classification of disease states, associating the first sample with each disease state of the classification of disease states; The method includes:

8. 8. The method of claim 7, wherein the disease state classification is human papillomavirus (HPV) cancer.

9. 8. The method of claim 7, wherein the classification of disease states includes gastric cancer and intestinal cancer.

10. 10. The method of claim 1, wherein the plurality of disease conditions comprises a non-cancerous condition.

11. 10. The method of claim 1, wherein the plurality of disease states comprises: The method includes one or more types of cancer selected from the group including anal cancer, breast cancer, uterine cancer, cervical cancer, ovarian cancer, bladder cancer, urothelial carcinoma of the renal pelvis and ureter, renal cancer other than urothelial carcinoma, prostate cancer, anorectal cancer, colorectal cancer, squamous cell carcinoma of the esophagus, esophageal cancer other than squamous cell carcinoma, gastric cancer, hepatobiliary cancer arising from hepatocytes, hepatobiliary cancer arising from cells other than hepatocytes, pancreatic cancer, head and neck cancer associated with human papillomavirus, head and neck cancer not associated with human papillomavirus, lung adenocarcinoma, small cell lung cancer, lung squamous cell carcinoma, lung cancer other than adenocarcinoma or small cell lung cancer, neuroendocrine cancer, melanoma, thyroid cancer, sarcoma, multiple myeloma, lymphoma, leukemia, kidney cancer, liver cancer, cholangiocarcinoma, plasma cell neoplasm cancer, upper gastrointestinal tract cancer, vulvar cancer, lung neuroendocrine tumors, and other high-grade neuroendocrine tumors.

12. 10. The method of claim 1 , providing, for display on the client device, a graphical comparison of each disease state corresponding to the subset of the plurality of disease states associated with the second sample.

13. 13. The method of claim 12, wherein the graphical comparison is a bar plot based on the probabilities of the second plurality of cancer signals.

14. 1. A method for cancer signal localization, comprising: receiving a plurality of cancer signals of a sample, each of the plurality of cancer signals indicative of a probability that the sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a highest probability among the plurality of cancer signals; in response to determining that the first cancer signal satisfies a criterion, associating the sample with a first disease state corresponding to the first cancer signal; responsive to determining that the first cancer signal does not meet the criterion, determining a second cancer signal having a second greatest probability among the plurality of cancer signals; Associating the sample with the first disease state corresponding to the first cancer signal and with a second disease state corresponding to the second cancer signal; A method comprising:

15. 15. The method of claim 14, responsive to said determining that the first cancer signal satisfies the criterion, providing said first cancer signal as an input to a machine learning model to determine a prediction of cancer in the sample; responsive to the determination that the first cancer signal does not meet the criterion, providing the first cancer signal and the second cancer signal as inputs to the machine learning model to determine the prediction of cancer in the sample; The method includes:

16. 15. The method of claim 14, responsive to determining that the first cancer signal meets the criterion, creating a first training set including associations between the samples and the first disease state corresponding to the first cancer signal to train a machine learning model for cancer signal localization; responsive to the determination that the first cancer signal does not meet the criterion, creating a second training set including associations of the samples with the first disease state corresponding to the first cancer signal and the second disease state corresponding to the second cancer signal to train the machine learning model; The method includes:

17. 15. The method of claim 14, wherein the criterion is a probability threshold, and the determining that the first cancer signal satisfies the criterion comprises determining that a maximum probability of the first cancer signal is higher than the probability threshold.

18. 15. The method of claim 14, determining said criterion based on accuracy and false positives of a cancer signal probability.

19. 15. The method of claim 14, determining said criterion based on a current residual risk of cancer associated with the sample.

20. 15. The method of claim 14, determining a subset of n cancer signals of the plurality of cancer signals having n highest probabilities among the plurality of cancer signals; in response to determining that at least one threshold number of the subset of the plurality of cancer signals is associated with a classification of disease states, associating the sample with each disease state of the classification of disease states; The method includes:

21. 21. The method of claim 20, wherein the disease state classification is human papillomavirus (HPV) cancer.

22. 21. The method of claim 20, wherein the classification of disease states includes gastric cancer and intestinal cancer.

23. 15. The method of claim 14, wherein the plurality of disease conditions comprises a non-cancerous condition.

24. 15. The method of claim 14, wherein the plurality of disease states comprises: The method includes one or more types of cancer selected from the group including anal cancer, breast cancer, uterine cancer, cervical cancer, ovarian cancer, bladder cancer, urothelial carcinoma of the renal pelvis and ureter, renal cancer other than urothelial carcinoma, prostate cancer, anorectal cancer, colorectal cancer, squamous cell carcinoma of the esophagus, esophageal cancer other than squamous cell carcinoma, gastric cancer, hepatobiliary cancer arising from hepatocytes, hepatobiliary cancer arising from cells other than hepatocytes, pancreatic cancer, head and neck cancer associated with human papillomavirus, head and neck cancer not associated with human papillomavirus, lung adenocarcinoma, small cell lung cancer, lung squamous cell carcinoma, lung cancer other than adenocarcinoma or small cell lung cancer, neuroendocrine cancer, melanoma, thyroid cancer, sarcoma, multiple myeloma, lymphoma, leukemia, kidney cancer, liver cancer, cholangiocarcinoma, plasma cell neoplasm cancer, upper gastrointestinal tract cancer, vulvar cancer, lung neuroendocrine tumors, and other high-grade neuroendocrine tumors.

25. 15. The method of claim 14, providing the disease state corresponding to the first cancer signal associated with a first sample for display on a client device to determine a diagnosis.

26. 15. The method of claim 14, providing a graphical comparison of each disease state corresponding to the subset of the plurality of disease states associated with a second sample for display on a client device to determine a diagnosis.

27. 27. The method of claim 26, wherein the graphical comparison is a bar plot based on the probabilities of the second plurality of cancer signals.

28. 1. A method for cancer signal localization, comprising: receiving a plurality of cancer signals of a sample, each of the plurality of cancer signals indicative of a probability that the sample is associated with a different one of a plurality of disease states; determining a first conditional probability that a first cancer signal of the plurality of cancer signals is a true positive given that remaining cancer signals of the plurality of cancer signals are inaccurate; in response to determining that the first conditional probability satisfies a criterion, associating the sample with at least one disease state corresponding to the first cancer signal; determining a subset of the plurality of cancer signals excluding the first cancer signal; determining a second conditional probability that a second cancer signal of the subset of the plurality of cancer signals is a true positive given that a remaining cancer signal of the subset of the plurality of cancer signals is inaccurate; in response to determining that the second conditional probability satisfies the criterion, associating the sample with at least one disease state corresponding to the second cancer signal; The method includes:

29. 1. A system including a computer processor and a memory, the memory storing computer program instructions that, when executed by the computer processor, cause the processor to perform steps including: receiving a first plurality of cancer signals of a first sample of a first individual, each of the first plurality of cancer signals indicative of a probability that the first sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a maximum probability among the first plurality of cancer signals; in response to determining that the first cancer signal satisfies a criterion, associating the first sample with a disease state corresponding to the first cancer signal; providing the disease status corresponding to the first cancer signal associated with the first sample for display on a client device to determine a first diagnosis for the first individual; receiving a second plurality of cancer signals of a second sample of a second individual, each of the second plurality of cancer signals indicative of a probability that the second sample is associated with a different one of the plurality of disease states; determining a second cancer signal having a maximum probability among the second plurality of cancer signals; in response to determining that the second cancer signal does not meet the criterion, associating the second sample with a subset of the plurality of disease states that corresponds to a subset of the second plurality of cancer signals that includes at least the second cancer signal; providing the subset of the plurality of disease states corresponding to the subset of the second plurality of cancer signals associated with the second sample for display on the client device to determine a second diagnosis for the second individual; Including, the system.

30. A non-transitory computer-readable medium containing instructions that, when executed by one or more processors, cause the one or more processors to perform steps, the steps including: receiving a first plurality of cancer signals of a first sample of a first individual, each of the first plurality of cancer signals indicative of a probability that the first sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a maximum probability among the first plurality of cancer signals; in response to determining that the first cancer signal satisfies a criterion, associating the first sample with a disease state corresponding to the first cancer signal; providing the disease status corresponding to the first cancer signal associated with the first sample for display on a client device to determine a first diagnosis for the first individual; receiving a second plurality of cancer signals of a second sample of a second individual, each of the second plurality of cancer signals indicative of a probability that the second sample is associated with a different one of the plurality of disease states; determining a second cancer signal having a maximum probability among the second plurality of cancer signals; in response to determining that the second cancer signal does not meet the criterion, associating the second sample with a subset of the plurality of disease states that corresponds to a subset of the second plurality of cancer signals that includes at least the second cancer signal; providing the subset of the plurality of disease states corresponding to the subset of the second plurality of cancer signals associated with the second sample for display on the client device to determine a second diagnosis for the second individual; Including, the system.

31. 1. A system including a computer processor and a memory, the memory storing computer program instructions that, when executed by the computer processor, cause the processor to perform steps including: receiving a plurality of cancer signals of a sample, each of the plurality of cancer signals indicative of a probability that the sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a highest probability among the plurality of cancer signals; in response to determining that the first cancer signal satisfies a criterion, associating the sample with a first disease state corresponding to the first cancer signal; responsive to determining that the first cancer signal does not meet the criterion, determining a second cancer signal having a second greatest probability among the plurality of cancer signals; Associating the sample with the first disease state corresponding to the first cancer signal and with a second disease state corresponding to the second cancer signal; Including, the system.

32. A non-transitory computer-readable medium containing instructions that, when executed by one or more processors, cause the one or more processors to perform steps, the steps including: receiving a plurality of cancer signals of a sample, each of the plurality of cancer signals indicative of a probability that the sample is associated with a different one of a plurality of disease states; determining a first cancer signal having a highest probability among the plurality of cancer signals; in response to determining that the first cancer signal satisfies a criterion, associating the sample with a first disease state corresponding to the first cancer signal; responsive to determining that the first cancer signal does not meet the criterion, determining a second cancer signal having a second greatest probability among the plurality of cancer signals; Associating the sample with the first disease state corresponding to the first cancer signal and with a second disease state corresponding to the second cancer signal; A non-transitory computer readable storage medium comprising:

33. 1. A system including a computer processor and a memory, the memory storing computer program instructions that, when executed by the computer processor, cause the processor to perform steps including: receiving a plurality of cancer signals of a sample, each of the plurality of cancer signals indicative of a probability that the sample is associated with a different one of a plurality of disease states; determining a first conditional probability that a first cancer signal of the plurality of cancer signals is a true positive given that remaining cancer signals of the plurality of cancer signals are inaccurate; in response to determining that the first conditional probability satisfies a criterion, associating the sample with at least one disease state corresponding to the first cancer signal; determining a subset of the plurality of cancer signals excluding the first cancer signal; determining a second conditional probability that a second cancer signal of the subset of the plurality of cancer signals is a true positive given that a remaining cancer signal of the subset of the plurality of cancer signals is inaccurate; in response to determining that the second conditional probability satisfies the criterion, associating the sample with at least one disease state corresponding to the second cancer signal; Including, the system.

34. A non-transitory computer-readable medium containing instructions that, when executed by one or more processors, cause the one or more processors to perform steps, the steps including: receiving a plurality of cancer signals of a sample, each of the plurality of cancer signals indicative of a probability that the sample is associated with a different one of a plurality of disease states; determining a conditional probability that a first cancer signal of the plurality of cancer signals is a true positive given that remaining cancer signals of the plurality of cancer signals are inaccurate; in response to determining that the first conditional probability satisfies a criterion, associating the sample with at least one disease state corresponding to the first cancer signal; determining a subset of the plurality of cancer signals excluding the first cancer signal; determining a second conditional probability that a second cancer signal of the subset of the plurality of cancer signals is a true positive given that a remaining cancer signal of the subset of the plurality of cancer signals is inaccurate; in response to determining that the second conditional probability satisfies the criterion, associating the sample with at least one disease state corresponding to the second cancer signal; A non-transitory computer readable storage medium comprising:

Images