Methods and systems for discriminating between clonal hematopoiesis-derived and tumor-derived variants

Abstract

This disclosure describes novel methods and systems for discriminating between tumor-origin variants and clonal hematopoiesis (CH)-derived variants across solid tumors and NGS sequencing platforms. The disclosed methods and systems utilize a machine learning model that integrates fragment-level, vanant-level, and patient-level features, allowing for predicting tumor vs. CH variant origin using plasma-only sequencing without additional white blood cell (WBC) or tumor sequencing. The ability to identify bona fide tumor- derived variants in plasma-only sequencing fills a critical need in the clinical implementation of liquid biopsy-guided cancer therapy by reducing misinterpretation due to CH contamination. The methods and systems as disclosed have improved precision and reliability' of prediction utilizing circulating-tumor DNA (ctDNA) for clinical prognosis purposes.

Description

[0001]Docket: 348358.17201 METHODS AND SYSTEMS FOR DISCRIMINATING BETWEEN CLONAL HEMATOPOIESIS-DERIVED AND TUMOR-DERIVED VARIANTS The present application clams the benefit of U.S. provisional application no. 63 / 548,318 filed November 13, 2023, which is incorporated by reference herein in its entirety. GOVERNMENT REIGHTS This invention was made with government support under grant CA121113 awarded by the National Institutes of Health. The government has certain rights in the invention. FIELD This invention relates generally to methods for discriminating between clonal hematopoiesis (CH)-derived and tumor-derived variants in plasma cell-free DNA (cfDNA) of a subject and methods for identifying a CH-derived variant in plasma cfDNA of a subject. BACKGROUND Analysis of cell-free DNA (cfDNA) from liquid biopsy samples has emerged as an efficient, non-invasive method for multiple purposes, including the early detection of cancers, tracking tumor heterogeneity and evolutionary dynamics to monitor disease progression, and detecting minimal residual disease to assess response to therapy. cfDNA consists of short, double-stranded DNA fragments released into circulation from cells undergoing apoptosis or necrosis. In healthy individuals, the majority of cfDNA is derived from hematopoietic cells, with a small percentage derived from other organs. In cancer patients, a very small fraction of the cfDNA is also tumor-derived (ctDNA). Due to the high percentage of hematopoietic and typically low concentrations of ctDNA within the population of cfDNA, a major challenge to analyzing liquid biopsy samples is to accurately determine the origin of the variants detected and avoid misclassification of clonal hematopoiesis (CH)-derived variants as tumor-derived. Clonal hematopoiesis of indeterminate potential (CHIP) is a phenomenon when a hematopoietic cell acquires a benign genetic mutation when dividing. These mutations often occur in leukemia-associated driver genes and may confer a fitness advantage to the cell, resulting in its disproportionate propagation and accumulation of the CH mutation. Since the 164810982.1 Docket: 348358.17201 majority of cfDNA is derived from hematopoietic cells, these CHIP-derived mutations may confound the interpretation of the cfDNA assay. Depending on the age and health condition of the patient, more than half of the mutations found in plasma cfDNA can also be detected in WBC sequencing. Furthermore, these CHIP mutations frequently occur in cancer-associated genes such as TP53, KRAS, or JAK2. The misidentification of CHIP mutations as tumor-derived can significantly impair the use of liquid biopsies for clinical prognoses, such as in cancer screenings, selecting treatment options, or detection of minimal residual disease and monitoring treatment response. As mentioned above, healthy individuals derive the majority of their cfDNA from hematopoietic cells. If the patient is in early stages of cancer, the amount of ctDNA will be very small, and some CH mutations may be misclassified as tumor mutations, leading to an unreliable diagnosis. Similarly, the consistent detection of CH mutations pre- and post-treatment may be incorrectly interpreted as the presence of minimal residual disease and ineffective treatment response. A recent study in patients with resectable gastric cancer has shown that without the removal of CH- derived variants, ctDNA analysis could not predict the risk of recurrence or survival post- operation. In order for ctDNA analysis to be meaningful, it is crucial to accurately differentiate between tumor and CH-derived variants. One way to circumvent this issue is to use a tumor-informed approach, which relies on the genotyping of the tumor tissue to correctly identify tumor-derived variants. However, this method reintroduces the issue of needing an adequate tissue sample, which may be difficult to obtain and may not capture the genetic heterogeneity of the cancer. Another approach utilizes parallel sequencing of patient-matched white blood cells (WBCs) to filter out CH variants from the cfDNA results, and can achieve comparable results to tumor-informed methods. Although this approach bypasses the requirement for a tumor biopsy, extra time and costs are still needed to extract, sequence, and analyze the WBC DNA. Furthermore, most commercial targeted sequencing panels do not include matched WBC sequencing, which may result in a high margin of error when interpreting cfDNA data. As a result, there is a critical need to devise a strategy capable of differentiating ctDNA variants from biological background noise independent of tumor tissue and WBC sequencing 164810982.1 Docket: 348358.17201 analysis. This will not only improve the use of cfDNA for clinical prognosis, but also will allow retrospective studies to be performed on plasma samples without matched WBC sequencing. SUMMARY This disclosure addresses the need mentioned above in a number of aspects. In one aspect, this disclosure provides a method of discriminating between clonal hematopoiesis- derived and tumor-derived variants in plasma cell-free DNA (cfDNA) of a subject. In some embodiments, the method comprises: (i) determining values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; (ii) determining a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and (iii) identifying a variant in the plasma cfDNA as a clonal hematopoiesis-derived variant if the classification score is equal to or greater than a threshold value. In another aspect, this disclosure also provides a method of identifying a clonal hematopoiesis-derived variant in plasma cell-free DNA (cfDNA) of a subject. In some embodiments, the method comprises: (a) determining values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; (b) determining a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and (c) identifying a variant in the plasma cfDNA as a clonal hematopoiesis-derived variant if the classification score is equal to or greater than a threshold value. In some embodiments, the fragment level features comprise differences in fragment length and cut points and endpoint motifs between mutant and wild type fragments at a locus. In some embodiments, the variant level features comprise variant allele frequency (VAF), scaled VAF, frequency of a variant in hematologic versus solid tumor malignancies in the Catalogue Of Somatic Mutations In Cancer (COSMIC), single base pair substitution, and estimate of driver function. In some embodiments, the patient level features comprise age. 164810982.1 Docket: 348358.17201 In some embodiments, the set of features comprise gene heme fraction, length cluster mutant, length MWU, length max delta S, and age. In some embodiments, the set of features further comprise endpoint motif down, endpoint location, plasma scaled VAF, length mean delta S, length cluster weight, variant COSMIC count, indel complex, mutant reads, or a combination thereof. In some embodiments, the classifier comprises a supervised or unsupervised Machine Learning or Deep Learning algorithm, Logistic Regression, Naive Bayes, Support Vector Machine, Decision Tree, Random Forest, Gradient Boosting, Regularizing Gradient Boosting, K- Nearest Neighbors, a continuous regression approach, Ridge Regression, Kernel Ridge Regression, Support Vector Regression, deep learning approach, Neural Networks, Convolutional Neural Network (CNNs), Recurrent Neural Networks (RNNs), Long Short Term Memory Networks (LSTMs), Generative Models, Generative Adversarial Networks (GANs), Deep Belief Networks (DBNs), Feedforward Neural Networks, Autoencoders, Radial Basis Function Networks (RBFNs), Multilayer Perceptrons (MLPs), Stochastic Neural Networks, or any combination thereof. In some embodiments, the classifier comprises an XGBoost algorithm. In some embodiments, the threshold value is about 0.5 (e.g., 0.5). In some embodiments, the method is capable of identifying the clonal hematopoiesis- derived variant with an accuracy of at least 80%. In some embodiments, the step of determining values of the set of features comprises extracting reads covering each variant locus from plasma-targeted next-generation sequencing (NGS) data. In some embodiments, the step of determining values of the set of features comprises calculating fragment summary statistics associated with fragment length, cut points, and endpoint motifs. In some embodiments, the step of determining values of the set of features comprises annotating variants using OpenCravat-COSMIC variant, COSMIC gene, CHASMMplus, and Cancer Gene Census (CGC). In some embodiments, the step of determining values of the set of features comprises calculating gene and variant heme fraction and variant counts using COSMIC counts. 164810982.1 Docket: 348358.17201 In some embodiments, the step of determining values of the set of features comprises determining driver status using CHASMMplus score, sequence ontology, and Cancer Gene Census (CGC) driver class. In some embodiments, the step of determining values of the set of features comprises calculating scaled plasma variant allele frequency (VAF). In some embodiments, the subject has no patient-matched white blood cell (WBC) sequencing data available. In some embodiments, the subject has cancer. In some embodiments, the subject has a solid tumor. In some embodiments, the plasma cfDNA comprises tumor- circulating DNA (ctDNA). In some embodiments, the method comprises obtaining from a sample from the subject comprising the plasma cfDNA. In some embodiments, the sample comprises a plasma sample, a blood sample, or a tissue sample. In another aspect, this disclosure also provides a system of discriminating between clonal hematopoiesis-derived and tumor-derived variants in plasma cell-free DNA (cfDNA) of a subject. In some embodiments, the system comprises one or more processors configured to: (a) determine values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; (b) determine a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and (c) identify a variant in the plasma cfDNA as a clonal hematopoiesis-derived variant if the classification score is equal to or greater than a threshold value. In yet another aspect, this disclosure further provides a system of identifying a clonal hematopoiesis-derived variant in plasma cell-free DNA (cfDNA) of a subject. In some embodiments, the system comprises one or more processors configured to: (a) determine values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; (b) determine a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and (c) identify a variant in the plasma cfDNA as a clonal hematopoiesis-derived variant if the classification score is equal to or greater than a threshold value. 164810982.1 Docket: 348358.17201 The foregoing summary is not intended to define every aspect of the disclosure, and additional aspects are described in other sections, such as the following detailed description. The entire document is intended to be related as a unified disclosure, and it should be understood that all combinations of features described herein are contemplated, even if the combinations of features are not found together in the same sentence, or paragraph, or section of this document. Other features and advantages of the invention will become apparent from the following detailed description. It should be understood, however, that the detailed description and the specific examples, while indicating specific embodiments of the disclosure, are given by way of illustration only, because various changes and modifications within the spirit and scope of the disclosure will become apparent to those skilled in the art from this detailed description. BRIEF DESCRIPTION OF THE DRAWINGS The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee. Figures 1a and 1b show example processes of variant origin prediction. Figure 2. Development of a machine learning algorithm to predict CH versus tumor origin for variants detected in plasma. Schematic illustrating algorithm development. For variants detected in plasma, fragment features are extracted from sequencing data and combined with variant features and patient features into a machine learning model that predicts if each variant is tumor or CH derived. Model performance is assessed by comparison to the reference plasma variant origin which is determined by comparison to matched WBC and tumor tissue sequencing. Figure 3 (includes Figures 3a-3e): Assessment of fragmentomic features. a-c. Illustrative examples of fragmentomic features seen between mutant and wild-type reads from CH-origin versus tumor origin variants. a. CH-origin variant with overlapping fragment length densities (top panel) and fragment endpoint locations (bottom panel) between mutant and wild- type reads. b. Tumor-derived plasma variant with fragment length density right-shifted (shorter) for mutant compared to wild-type fragments (top panel), and with the fragment endpoints relative to mutation location (set at 0) closer together in mutant versus wild-type fragments (bottom panel). c. Tumor-derived plasma variant that is not distinguishable by fragment length 164810982.1 Docket: 348358.17201 alone. Fragment length density is the same for both mutant and wild-type fragments (top panel), while locations of the relative fragment endpoints are shifted in mutant versus wild-type fragments (bottom panel). d. The serial cohort used to assess fragmentomic features over time. Each column represents a patient-variant, with points on the y-axis depicting VAF at baseline (black) and at follow-up (gray) time points. e. Pearson’s product moment correlation of fragmentomic feature values at baseline and follow-up timepoints. Features that were correlated over time with p-value <0.05 were subsequently used for development of the machine learning model. Figure 4 (includes Figures 4a-4c): Description of the training cohort. a. Training cohort of 426 variants identified in 225 patients with stage I-IV breast, colorectal, esophageal, NSCLC, ovarian and SCLC. Each column represents a patient, with plasma VAF along the y- axis and points color-coded by reference variant origin (tumor vs WBC). b. Plasma variant allele frequencies (VAF) for tumor-origin and WBC-origin variants. c. Ability of plasma to distinguish variant origin in the training cohort. AUC= area under the curve, WBC = white blood cell Figure 5 (includes Figures 5a-5c): Model performance in the training cohort. a. Heatmap with rows depicting the scaled values of features inputted into the machine learning algorithm and each column representing a single variant identified in a given patient. The second from the bottom row shows the reference variant origin where red is tumor-origin and blue is CH-origin. The bottom row shows the variant origin predicted by the machine learning algorithm. b. Receiver-operator curve (ROC) for the machine learning algorithm to predict variant origin from 10-fold cross-validation repeated 10 times. c. Feature importance in the machine learning model. AUC=area under ROC. Figure 6 (includes Figures 6a-6d). Machine learning model performance in an independent validation cohort. a. The independent validation cohort of 1,412 variants (507 tumor-derived, 911 CH-derived) found in 114 patients with metastatic breast, non-small cell lung, and prostate cancer. Each column represents a patient with variant allele frequency for variants detected in plasma in the y-axis. Plasma variants that were determined by matched sequencing to have reference origin of tumor are red, versus WBC blue. b. For the validation cohort, heatmap with rows depicting the scaled values of features inputted into the fixed machine 164810982.1 Docket: 348358.17201 learning algorithm. Each column represents a single variant with reference and predictive variant origin in the second from bottom and bottom rows respectively. c. Distribution of actual and predicted variant origins. d. Overall performance of the fixed machine learning model in the independent validation cohort across range of possible cutoff values. AUC = area under the curve Figure 7(includes Figures 7a-7c): Model performance in cancer-type and gene based subsets of independent validation cohort. a. Distribution of actual and predicted variant origin according to cancer type in the validation cohort. b. Accuracy of the model origin prediction versus null hypothesis that all variants are tumor-derived for gene subsets including clinically relevant mixed origin genes, TP53, and excluding genes commonly associated with CH. c. Distribution of location and origin of alterations in TP53. Points filled in red are tumor-derived by reference origin and filled in blue are WBC-derived by reference origin. Outlines are red if predicted tumor-derived by the machine learning model and blue if predicted WBC-derived by reference origin. Figure 8 shows an example computing system for implementing the disclosed methods. Figure 9: Correlation between fragmentomic statistics at baseline and follow-up time points in the serial cohort. Pearson correlation for different fragmentomic statistics between baseline and follow-up time points as determined using the serial cohort. Figure 10 (includes Figures 10a-10c): Description of training cohort a. Distribution of cancer types, stage, and patient age in the training cohort. b. Distribution of WBC only versus WBC and tumor matched sequencing and determined reference variant origin. c. Distribution of variants included in training cohort by gene and reference variant origin. Figure 11: Performance of individual features in training cohort Receiver-operator curves and AUC values individual features for determining CH versus tumor-origin. Figure 12: Model performance in training cohort by number of mutant reads Receiver operator curve and AUC values for subsets of training cohort as determined by the number of mutant reads. Figure 13 (includes Figures 13a-13c): Description of Independent validation cohort. a. Distribution of plasma variant allele frequency for tumor-origin and WBC-origin variants in the 164810982.1 Docket: 348358.17201 validation cohort. b. Distribution of variant origin in the validation cohort. biopsy_matched and IMPACT-BAM_matched wre combined as as tumor-derived. C. Distribution of alterations and variant origins in genes with more than 10 variants. Figure 14 shows classification schema for determining reference origin from matched WBC and tumor tissue sequencing. DETAILED DESCRIPTION This disclosure describes novel methods and systems for discriminating between tumor- origin variants and clonal hematopoiesis (CH)-derived variants across solid tumors and NGS sequencing platforms. The disclosed methods and systems utilize a machine learning model that integrates fragment-level, variant-level, and patient-level features, allowing for predicting tumor vs. CH variant origin using plasma-only sequencing without additional white blood cell (WBC) or tumor sequencing. The ability to identify bona fide tumor-derived variants in plasma-only sequencing fills a critical need in the clinical implementation of liquid biopsy-guided cancer therapy by reducing misinterpretation due to CH contamination. Accordingly, the methods and systems as disclosed have improved precision and reliability of prediction utilizing circulating- tumor DNA (ctDNA) for clinical prognosis purposes. Methods and Systems for Discriminating between CH-Derived and Tumor-Derived Variants In one aspect, this disclosure provides a method of discriminating between clonal hematopoiesis (CH)-derived and tumor-derived variants in plasma cell-free DNA (cfDNA) of a subject. In some embodiments, the method comprises: (i) determining values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; (ii) determining a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and (iii) identifying a variant in the plasma cfDNA as a CH-derived variant if the classification score is equal to or greater than a threshold value. In another aspect, this disclosure also provides a method of identifying a CH-derived variant in plasma cfDNA of a subject. In some embodiments, the method comprises: (a) determining values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; (b) determining a classification score 164810982.1 Docket: 348358.17201 by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and (c) identifying a variant in the plasma cfDNA as a CH-derived variant if the classification score is equal to or greater than a threshold value. In one aspect, we now provide a novel machine-learning model to distinguish tumor- origin variants from CH-associated variants in liquid biopsies from patients with solid tumors. By integrating information derived from fragment-, variant-, and patient-level features, our systems can effectively predict tumor vs. CH variant origin from plasma-only sequencing with high concordance with gold-standard variant origin determined from matched WBC and tumor sequencing (see, for instance Figure 2). This ability to identify bona fide tumor variants in plasma-only sequencing enhances the precision and reliability of utilizing ctDNA for clinical prognosis purposes. As used herein, the term “cell free DNA,” “cell-free DNA,” or “cfDNA” refers to strands of deoxyribose nucleic acids (DNA) found free of cells, for example, as extracted or isolated from plasma / serum of circulating blood, extracted from urine or other bodily fluids. As used herein, the terms “tumor-derived,” “tumor-associated,” or “tumor-related” in reference to variant in cfDNA refers to differences in DNA sequences of cfDNA in a patient whose cancer formed a tumor, when compared to reference DNA, such as when cfDNA is compared to control DNA (gDNA) from a cell that is not a tumor as described herein, or when pre-treatment cfDNA is compared to cfDNA collected during or after treatment. A “variant” of a first nucleotide sequence is defined as a nucleotide sequence that differs from a similar reference sequence or control sequence, e.g., by having one or more deletions, insertions, or substitutions that may be detected using DNA sequencing and / or digital DNA sequence comparison. For example, comparative digital methods may be used to match an entire region, loci, gene, or selected fragment of a first DNA sequence to a second DNA sequence. A “deletion” is defined as a change in either nucleotide in which one or more nucleotides are absent as compared to, for example, a reference sequence or control sequence. An “insertion” or “addition” is a change in a nucleotide sequence, which has resulted in the addition of one or more nucleotides as compared to, for example, a reference sequence or control sequence. A “substitution” results from the replacement of one or more nucleotides by different nucleotides as compared to, for example, a reference sequence or control sequence. 164810982.1 Docket: 348358.17201 In some embodiments, the fragment level features comprise differences in fragment length and cut points and endpoint motifs between mutant and wild type fragments at a locus. In some embodiments, the variant level features comprise variant allele frequency (VAF), scaled VAF, frequency of a variant in hematologic versus solid tumor malignancies in the Catalogue Of Somatic Mutations In Cancer (COSMIC), single base pair substitution, and estimate of driver function. In some embodiments, the patient level features comprise age. In some embodiments, the set of features comprise gene heme fraction, length cluster mutant, length MWU, length max delta S, and age. In some embodiments, the set of features further comprise endpoint motif down, endpoint location, plasma scaled VAF, length mean delta S, length cluster weight, variant COSMIC count, indel complex, mutant reads, or a combination thereof. Table 1. Description of Representative Features 164810982.1 Docket: 348358.17201 In some embodiments, the threshold value is about 0.5 (e.g., 0.5). In some embodiments, the method is capable of identifying the clonal hematopoiesis- derived variant with an accuracy of at least 75% (such as 75%, 78%, 80%, 82%, 84%, 86%, 88%, 90%, 92%, 94%, 95% or higher). 164810982.1 Docket: 348358.17201 An example process of variant origin prediction or discriminating between clonal hematopoiesis-derived and tumor-derived variants is illustrated in Figures 1a-b. For example, as shown in Fig. 1a, the process at 101 starts with receiving next-generation sequencing data, including alignment files and a list of mutations detected in the sequencing. At 102, the process may include extracting the sequencing reads that pertain to the list of detected mutations. For each mutation, the process at 103 may include determining which reads are mutant versus wild-type, and at 104, it may include calculating fragment-level summary statistics using read data. In addition, the process at 121 may include Annotating reported mutations with OpenCravat, and at 122, it may include calculating variant-level summary statistics. In addition, at 105, the process may include calculating a score for each reported mutation using fragment-level summary statistics, variant-level summary statistics, and patient-level summary statistics using a machine learning model. At 106, the process may further include annotating reported mutations as CH or tumor- derived using the calculated score. In some embodiments, the step of determining values of the set of features comprises extracting reads covering each variant locus from plasma-targeted next-generation sequencing (NGS) data. In some embodiments, the step of determining values of the set of features comprises calculating fragment summary statistics associated with fragment length, cut points, and endpoint motifs. In some embodiments, the step of determining values of the set of features comprises annotating variants using OpenCravat – COSMIC variant, COSMIC gene, CHASMMplus, and Cancer Gene Census (CGC). In some embodiments, the step of determining values of the set of features comprises calculating gene and variant heme fraction and variant counts using COSMIC counts. In some embodiments, the step of determining values of the set of features comprises determining driver status using CHASMMplus score, sequence ontology, and Cancer Gene Census (CGC) driver class. In some embodiments, the step of determining values of the set of features comprises calculating scaled plasma variant allele frequency (VAF). As used herein, the term “classifiers” refers generally to various types of classifier frameworks, such as neural network classifiers, hierarchical classifiers, ensemble classifiers, etc. 164810982.1 Docket: 348358.17201 In addition, a classifier design can include a multiplicity of classifiers that attempt to partition data into two groups, either organized hierarchically or run in parallel, and then combined to find the best classification. Further, a classifier can include ensemble classifiers wherein a large number of classifiers all attempting to perform the same classification task are learned, but trained with different data / variables / parameters, and then combined to produce a final classification label. The classification methods implemented may be “black boxes” that are unable to explain their prediction to a user (which is the case if classifiers are built using neural networks, for example). The classification methods may be “white boxes” that are in a human-readable form (which is the case if classifiers are built using decision trees, for example). In other embodiments, the classification models may be “gray boxes” that can partially explain how solutions are derived (e.g., a combination of “white box” and “black box” type classifiers). As used herein, the term “classification” refers to any number or other characters that are associated with a particular property of a sample. The classification can be binary (e.g., positive or negative) or have more levels of classification (e.g., a scale from 1 to 10 or 0 to 1). The term “cutoff” or “threshold” refers to a predetermined number used in an operation. For example, a cutoff value can refer to a classification score as used above. A threshold value may be a value above or below which a particular classification applies. Either of these terms can be used in either of these contexts. As used herein, a “machine learning model,” a “model,” or a “classifier” refers to a set of algorithmic routines and parameters that can predict an output(s) for a process input based on a set of input features, with or without being explicitly programmed. A structure of the software routines (e.g., number of subroutines and relation between them) and / or the values of the parameters can be determined in a training process, which can use actual results of the process that is being modeled. Such systems or models are understood to be necessarily rooted in computer technology, and, in fact, cannot be implemented or even exist in the absence of computing technology. While machine learning systems utilize various types of statistical analyses, machine learning systems are distinguished from statistical analyses by virtue of the ability to learn without explicit programming and being rooted in computer technology. A neural network or an artificial neural network is one set of algorithms used in machine learning for modeling the data using graphs of neurons. Any network structure may be used. Any number of layers, nodes within layers, types of nodes (activations), types of layers, interconnections, learnable parameters, and / or 164810982.1 Docket: 348358.17201 other network architectures may be used. Machine training uses the defined architecture, training data, and optimization to learn values of the learnable parameters of the architecture based on the samples and ground truth of training data. A typical machine learning pipeline may include building a machine learning model from a sample dataset (referred to as a “training set”), evaluating the model against one or more additional sample datasets (referred to as a “validation set” and / or a “test set”) to decide whether to keep the model and to benchmark how good the model is, and using the model in “production” to make predictions or decisions against live input data captured by an application service. For training the model to be applied as a machine-learned model, training data is acquired and stored in a database or memory. The training data is acquired by aggregation, mining, and loading from a publicly or privately formed collection, transfer, and / or access. Ten, hundreds, or thousands of samples of training data are acquired. The samples are from scans of different patients and / or phantoms. Simulation may be used to form the training data. The training data includes the desired output (ground truth), such as segmentation, and the input, such as protocol data and imaging data. In some embodiments, the training set will be used to create a single classifier using any now or hereafter-known methods. In other embodiments, a plurality of training sets will be created to generate a plurality of corresponding classifiers. Each of the plurality of classifiers can be generated based on the same or different learning algorithm that utilizes the same or different features in the corresponding one of the pluralities of training sets. For example, each of the plurality of neural network models can be trained on a training set classified on sequence type, view type, anatomy type and / or other image classifying data as discussed in conjunction with the disclosure. Once trained, the machine-learned or trained classifier is stored for later application. The training determines the values of the learnable parameters of the network. The network architecture, values of non-learnable parameters, and values of the learnable parameters are stored as the machine-learned network. Once stored, the machine-learned network may be fixed. The same machine-learned network may be applied to different patients, different scanners, and / or with different imaging protocols for the scanning. The machine-learned network may be updated. As additional training data is acquired, such as through application of the network for patients and 164810982.1 Docket: 348358.17201 corrections by experts to that output, the additional training data may be used to re-train or update the training. For the machine learning model, input data structures of subreads can be used for the training. The input data structure may correspond to a window of nucleotides sequenced in a sample nucleic acid molecule. The training set can have sites with known mutation status. Each training sample can include one of the first plurality of first data structures and a label indicating the first state for the modification (e.g., mutation) of the nucleotide at the target position. The training is performed by optimizing parameters of the model based on outputs of the model matching or not matching corresponding labels of the first labels and optionally the second labels when the first plurality of first data structures and optionally the second plurality of second data structures are input to the model. An output of the model specifies whether the nucleotide at the target position in the respective window has the modification. In some embodiments, the output of the model may include a probability of being in each of a plurality of states. The state with the highest probability can be taken as the state. The model may include a convolutional neural network (CNN). The CNN may include a set of convolutional filters configured to filter the first plurality of data structures and, optionally, the second plurality of data structures. The filter may be any filter described herein. The number of filters for each layer may be from 10 to 20, 20 to 30, 30 to 40, 40 to 50, 50 to 60, 60 to 70, 70 to 80, 80 to 90, 90 to 100, 100 to 150, 150 to 200, or more. The kernel size for the filters can be 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, from 15 to 20, from 20 to 30, from 30 to 40, or more. The CNN may include an input layer configured to receive the filtered first plurality of data structures and, optionally, the filtered second plurality of data structures. The CNN may also include a plurality of hidden layers, including a plurality of nodes. The first layer of the plurality of hidden layers is coupled to the input layer. The CNN may further include an output layer coupled to a last layer of the plurality of hidden layers and configured to output an output data structure. The output data structure may include the properties. The model may include a supervised learning model. Supervised learning models may include different approaches and algorithms including analytical learning, artificial neural network, backpropagation, boosting (meta-algorithm), Bayesian statistics, case-based reasoning, decision tree learning, inductive logic programming, Gaussian process regression, genetic programming, 164810982.1 Docket: 348358.17201 group method of data handling, kernel estimators, learning automata, learning classifier systems, minimum message length (decision trees, decision graphs, etc.), multilinear subspace learning, naive Bayes classifier, maximum entropy classifier, conditional random field, Nearest Neighbor Algorithm, probably approximately correct learning (PAC) learning, ripple down rules, a knowledge acquisition methodology, symbolic machine learning algorithms, subsymbolic machine learning algorithms, support vector machines, Minimum Complexity Machines (MCM) , random forests, ensembles of classifiers, ordinal classification, data pre-processing, handling imbalanced datasets, statistical relational learning, or Proaftn, a multicriteria classification algorithm The model may linear regression, logistic regression, deep recurrent neural network (e.g., long short term memory, LSTM), Bayes classifier, hidden Markov model (HMM) , linear discriminant analysis (LDA) , k-means clustering, density-based spatial clustering of applications with noise (DBSCAN) , random forest algorithm, support vector machine (SVM), or any model described herein. In some embodiments, the classifier comprises a supervised or unsupervised Machine Learning or Deep Learning algorithm, Logistic Regression, Naive Bayes, Support Vector Machine, Decision Tree, Random Forest, Gradient Boosting, Regularizing Gradient Boosting, K-Nearest Neighbors, a continuous regression approach, Ridge Regression, Kernel Ridge Regression, Support Vector Regression, deep learning approach, Neural Networks, Convolutional Neural Network (CNNs), Recurrent Neural Networks (RNNs), Long Short Term Memory Networks (LSTMs), Generative Models, Generative Adversarial Networks (GANs), Deep Belief Networks (DBNs), Feedforward Neural Networks, Autoencoders, Radial Basis Function Networks (RBFNs), Multilayer Perceptrons (MLPs), Stochastic Neural Networks, or any combination thereof. In some embodiments, the classifier comprises an XGBoost algorithm. In some embodiments, the subject has no patient-matched white blood cell (WBC) sequencing data available. In some embodiments, the subject has cancer. In some embodiments, the cancer is a carcinoma, a sarcoma, a lymphoma, a melanoma, a pediatric tumor, or a leukemia. In some embodiments, the subject has a solid tumor. In some embodiments, the plasma cfDNA comprises tumor-circulating DNA (ctDNA). 164810982.1 Docket: 348358.17201 The terms “patient,” “subject,” “host,” and “individual” are used interchangeably herein and refer to any subject, particularly a vertebrate subject, and even more particularly a mammalian subject, for whom therapy or prophylaxis is desired. Suitable vertebrate animals that fall within the scope of the invention include, but are not restricted to, any member of the subphylum Chordata including primates (e.g., humans, monkeys and apes, and includes species of monkeys such from the genus Macaca (e.g., cynomolgus monkeys such as Macaca fascicularis, and / or rhesus monkeys (Macaca mulatta )) and baboon (Papio ursinus), as well as marmosets (species from the genus Callithrix), squirrel monkeys (species from the genus Saimiri) and tamarins (species from the genus Saguinus), as well as species of apes such as chimpanzees ( Pan troglodytes)), rodents (e.g., mice rats, guinea pigs), lagomorphs (e.g., rabbits, hares), bovines (e.g., cattle), ovines (e.g., sheep), caprines (e.g., goats), porcines (e.g., pigs), equines (e.g., horses), canines (e.g., dogs), felines (e.g., cats), avians (e.g., chickens, turkeys, ducks, geese, companion birds such as canaries, budgerigars etc.), marine mammals (e.g., dolphins, whales), reptiles (snakes, frogs, lizards etc.), and fish. In some embodiments, a subject is a human with cancer. In some embodiments, the cancer is selected from adrenal gland tumors, biliary cancer, bladder cancer, brain cancer, breast cancer, carcinoma, central or peripheral nervous system tissue cancer, cervical cancer, colon cancer, endocrine or neuroendocrine cancer or hematopoietic cancer, esophageal cancer, fibroma, gastrointestinal cancer, glioma, head and neck cancer, Li-Fraumeni tumors, liver cancer, lung cancer, lymphoma, melanoma, meningioma, multiple neuroendocrine type I and type II tumors, nasopharyngeal cancer, oral cancer, oropharyngeal cancer, osteogenic sarcoma tumors, ovarian cancer, pancreatic cancer, pancreatic islet cell cancer, parathyroid cancer, pheochromocytoma, pituitary tumors, prostate cancer, rectal cancer, renal cancer, respiratory cancer, sarcoma, skin cancer, stomach cancer, testicular cancer, thyroid cancer, tracheal cancer, urogenital cancer, and uterine cancer. In some embodiments, the method comprises obtaining from a sample from the subject comprising the plasma cfDNA. In some embodiments, the sample comprises a biological sample that comprises a plasma sample, a blood sample, or a tissue sample. The terms “sample” or “biological sample,” as used herein, include any biological specimen obtained (isolated, removed) from a subject. Samples may include, without limitation, organ tissue (e.g., primary or metastatic tumor tissue), whole blood, plasma, serum, whole blood 164810982.1 Docket: 348358.17201 cells, red blood cells, white blood cells (e.g., peripheral blood mononuclear cells), saliva, urine, stool (feces), tears, sweat, sebum, nipple aspirate, ductal lavage, tumor exudates, synovial fluid, cerebrospinal fluid, lymph, fine needle aspirate, amniotic fluid, any other bodily fluid, exudate or secretory fluid, cell lysates, cellular secretion products, inflammation fluid, semen, and vaginal secretions. In some embodiments, a sample may be readily obtainable by non-invasive or minimally invasive methods, such as blood collection (“liquid biopsy”), urine collection, feces collection, tissue (e.g., tumor tissue) biopsy or fine-needle aspiration, allowing the provision / removal / isolation of the sample from a subject. The term “tissue,” as used herein, encompasses all types of cells of the body, including cells of organs but also including blood and other body fluids recited above. The tissue may be healthy or affected by pathological alterations, e.g., tumor tissue. The tissue may be from a living subject or may be cadaveric tissue. In some embodiments, useful samples are those known to comprise, expected, or predicted to comprise, known to potentially comprise, or expected or predicted to potentially comprise tumor cells. A sample can be obtained from a subject in any way typically used in clinical settings for obtaining a sample comprising the required cells or nucleic acid, including RNA, genomic DNA, mitochondrial DNA, and protein-associated nucleic acids. For example, the sample can be obtained from fresh, frozen, or paraffin-embedded surgical samples or biopsies of an organ or tissue comprising the suitable cells or nucleic acid to be tested. If desired, the sample can be mixed with a fluid or purified or amplified or otherwise treated. For examples, samples may be treated in one or more purification steps to increase the purity of the desired cells or nucleic acid in the sample, or they may be examined without any purification steps. Any nucleic acid specimen in purified or non-purified form obtained from such sample can be utilized in the methods as taught herein. In some embodiments, the disclosed methods may be used for prognosis or predicting responsiveness to a therapy. The terms “determining responsiveness,” “predicting responsiveness,” and “assessing a likelihood of a therapeutic response” may be used interchangeably herein. The term “prognosis,” as used herein, refers to anticipation of progression of a disease (e.g., cancer) or condition and prospect (e.g., the probability, duration, and / or extent) of recovery. A good prognosis of the diseases or conditions may generally encompass anticipation of a 164810982.1 Docket: 348358.17201 satisfactory partial or complete recovery from the diseases or conditions, such as within an acceptable time period. A good prognosis of such may more commonly encompass anticipation of not further worsening or aggravating within a given time period. A poor prognosis of the diseases or conditions as taught herein may generally encompass anticipation of a substandard recovery and / or unsatisfactorily slow recovery, or to substantially no recovery or even further worsening of such. In some embodiments, a therapeutic response may include an anti-tumor response when referring to a cancer patient treated with a cancer therapy. For example, an anti-tumor response may include at least one positive therapeutic effect, such as a reduced number of cancer cells, reduced tumor size, reduced rate of cancer cell infiltration into peripheral organs, reduced rate of tumor metastasis or tumor growth, or progression-free survival. Positive therapeutic effects in cancer can be measured in a number of ways (see, e.g., W. A. Weber, J. Null. Med. 50:1S-10S (2009); Eisenhauer et al., 2009 European Journal of Cancer, 45: 228-247). In some embodiments, an anti-tumor response to a cancer therapy is assessed using RECIST 1.1 criteria, bidimensional irRC, or unidimensional irRC. In some embodiments, an anti-tumor response is any of stable disease (SD), partial response (PR), complete response (CR), progression-free survival (PFS), and disease-free survival (DFS). In some embodiments, one or more biomarkers of this disclosure predict whether a subject with a solid tumor is likely to achieve a complete response or a partial response. The term “predicting,” “prediction,” or “predictive,” as used herein, refers to an advance declaration, indication, or foretelling of a response or reaction to a therapy (e.g., chemotherapy, immunotherapy, immunochemotherapy) in a subject not (yet) having been treated with the therapy. For example, a prediction of responsiveness (or sensitivity or susceptibility) to a cancer therapy in a subject may indicate that the subject will respond or react to the cancer therapy, for example, within a certain time period, e.g., so that the subject will have a clinical benefit from the cancer therapy. A prediction of unresponsiveness (or insensitivity or insusceptibility) to a cancer therapy in a subject may indicate that the subject will minimally or not respond or react to the cancer therapy, for example, within a certain time period, e.g., so that the subject will have no clinical benefit from the cancer therapy. 164810982.1 Docket: 348358.17201 In another aspect, this disclosure also provides a system of discriminating between CH- derived and tumor-derived variants in plasma cfDNA of a subject. In some embodiments, the system comprises one or more processors configured to: (a) determine values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; (b) determine a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and (c) identify a variant in the plasma cfDNA as a CH-derived variant if the classification score is equal to or greater than a threshold value. In yet another aspect, this disclosure further provides a system of identifying a CH-derived variant in plasma cfDNA of a subject. In some embodiments, the system comprises one or more processors configured to: (a) determine values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; (b) determine a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and (c) identify a variant in the plasma cfDNA as a CH-derived variant if the classification score is equal to or greater than a threshold value. Figure 8 is a functional diagram illustrating a programmed computer system in accordance with some embodiments. As will be apparent, other computer system architectures and configurations can be used to perform the described methods. Computer system 600, which includes various subsystems as described below, includes at least one microprocessor subsystem (also referred to as a processor or a central processing unit (CPU) 606). For example, processor 606 can be implemented by a single-chip processor or by multiple processors. In some embodiments, processor 606 is a general-purpose digital processor that controls the operation of the computer system 600. In some embodiments, processor 606 also includes one or more coprocessors or special purpose processors (e.g., a graphics processor, a network processor, etc.). Using instructions retrieved from memory 607, processor 606 controls the reception and manipulation of input data received on an input device (e.g., image processing device 603, I / O device interface 602), and the output and display of data on output devices (e.g., display 601). Processor 606 is coupled bi-directionally with memory 607, which can include, for example, one or more random access memories (RAM) and / or one or more read-only memories 164810982.1 Docket: 348358.17201 (ROM). As is well known in the art, memory 607 can be used as a general storage area, a temporary scratchpad) memory, and / or a cache memory. Memory 607 can also be used to store input data and processed data, as well as to store programming instructions and data, in the form of data objects and text objects, in addition to other data and instructions for processes operating on processor 606. Also, as is well known in the art, memory 607 typically includes basic operating instructions, program code, data, and objects used by the processor 606 to perform its functions (e.g., programmed instructions). For example, memory 607 can include any suitable computer-readable storage media described below, depending on whether, for example, data access needs to be bi-directional or uni-directional. For example, processor 606 can also directly and very rapidly retrieve and store frequently needed data in a cache memory included in memory 607. A removable mass storage device 608 provides additional data storage capacity for the computer system 600 and is optionally coupled either bi-directionally (read / write) or uni- directionally (read-only) to processor 606. A fixed mass storage 609 can also, for example, provide additional data storage capacity. For example, storage devices 608 and / or 609 can include computer-readable media such as magnetic tape, flash memory, PC-CARDS, portable mass storage devices such as hard drives (e.g., magnetic, optical, or solid-state drives), holographic storage devices, and other storage devices. Mass storages 608 and / or 609 generally store additional programming instructions, data, and the like that typically are not in active use by the processor 606. It will be appreciated that the information retained within mass storages 608 and 609 can be incorporated, if needed, in a standard fashion as part of memory 607 (e.g., RAM) as virtual memory. In addition to providing processor 606 access to storage subsystems, bus 610 can be used to provide access to other subsystems and devices as well. As shown, these can include a display 601, a network interface 604, an input / output (I / O) device interface 602, an image processing device 603, as well as other subsystems and devices. For example, image processing device 603 can include a camera, a scanner, etc.; I / O device interface 602 can include a device interface for interacting with a touchscreen (e.g., a capacitive touch sensitive screen that supports gesture interpretation), a microphone, a sound card, a speaker, a keyboard, a pointing device (e.g., a mouse, a stylus, a human finger), a global positioning system (GPS) receiver, a differential global positioning system (DGPS) receiver, an accelerometer, and / or any other appropriate device 164810982.1 Docket: 348358.17201 interface for interacting with system 600. Multiple I / O device interfaces can be used in conjunction with computer system 600. The I / O device interface can include general and customized interfaces that allow the processor 606 to send and, more typically, receive data from other devices such as keyboards, pointing devices, microphones, touchscreens, transducer card readers, tape readers, voice or handwriting recognizers, biometrics readers, cameras, portable mass storage devices, and other computers. The network interface 604 allows processor 606 to be coupled to another computer, computer network, or telecommunications network using a network connection as shown. For example, through the network interface 604, the processor 606 can receive information (e.g., data objects or program instructions) from another network, or output information to another network in the course of performing method / process steps. Information, often represented as a sequence of instructions to be executed on a processor, can be received from and outputted to another network. An interface card or similar device and appropriate software implemented by (e.g., executed / performed on) processor 606 can be used to connect the computer system 600 to an external network and transfer data according to standard protocols. For example, various process embodiments disclosed herein can be executed on processor 606 or can be performed across a network such as the Internet, intranet networks, or local area networks, in conjunction with a remote processor that shares a portion of the processing. Additional mass storage devices (not shown) can also be connected to processor 606 through network interface 604. In addition, various embodiments disclosed herein further relate to computer storage products with a computer-readable medium that includes program code for performing various computer-implemented operations. The computer-readable medium includes any data storage device that can store data that can thereafter be read by a computer system. Examples of computer- readable media include, but are not limited to: magnetic media such as disks and magnetic tape; optical media such as CD-ROM disks; magneto-optical media such as optical disks; and specially configured hardware devices such as application-specific integrated circuits (ASICs), programmable logic devices (PLDs), and ROM and RAM devices. Examples of program code include both machine code as produced, for example, by a compiler, or files containing higher level code (e.g., script) that can be executed using an interpreter. 164810982.1 Docket: 348358.17201 The computer system as shown in Figure 8 is an example of a computer system suitable for use with the various embodiments disclosed herein. Other computer systems suitable for such use can include additional or fewer subsystems. In some computer systems, subsystems can share components (e.g., for touchscreen-based devices such as smartphones, tablets, etc., I / O device interface 602 and display 601 share the touch-sensitive screen component, which both detects user inputs and displays outputs to the user). In addition, bus 610 is illustrative of any interconnection scheme serving to link the subsystems. Other computer architectures having different configurations of subsystems can also be utilized. Additional Definitions To aid in understanding the detailed description of the compositions and methods according to the disclosure, a few express definitions are provided to facilitate an unambiguous disclosure of the various aspects of the disclosure. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. Unless defined otherwise, all technical and scientific terms used herein have the meaning commonly understood by a person skilled in the art to which this invention belongs. The following references provide one of skill with a general definition of many of the terms used in this invention: Singleton et al., Dictionary of Microbiology and Molecular Biology (2nd ed. 1994); The Cambridge Dictionary of Science and Technology (Walker ed., 1988); The Glossary of Genetics, 5th Ed., R. Rieger et al. (eds.), Springer Verlag (1991); and Hale & Marham, The Harper Collins Dictionary of Biology (1991). As used herein, the following terms have the meanings ascribed to them below, unless specified otherwise. Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. In some embodiments, the flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, a segment, or a portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some 164810982.1 Docket: 348358.17201 alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions. These computer readable program instructions may be provided to a processor of a general- purpose computer, a special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks. The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks. It will be understood that, although the terms “first,” “second,” etc., may be used herein to describe various elements, components, regions, layers and / or sections. These elements, components, regions, layers and / or sections should not be limited by these terms. These terms are only used to distinguish one element, component, region, layer or section from another element, component, region, layer or section. Thus, a first element, component, region, layer or section discussed below could be termed a second element, component, region, layer or section without departing from the teachings of example embodiments. 164810982.1 Docket: 348358.17201 Unless specifically stated otherwise, as apparent from the above discussion, it is appreciated that throughout the description, discussions utilizing terms such as “processing,” “performing,” “receiving,” “computing,” “calculating,” “determining,” “identifying,” “displaying,” “providing,” “merging,” “combining,” “running,” “transmitting,” or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (or electronic) quantities within the computer system memories or registers or other such information storage, transmission or display devices. As used herein, the term “if may be construed to mean “when” or “upon” or “in response to determining” or “in response to detecting,” depending on the context. Similarly, the phrase “if it is determined” or “if [a stated condition or event] is detected” may be construed to mean “upon determining” or”in response to determining” or “upon detecting [the stated condition or event]” or “in response to detecting [the stated condition or event],” depending on the context. The terms or acronyms like “convolutional neural network,” “CNN,” “neural network,” “NN,” “deep neural network,” “DNN,” “recurrent neural network,” “RNN,” and / or the like may be interchangeably referenced throughout this document. An “electronic device” or a “computing device” refers to a device that includes a processor and memory. Each device may have its own processor and / or memory, or the processor and / or memory may be shared with other devices as in a virtual machine or container arrangement. The memory will contain or receive programming instructions that, when executed by the processor, cause the electronic device to perform one or more operations according to the programming instructions. The terms “memory,” “memory device,” “computer-readable medium,” “data store,” “data storage facility” and the like each refer to a non-transitory device on which computer-readable data, programming instructions or both are stored. Except where specifically stated otherwise, the terms “memory,” “memory device,” “computer-readable medium,” “data store,” “data storage facility” and the like are intended to include single device embodiments, embodiments in which multiple memory devices together or collectively store a set of data or instructions, as well as individual sectors within such devices. The terms “processor” and “processing device” refer to a hardware component of an electronic device that is configured to execute programming instructions, such as a microprocessor 164810982.1 Docket: 348358.17201 or other logical circuit. A processor and memory may be elements of a microcontroller, custom configurable integrated circuit, programmable system-on-a-chip, or other electronic device that can be programmed to perform various functions. Except where specifically stated otherwise, the singular term “processor” or “processing device” is intended to include both single-processing device embodiments and embodiments in which multiple processing devices together or collectively perform a process. In this document, the terms “communication link” and “communication path” mean a wired or wireless path via which a first device sends communication signals to and / or receives communication signals from one or more other devices. Devices are “communicatively connected” if the devices are able to send and / or receive data via a communication link. “Electronic communication” refers to the transmission of data via one or more signals between two or more electronic devices, whether through a wired or wireless network, and whether directly or indirectly via one or more intermediary devices. As used herein, “in vitro” refers to events that occur in an artificial environment, e.g., in a test tube or reaction vessel, in cell culture, etc., rather than within a multi-cellular organism. As used herein, “in vivo” refers to events that occur within a multi-cellular organism, such as a non-human animal. It is noted here that, as used in this specification and the appended claims, the singular forms “a,” “an,” and “the” include plural reference unless the context clearly dictates otherwise. As used herein, “plurality” means two or more. As used herein, a “set” of items may include one or more of such items. As used herein, “including,” “comprising,” “containing,” or “having” and variations thereof are meant to encompass the items listed thereafter and equivalents thereof as well as additional subject matter unless otherwise noted. As used herein, the phrases “in one embodiment,” “in various embodiments,” “in some embodiments,” and the like do not necessarily refer to the same embodiment, but may unless the context dictates otherwise. As used herein, the terms “and / or” or “ / ” means any one of the items, any combination of the items, or all of the items with which this term is associated. 164810982.1 Docket: 348358.17201 As used herein, the term “substantially” does not exclude “completely,” e.g., a composition which is “substantially free” from Y may be completely free from Y. Where necessary, the word “substantially” may be omitted from the definition of the present disclosure. As used herein, the term “approximately” or “about,” as applied to one or more values of interest, refers to a value that is similar to a stated reference value. In some embodiments, the term “approximately” or “about” refers to a range of values that fall within 25%, 20%, 19%, 18%, 17%, 16%, 15%, 14%, 13%, 12%, 11%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, or less in either direction (greater than or less than) of the stated reference value unless otherwise stated or otherwise evident from the context (except where such number would exceed 100% of a possible value). Unless indicated otherwise herein, the term “about” is intended to include values, e.g., weight percents, proximate to the recited range that are equivalent in terms of the functionality of the individual ingredient, the composition, or the embodiment. As used herein, the term “each,” when used in reference to a collection of items, is intended to identify an individual item in the collection but does not necessarily refer to every item in the collection. Exceptions can occur if explicit disclosure or context clearly dictates otherwise. As disclosed herein, a number of ranges of values are provided. It is understood that each intervening value, to the tenth of the unit of the lower limit, unless the context clearly dictates otherwise, between the upper and lower limits of that range is also specifically disclosed. Each smaller range between any stated value or intervening value in a stated range and any other stated or intervening value in that stated range is encompassed within the present disclosure. The upper and lower limits of these smaller ranges may independently be included or excluded in the range, and each range where either, neither, or both limits are included in the smaller ranges is also encompassed within the present disclosure, subject to any specifically excluded limit in the stated range. Where the stated range includes one or both of the limits, ranges excluding either or both of those included limits are also included in the present disclosure. The use of any and all examples, or exemplary language (e.g., “such as”) provided herein, is intended merely to better illuminate the invention and does not pose a limitation on the scope of the present disclosure unless otherwise claimed. No language in the specification should be construed as indicating any non-claimed element as essential to the practice of the present disclosure. 164810982.1 Docket: 348358.17201 All methods described herein are performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. In regard to any of the methods provided, the steps of the method may occur simultaneously or sequentially. When the steps of the method occur sequentially, the steps may occur in any order, unless noted otherwise. In cases in which a method comprises a combination of steps, each and every combination or sub- combination of the steps is encompassed within the scope of the disclosure, unless otherwise noted herein. Each publication, patent application, patent, and other reference cited herein is incorporated by reference in its entirety to the extent that it is not inconsistent with the present disclosure. Publications disclosed herein are provided solely for their disclosure prior to the filing date of the present disclosure. Nothing herein is to be construed as an admission that the present disclosure is not entitled to antedate such publication by virtue of prior disclosure. Further, the dates of publication provided may be different from the actual publication dates, which may need to be independently confirmed. It is understood that the examples and embodiments described herein are for illustrative purposes only and that various modifications or changes in light thereof will be suggested to persons skilled in the art and are to be included within the spirit and purview of this application and scope of the appended claims. Examples EXAMPLE 1 This example describes the materials and methods used in EXAMPLE 2. Evaluation of fragmentomic features In order to facilitate incorporating fragmentomics into the machine learning model, we sought to identify the summary statistics that would consistently capture differences in fragmentation patterns in mutant versus non-mutant reads that could be derived from standard plasma NGS. Though fragment length and location of fragment endpoints are intrinsically intertwined, we examined summary statistics for both of these features as the information contained is not entirely overlapping. Figures 3a-c show illustrative examples of potential fragmentation patterns. In Figure 3a, which is CH-derived variant, the distribution of read length 164810982.1 Docket: 348358.17201 for mutant fragments overlaps that of wild-type fragments, and the location of fragment end- points relative to the location of the mutation is similar. In contrast, Figure 3b shows a tumor- derived variant where mutant read length is shorter than wild-type, resulting in a left-shift of the length cumulative density function and a shifting of the relative endpoint peaks closer together. While shorter fragment length distribution is often reported in tumor-derived variants, it is not seen universally. Figure 3c shows a different tumor-derived variant and here the mutant read length is similar to wild-type, as seen in overlapping length cumulative density, but the relative endpoint locations for mutant reads are shifted, which we hypothesize may be the result of the tumor-derived cfDNA is being cleaved at different locations than the non-tumor cfDNA. Thus, we decided to incorporate multiple fragmentomics-based summary statistics in order to more fully capture the diverse potential differences in fragmentation patterns. In an effort to limit noise, we reasoned that the fragmentation summary statistics that were more representative of biologic differences in cfDNA fragmentation between cancer and non-cancer would be consistent over time. We examined the correlation over time of different fragmentation summary statistics using a cohort of 41 variants found in 29 patients with stage I- III colorectal, esophageal, or NSCLC who had serial cfDNA plasma sequencing performed while undergoing treatment as part of clinical trials (Figure 3d). Potential summary statistics that could be used to describe differences between mutant and wild-type fragments were gleaned from the literature as well as statistical theory, and included length, endpoint location, and endpoint 4-mer motif. Each statistical value was calculated by comparing wild-type to mutant reads for each patient-locus at each time point independently. We then calculated the correlation of the statistical values between baseline and follow-up time points (Figure 9), regardless of variant origin. The correlation between the value of each statistic at baseline and at follow-up timepoints is shown in Figure 3d. Notably, two summary statistics commonly used to describe differences in fragment lengths, the F-statistic for variance and T-statistic for mean, were not highly correlated between baseline and follow-up timepoints, and so were excluded from the model. Fragmentation summary statistics that were significantly correlated between baseline and follow-up timepoints were then incorporated as features into the machine learning model. Model development and performance in the training cohort 164810982.1 Docket: 348358.17201 The training set consisted of 426 plasma variants from 225 patients with stage I-IV breast, colorectal, esophageal, ovarian, non-small cell (NSCLC) and small-cell lung (SCLC) cancer, all of whom had also had matched WBC DNA sequencing, and 67.6% had undergone tumor tissue sequencing, allowing determination of reference origin (Figure 4a, Figure 10). Tumor-origin variants had statistically significant higher variant allele frequency (VAF) compared to CH-origin variants (mean 0.084 versus 0.023, Welch two sample t-test p-value = 4.634e-07, Figure 3b). However, given significant overlap in distribution, VAF alone was only able to distinguish variant origin with AUC 0.61 (95% confidence interval 0.56 - 0.67, Figure 4c). We then combined variant and patient level features with the fragmentomic features identified using the serial cohort into a machine learning model to predict tumor- vs CH-origin. Variant level features included the variant allele frequency, the scaled variant allele frequency relative to other variants identified in a patient’s plasma, and variant and gene annotations of the reported hematologic versus solid tumor frequency in COSMIC and the patient level feature was age. Individual features distinguished plasma variant origin with AUC ranging from 0.55 (plasma scaled VAF) to 0.86 (gene heme fraction, Figure 5a, Figure 11). By integrating the features into a machine learning model using XGboost, we were able to predict variant origin with an AUC 0.93788 (95% confidence interval 0.85288-1, 10-fold cross validation repeated 10 times, Figure 5b). Model feature importance analysis revealed that the top five most important features included features from all three levels - fragment, variant, and patient (Figure 5c). The best performing model determined using the training cohort was then locked for evaluation in the validation cohort. Additionally, we examined the impact of number of mutant reads on model performance, as there is increased uncertainty in the summary statisitics used to represent fragmentomic differences when there are small numbers of mutant reads. Model performance decreased to AUC of 0.841 when only 3-5 mutant reads were detected (Figure 12). When more than five mutant reads were detected, model performance had AUC of at least 0.89 in the training cohort. Model performance in the independent validation cohort In order to further assess our model, we examined performance in an independent validation cohort which included patients with metastatic breast, prostate and NSCL cancer who 164810982.1 Docket: 348358.17201 had undergone sequencing of plasma, WBC, and tumor tissue to determine reference origin14. Using the available the plasma sequencing files and published clinical data, we were able calculate model features for 1,412 variants (507 tumor-derived, 911 CH-derived) found in 114 patients (Figure 6a-b, Figure 13). Using a pre-specified cutoff score of 0.5, the model predicted the plasma variant was tumor-origin with sensitivity of 82%, specificity 80.3%, and accuracy of 80.2% (p-value <2e-16 compared to null information rate of 0.6112, Figure 6c-d). The locked machine learning model derived from training cohort predicted variant origin in the independent validation cohort with an AUC of 0.902. We further examined the performance of the model in key subsets by cancer type and gene. The model performed well across all three cancer types included in the validation cohort (Figure 7a): breast (accuracy 77.7%, p=1.02e-12), NSCLC (accuracy 82.4%, p=<2.2e-16), and prostate (accuracy 81.8%, p=3.14e-7). Notably, the training cohort did not include any patients with prostate cancer, but there was not a significant decrease in performance in patients with prostate cancer in the validation cohort compared to other cancer types that had been included in the training set. Within gene sets, we compared the accuracy using model predicted origin compared to null hypothesis that all variants are tumor-derived (Figure 7b. For clinically relevant genes that can be commonly mutated in both CH and solid tumors (AKT1, ATM, BRCA1, BRCA2, EGFR), the model predicted variant origin with accuracy 82.8% compared to null assumption with accuracy 48.4% (p=1.84e-7). Similarly, TP53 is commonly mutated in both CH and solid tumors, and accuracy in predicting variant origin increased to 76.9% using the model compared to 65.4% under the null assumption (p=0.019). In the validation cohort, CH- origin and tumor-origin variants in TP53 were distributed across the gene, highlighting the difficulty of determining variant origin in this gene (Figure 7c). The model continued to have high accuracy even when excluding CH blacklist genes (i.e. genes that are commonly mutated in CH and less commonly mutated in solid tumors) with accuracy 75.7% versus 48.0% (p<2.2e-16). Targeted error-correction sequencing and sequence data processing For training and serial sets, plasma cfDNA and matched buffy coat WBC samples were sequenced using TECSeq (Jaiswal S, et al. Science. 2019;366(6465):eaan4673). Mean depth of coverage for both cfDNA and WBC was 5000x for panels of 63-219 genes associated with cancer. Details of sequencing approaches, including specifics of panels employed, have been described in 164810982.1 Docket: 348358.17201 detail for each cohort previously (Parikh AR, et al. Nat Med.2019;25(9):1415-1421). For patients from the MEDOCC subcohort, tissue sequencing was performed using a targeted panel of cancer- associated genes to an average depth of 750x. For the remaining patients with available matched tumor tissue, whole exome sequencing was performed on matched tumor and normal tissue to a mean depth of 100x, with sequence data processing and variant calling as described (Parikh AR, et al. Nat Med. 2019;25(9):1415-1421). A total of 426 variants with definitive origin were identified and used in the training cohort. The independent validation cohort included 124 patients with stage IV breast, NSCLC, or prostate cancer who had cfDNA, WBC (buffy coat) and tumor tissue obtained as part of a prospective observational study described by Razavi et al (Razavi P, et al. Nat Med. 2019;25(12):1928-1937). We included only patients that had both plasma variants with well- characterized reference origins (classified as WBC-matched, biopsy-matched, or IMPACT-bam- matched) and available age data. In total, 1,412 variants (507 tumor-derived, 911 CH-derived) found in 114 patients were included. All patients in this cohort had blood drawn within 6 weeks of tumor biopsy. No patients in this cohort were part of any of the 6 studies aggregated to create the training cohort, nor was any of the data used for informing the model parameters. Evaluation of mutation origin For the training and serial cohorts, the reference origin was determined for each variant identified in plasma sequencing by comparison to the matched WBC and tissue sequencing using a custom pipeline. First, all somatic variants are identified and annotated using the COSMIC database. Cancer hotspots are identified using a COSMIC threshold frequency of 25 or greater occurrences. All variants not identified as hotspots were classified as germline if they have a variant allele frequency greater than 25% for every plasma and WBC sample per patient. Variants detected in plasma were classified as WBC origin if the same mutation was detected in WBC sequencing with at least 5 supermutant families, or detected in WBC with 2-4 supermutant families and either reported as a hemotapoietic variant in COSMIC or not detected in tumor tissue sequencing with at least 100x coverage at that locus. Variants were classified as likely WBC origin if they were detected in WBC sequencing with 2-4 supermutant families and did not meet criteria for WBC origin, or were detected in 1 supermutant family and either not detected in tumor tissue 164810982.1 Docket: 348358.17201 sequencing with at least 100x coverage at that locus or reported predominantly in hematopoietic malignancies in COSMIC. Variants detected in plasma were classified as tumor origin if they were detected in tumor sequencing and found in less than or equal to one supermutant in WBC sequencing. Plasma variants were classified as likely tumor origin if they met at least two of the following criteria: no WBC supermutants with a read depth of at least 1000x at that locus, at least 10 reports in COSMIC with the primary tumor having a higher number of reports that hematologic malignancy, or at least two read families identified in tumor sequencing by IGV look-up. Plasma variants that were not classified by this method were of unknown origin and were removed from further analysis. For the independent validation cohort, the mutation origin for each variant described in the Supplementary Table 11 of Razavi et al. (Razavi P, et al. Nat Med. 2019;25(12):1928-1937) was used. Only mutations detected in patients with cancer and classified as phenotype WBC matched, biopsy matched, or IMPACT-bam matched were included; variants classified as uncertain origin were excluded. Both biopsy matched and IMPACT-bam matched were classified as tumor origin. Longitudinal evaluation In the longitudinal cohort, 41 variants were identified at two separate timepoints in a given patient. Of these, 23 variants detected at baseline timepoints were included in the training cohort. These variants underwent evaluation of origin by comparison to matched sequencing as detailed above, but variant origin was not considered in assessment of longitudinal stability of fragmentomic summary statistics. Fragmentomic summary statistics were calculated for each variant at a given timepoint. Then, the correlation between the summary statistics across timepoints was determined using Pearson’s product moment correlation between paired samples. Fragmentomic summary statistics whose correlation over time had p-value <0.05 were used for development of machine learning model. Machine learning The machine learning algorithm integrates patient age along with fragment level features and variant level features derived from plasma sequencing data to create a score from 0 to 1 for each variant. Features are described in detail in Table 1. Input features are pre-processed with centering, scaling, and Yeo Johnson transformation (Yeo and R.A. Johnson, Biometrika, 87(4), pp.954-959, (2000)), with all pre-processing done within each cross-validation fold. Output scores 164810982.1 Docket: 348358.17201 are binarized at a cutoff (default 0.5) into CH-predicted origin and tumor-predicted origin. The algorithm was implemented using XGBoost (version 1.6.0.1) and caret package (version 6.0-92) in R. Model training was performed using 10-fold cross validation repeated 10 times, optimizing ROC. Feature importance was extracted using built-in caret feature importance function. Statistical Analyses All statistical analysis was conducted in R Version 4.2.1. For the serial cohort, a correlation between the summary statistics across timepoints was determined using Pearson’s product moment correlation between paired samples. EXAMPLE 2 Creation of training cohort It was hypothesized that integration of fragmentomic, variant, and patient level features into a machine learning model would enable accurate prediction of tumor versus CH origin for variants identified in plasma-only sequencing (Figure 2a). In order to develop this approach, sequencing data was assembled, which are from 225 patients with stage I-IV breast, colorectal, esophageal, ovarian, non-small cell (NSCLC), and small-cell lung (SCLC) cancer, all of whom had undergone plasma cfDNA and WBC DNA sequencing, and 67.6% had undergone tumor tissue sequencing. By comparing across matched sequencing samples, a reference origin was identified for 426 plasma variants, which became the training cohort (Figure 2b). Tumor-origin variants had statistically significant higher variant allele frequency (VAF) compared to CH-origin variants (mean 0.084 versus 0.023, Welch two sample t-test p-value = 4.634e-07). However, given significant overlap in distribution, VAF alone was a poor predictor of variant origin with AUC 0.61433 (95% confidence interval 0.55979 - 0.66886). Identification of serially stable fragmentation statistics In order to facilitate the incorporation of fragmentomics into the development of machine learning model, it was sought to identify summary statistics that could be used to describe fragmentation patterns in mutant versus non-mutant reads, regardless of variant origin. Figures 3a and 3b show two illustrative examples of potential fragmentation patterns. In Figure 3a, which is a tumor-derived variant, mutant read length is shorter than wild-type, resulting in a left-shift of the length cumulative density function and a shifting of the relative endpoint peaks closer together. In 164810982.1 Docket: 348358.17201 Figure 3b, which is also a tumor-derived variant, mutant read length is similar to wild-type, as seen in overlapping length cumulative density, but the relative endpoint locations for mutant reads are shifted. Thus, in order to capture multiple possible differences in fragmentation patterns, there was a need to incorporate multiple summary statistics. Applying the assumption that fragmentation summary statistics that were representative of biology would be consistent over time, the stability of different fragmentation summary statistics was examined in a cohort of variants that were detected in patients at baseline and follow-up timepoints. This serial cohort included 41 variants found in 29 patients with stage I-III colorectal, esophageal, or NSCLC who had serial cfDNA plasma sequencing performed while undergoing treatment as part of clinical trials (Figure 3c). Summary statistics were calculated by comparing mutant versus wild-type reads for a given patient locus and did not take into account variant origin. Length statistics included F-statistic ratio of variances and T-statistic. Other statistics calculated related to length, endpoint location, and endpoint 4-mer motif are described in supplemental information. Each statistical value was calculated by comparing wild-type to mutant reads for each patient locus at each time point independently. The correlation of the statistical values between baseline and follow-up time points was then calculated. Notably, the F-statistic and T- statistic were not well correlated over time (Figure 3d). Fragmentation summary statistics that were significantly correlated using Pearson correlation over time were then incorporated as features into the machine learning model. Model development and performance in the training cohort The fragmentomic features identified as stable over time in the serial cohort were then combined with variant and patient level features to develop a machine learning model to predict tumor versus CH origin. Variant level features included the variant allele frequency, the scaled variant allele frequency relative to other variants identified in a patient’s plasma, and variant and gene annotations of the reported hematologic versus solid tumor frequency in COSMIC. The patient level feature was age. As seen in the heatmap in Figure 4a, no single feature clearly distinguished plasma variant origin. However, when integrated together into a machine learning model using XGboost, variant origin was predicted with an AUC 0.93788 (95% confidence interval 0.85288-1, 10-fold cross validation repeated 10 times, Figure 4b). A model feature 164810982.1 Docket: 348358.17201 importance analysis revealed that the top five most important features included features from all three levels - fragment, variant, and patient (Figure 4c). The best performing model was locked. Model performance in the independent validation cohort In order to further assess the model in an independent validation cohort, data related to Razavi et al. (Razavi P, et al. Nat Med. 2019;25(12):1928-1937) was obtained. This cohort is discussed above. Exemplary model performance in other gene sets, including all gene except DMT3A, genes associated with CHIP, and genes where the reference origin was split between tumor and CH- derived is shown in Table 3. Discusison Contamination of liquid biopsies by CH-origin variants for patients with solid tumors is extremely common. In preferred asepcts, the present systems provide a machine learning systems that incorporates fragmentomic, variant, and patient level features in order to distinguish CH-origin variants from bona fide tumor-origin variants with high accuracy in both training and independent validation cohorts. Other approaches for identifying potential CH-derived variants from plasma-only sequencing typically rely upon gene-based or variant allele frequency (VAF)- based filtering. Plasma VAF is suggested as a filtering method since CH-derived mutations typically have lower VAFs than tumor-derived mutations28. However, as we have shown in both the training and validation cohorts, VAF alone is a poor predictor of origin given the large overlap in distributions of VAF between CH and tumor-origin variants. Similarly, certain genes associated with hematologic malignancies, such as DNMT3A, TET2, and ASXL1 are commonly mutated in CH, and very uncommonly altered in solid tumors, and therefore are often assumed to be CH-derived2540. While this gene-list based approach can identify many common CH-derived alterations, it does not allow for distinguishing origin for genes that are frequently altered in both CH and solid tumors, such as TP5326. In contrast, the present systems can allow for determination of origin in genes outside those commonly considered CH- associated. Additionally, the present systems can allow for more accurate identification of origin of clinically relevant genes, such as BRCA1 / 2, which though less commonly associated with CH, can be CH-derived, and if erroneously assumed to be tumor-derived, could lead to the harmful misapplication of targeted therapy.27164810982.1 Docket: 348358.17201 The issue of CH-contamination in liquid biopsies can be potentially addressed by performing matched tumor tissue and / or white blood cell (WBC) sequencing, which we used here to derive a reference origin1718. Applying a tumor-informed approach does allow for the detection of bona fide tumor-derived mutations in plasma; however, this method reintroduces the issue of needing an adequate tissue sample, which may be difficult to obtain, and it does not enable assessment of the genetic heterogeneity of the cancer, which may be particularly important for patients with metastatic disease119. Similarly, a WBC-informed approach utilizes parallel sequencing of patient-matched WBCs to identify CH variants from the cfDNA results, and can achieve comparable results to tumor-informed methods1720. Although this approach bypasses the requirement for a tumor biopsy, extra time and costs are still needed to extract, sequence, and analyze the WBC DNA21. Currently, commercial targeted sequencing panels do not include matched WBC sequencing222324. Our model demonstrates significant utility in accurately identifying the origin of variants from plasma cfDNA without requiring matched WBC or tumor tissue sequencing. This approach offers a scalable and less invasive solution in clinical settings, where comprehensive sequencing of multiple tissue types may not be feasible. Additionally, applying this model can enable more accurate interpretation of retrospective research studies that generated cfDNA results without matched WBC sequencing44. Performance in the independent validation dataset demonstrated that our model can be generalizable across different sequencing platforms, cancer types and patient populations. Importantly, in preferred aspects, the present systems have shown to be highly reliable in classifying variants found in clinically actionable genes of mixed origin. Among others, in preferred aspects, the present systems can achieve consistent performance across a wide range of allele fractions by reducing dependency on VAF. By integrating fragmentomic data, the present systems can enhance the accuracy and reliability of variant classification. The present disclosure is not to be limited in scope by the specific embodiments described herein. Indeed, various modifications of the invention, in addition to those described herein, will become apparent to those skilled in the art from the foregoing description and the accompanying figures. Such modifications are intended to fall within the scope of the appended claims. 164810982.1 Docket: 348358.17201 Table 2. Performance of machine learning model in independent validation cohort by cancer type. Table 3. Performance of machine learning algorithm in independent validation cohort by gene or gene sets. *CHIP-associated genes include DNMT3A, TET2, ASXL1, PPM1D, TP53, JAK2, RUNX1, SF3B1, SRSF2, IDH1, IDH2, U2AF1, CBL, ATM, CHEK2;**Mixed origin genes include TP53, ATM, KMT2D, PPM1D, GNAS 164810982.1 Docket: 348358.17201 Code Availability All code that is necessary to run the machine learning model in R, as well as, additional statistical analyses are publicly available on . References 1. Parikh, A. R. et al. Liquid versus tissue biopsy for detecting acquired resistance and tumor heterogeneity in gastrointestinal cancers. Nat. Med.25, 1415–1421 (2019). 2. Campos-Carrillo, A. et al. Circulating tumor DNA as an early cancer detection tool. Pharmacol. Ther. 207, 107458 (2020). 3. Moding, E. J., Nabet, B. Y., Alizadeh, A. A. & Diehn, M. Detecting liquid remnants of solid tumors: circulating tumor DNA minimal residual disease. Cancer discovery 11, 2968–2986 (2021). 4. Yan, W., Zhang, A. & Powell, M. J. Genetic alteration and mutation profiling of circulating cell-free tumor DNA (cfDNA) for diagnosis and targeted therapy of gastrointestinal stromal tumors. Chinese journal of cancer 35, 1–8 (2016). 5. Sivapalan, L. et al. Liquid biopsy approaches to capture tumor evolution and clinical outcomes during cancer immunotherapy. Journal for immunotherapy of cancer 11 (2023). 6. Liu, J. et al. Biological background of the genomic variations of cf-DNA in healthy individuals. Ann. Oncol.30, 464–470 (2019). 7. Barbany, G. et al. Cell-free tumour DNA testing for early detection of cancer–a potential future tool. J. Intern. Med.286, 118–136 (2019). 8. Hu, Z., Chen, H., Long, Y., Li, P. & Gu, Y. The main sources of circulating cell-free DNA: Apoptosis, necrosis and active secretion. Critical reviews in oncology / hematology 157, 103166 (2021). 9. Chan, H. T. et al. Clinical significance of clonal hematopoiesis in the interpretation of blood liquid biopsy. Molecular oncology 14, 1719–1730 (2020). 10. Jaiswal, S. & Ebert, B. L. Clonal hematopoiesis in human aging and disease. Science 366, eaan4673 (2019). 11. Abbosh, C., Swanton, C. & Birkbak, N. J. Clonal haematopoiesis: a source of biological noise in cell- free DNA analyses. Ann. Oncol.30, 358–359 (2019). 12. Chan, H. T., Chin, Y. M., Nakamura, Y. & Low, S. Clonal Hematopoiesis in Liquid Biopsy: From Biological Noise to Valuable Clinical Implications. Cancers (Basel) 12, 2277 (2020). 13. Fairchild, L. et al. Clonal hematopoiesis detection in patients with cancer using cell-free DNA sequencing. Science translational medicine 15, eabm8729 (2023). 14. Razavi, P. et al. High-intensity sequencing reveals the sources of plasma circulating cell-free DNA variants. Nat Med 25, 1928–1937 (2019). 15. Hu, Y. et al. False-Positive Plasma Genotyping Due to Clonal Hematopoiesis. Clin. Cancer Res.24, 4437–4443 (2018). 16. Huang, F. et al. Chemotherapy-associated clonal hematopoiesis mutations should be taken seriously in plasma cell-free DNA KRAS / NRAS / BRAF genotyping for metastatic colorectal cancer. Clin. Biochem.92, 46–53 (2021). 17. Leal, A. et al. White blood cell and cell-free DNA analyses for detection of residual disease in gastric cancer. Nature communications 11, 525 (2020). 18. Chan, H. T. et al. Tumor-informed or tumor-agnostic circulating tumor DNA as a biomarker for risk of recurrence in resected colorectal cancer patients. Frontiers in Oncology 12, 1055968 (2023). 164810982.1 Docket: 348358.17201 19. Pereira, B. et al. Cell-free DNA captures tumor heterogeneity and driver alterations in rapid autopsies with pre-treated metastatic cancer. Nature communications 12, 3199 (2021). 20. van 't Erve, I. et al. Metastatic Colorectal Cancer Treatment Response Evaluation by Ultra-Deep Sequencing of Cell-Free DNA and Matched White Blood Cells. Clin. Cancer Res.29, 899–909 (2023). 21. American Cancer Society Facts and Figures 2020. 22. Stetson, D. et al. Orthogonal comparison of four plasma NGS tests with tumor suggests technical factors are a major source of assay discordance. JCO Precision Oncology 3, 1–9 (2019). 23. Weber, S. et al. Technical evaluation of commercial mutation analysis platforms and reference materials for liquid biopsy profiling. Cancers 12, 1588 (2020). 24. Chin, R. et al. Detection of solid tumor molecular residual disease (MRD) using circulating tumor DNA (ctDNA). Molecular diagnosis & therapy 23, 311–331 (2019). 25. Watson, C. J. et al. The evolutionary dynamics and fitness landscape of clonal hematopoiesis. Science 367, 1449–1454 (2020). 26. Park, S. J. & Bejar, R. Clonal hematopoiesis in cancer. Exp. Hematol.83, 105–112 (2020). 27. Ptashkin, R. N. et al. Prevalence of Clonal Hematopoiesis Mutations in Tumor-Only Clinical Genomic Profiling of Solid Tumors. JAMA Oncol.4, 1589–1593 (2018). 28. Yaung, S. J. et al. Clonal hematopoiesis in late-stage non–small-cell lung cancer and its impact on targeted panel next-generation sequencing. JCO precision oncology 4, 1271–1279 (2020). 29. Mack, T. et al. Cost-effective and scalable clonal hematopoiesis assay provides insight into clonal dynamics. The Journal of Molecular Diagnostics (2024). 30. Gutierrez-Rodrigues, F. et al. Utility of plasma cell-free DNA for de novo detection and quantification of clonal hematopoiesis. Haematologica 107, 1815 (2022). 31. Ding, S. C. & Lo, Y. D. Cell-free DNA fragmentomics in liquid biopsy. Diagnostics 12, 978 (2022). 32. Che, H. et al. Genomic origin, fragmentomics, and transcriptional properties of long cell-free DNA molecules in human plasma. Genome Res.34, 189–200 (2024). 33. Lo, Y. M. D., Han, D. S. C., Jiang, P. & Chiu, R. W. K. Epigenetics, fragmentomics, and topology of cell- free DNA in liquid biopsies. Science 372 (2021). 34. Tracing the Origin of Cell-Free DNA Molecules through Tissue-Specific Epigenetic Signatures - PMC. 35. Snyder, M. W., Kircher, M., Hill, A. J., Daza, R. M. & Shendure, J. Cell-free DNA comprises an in vivo nucleosome footprint that informs its tissues-of-origin. Cell 164, 57–68 (2016). 36. Mouliere, F. et al. Enhanced detection of circulating tumor DNA by fragment size analysis. Sci. Transl. Med.10, eaat4921. doi: 10.1126 / scitranslmed.aat4921 (2018). 37. Jiang, P. et al. Lengthening and shortening of plasma DNA in hepatocellular carcinoma patients. Proceedings of the National Academy of Sciences 112, E1317–E1325 (2015). 38. Cristiano, S. et al. Genome-wide cell-free DNA fragmentation in patients with cancer. Nature 570, 385–389 (2019). 39. Marass, F. et al. Fragment Size Analysis May Distinguish Clonal Hematopoiesis from Tumor-Derived Mutations in Cell-Free DNA. Clinical chemistry (Baltimore, Md.) 66, 616–618 (2020). 40. Reed, S. C., Croessmann, S. & Park, B. H. CHIP happens: clonal hematopoiesis of indeterminate potential and its relationship to solid tumors. Clinical Cancer Research 29, 1403–1411 (2023). 41. Crants, S. A. et al. Radiation Therapy and Subsequent Clonal Hematopoiesis: An Analysis of a Biorepository of 89,782 Patients. Int. J. Radiat. Oncol. Biol. Phys.114, e504 (2022). 42. Hsu, J. I. et al. PPM1D mutations drive clonal hematopoiesis in response to cytotoxic chemotherapy. Cell stem cell 23, 700–713. e6 (2018). 43. Olszewski, A. J. et al. Clonal haematopoiesis of indeterminate potential among cancer survivors exposed to myelotoxic chemotherapy. Br. J. Haematol.186, e31 (2019). 164810982.1 Docket: 348358.17201 44. Gimeno-Valiente, F. et al. Sequencing paired tumor DNA and white blood cells improves circulating tumor DNA tracking and detects pathogenic germline variants in localized colon cancer. ESMO open 8, 102051 (2023). 45. Rolfo, C. D. et al. Measurement of ctDNA tumor fraction identifies informative negative liquid biopsy results and informs value of tissue confirmation. Clinical Cancer Research 30, 2452–2460 (2024). 46. Chen, K. et al. Commercial ctDNA assays for minimal residual disease detection of solid tumors. Molecular diagnosis & therapy, 1–18 (2021). 47. Newman, A. M. et al. Integrated digital error suppression for improved detection of circulating tumor DNA. Nat. Biotechnol.34, 547–555 (2016). 48. Shen, W. et al. Detecting mosaic variants in patients with somatic overgrowth syndromes using cell- free circulating DNA and deep sequencing. J. Med. Genet.57, 794–796 (2020). 49. Reuss, J. E. et al. Neoadjuvant nivolumab plus ipilimumab in resectable non-small cell lung cancer; PMC7488786. J Immunother Cancer 8 (2020). 50. Reuss, J. E. et al. Neoadjuvant nivolumab in resectable non-small cell lung cancer: Extended follow- up and molecular markers of response. JCO 37, 8524 (2019). 51. Forde, P. M. et al. Neoadjuvant PD-1 Blockade in Resectable Lung Cancer. N. Engl. J. Med.378, 1976– 1986 (2018). 52. Hwang, M. W. et al. Early dynamics in peripheral blood immune cell subsets and ctDNA are predictive of outcome to immunotherapy. . (2021). 53. Phallen, J. et al. Early Noninvasive Detection of Response to Targeted Therapy in Non-Small Cell Lung Cancer. Cancer Res.79, 1204–1213 (2019). 54. Sivapalan, L. et al. Dynamics of Sequence and Structural Cell-Free DNA Landscapes in Small-Cell Lung Cancer. Clin. Cancer Res., OF1–OF14 (2023). 55. Phallen, J. et al. Direct detection of early-stage cancers using circulating tumor DNA; PMC6714979. Sci.Transl.Med.9 (2017). 56. Razavi, P. et al. High-intensity sequencing reveals the sources of plasma circulating cell-free DNA variants. Nat. Med.25, 1928–1937 (2019). 164810982.1

Claims

Docket: 348358.17201 CLAIMS What is claimed is:

1. A method of discriminating between clonal hematopoiesis-derived and tumor-derived variants in plasma cell-free DNA (cfDNA) of a subject, comprising: determining values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; determining a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and identifying a variant in the plasma cfDNA as a clonal hematopoiesis-derived variant if the classification score is equal to or greater than a threshold value.

2. A method of identifying a clonal hematopoiesis-derived variant in plasma cell-free DNA (cfDNA) of a subject, comprising: determining values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; determining a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and identifying a variant in the plasma cfDNA as a clonal hematopoiesis-derived variant if the classification score is equal to or greater than a threshold value.

3. The method of claim 1 or 2, wherein the fragment level features comprise differences in fragment length and cut points and endpoint motifs between mutant and wild type fragments at a locus.

4. The method of any one of claims 1 to 3, wherein the variant level features comprise variant allele frequency (VAF), scaled VAF, frequency of a variant in hematologic versus solid tumor malignancies in the Catalogue Of Somatic Mutations In Cancer (COSMIC), single base pair substitution, and estimate of driver function. 164810982.1Docket: 348358.17201 5. The method of any one of claims 1 to 4, wherein the patient level features comprise age.

6. The method of any one of claims 1 to 5, wherein the set of features comprise gene heme fraction, length cluster mutant, length MWU, length max delta S, and age.

7. The method of any one of claims 1 to 6, wherein the set of features further comprise endpoint motif down, endpoint location, plasma scaled VAF, length mean delta S, length cluster weight, variant COSMIC count, indel complex, mutant reads, or a combination thereof.

8. The method of any one of claims 1 to 7, wherein the classifier comprises a supervised or unsupervised Machine Learning or Deep Learning algorithm, Logistic Regression, Naive Bayes, Support Vector Machine, Decision Tree, Random Forest, Gradient Boosting, Regularizing Gradient Boosting, K-Nearest Neighbors, a continuous regression approach, Ridge Regression, Kernel Ridge Regression, Support Vector Regression, deep learning approach, Neural Networks, Convolutional Neural Network (CNNs), Recurrent Neural Networks (RNNs), Long Short Term Memory Networks (LSTMs), Generative Models, Generative Adversarial Networks (GANs), Deep Belief Networks (DBNs), Feedforward Neural Networks, Autoencoders, Radial Basis Function Networks (RBFNs), Multilayer Perceptrons (MLPs), Stochastic Neural Networks, or any combination thereof.

9. The method of claim 8, wherein the classifier comprises an XGBoost algorithm.

10. The method of any one of claims 1 to 9, wherein the threshold value is about 0.

5.

11. The method of any one of claims 1 to 10, wherein the method is capable of identifying the clonal hematopoiesis-derived variant with an accuracy of at least 80%.

12. The method of any one of claims 1 to 11, wherein the step of determining values of the set of features comprises extracting reads covering each variant locus from plasma-targeted next- generation sequencing (NGS) data. 164810982.1Docket: 348358.17201 13. The method of any one of claims 1 to 12, wherein the step of determining values of the set of features comprises calculating fragment summary statistics associated with fragment length, cut points, and endpoint motifs.

14. The method of any one of claims 1 to 13, wherein the step of determining values of the set of features comprises annotating variants using OpenCravat – COSMIC variant, COSMIC gene, CHASMMplus, and Cancer Gene Census (CGC).

15. The method of any one of claims 1 to 14, wherein the step of determining values of the set of features comprises calculating gene and variant heme fraction and variant counts using COSMIC counts.

16. The method of any one of claims 1 to 15, wherein the step of determining values of the set of features comprises determining driver status using CHASMMplus score, sequence ontology, and Cancer Gene Census (CGC) driver class.

17. The method of any one of claims 1 to 16, wherein the step of determining values of the set of features comprises calculating scaled plasma variant allele frequency (VAF).

18. The method of any one of claims 1 to 17, wherein the subject has no patient-matched white blood cell (WBC) sequencing data available.

19. The method of any one of claims 1 to 18, wherein the plasma cfDNA comprises tumor- circulating DNA (ctDNA).

20. The method of any one of claims 1 to claims 1 to 19, comprising obtaining from a sample from the subject comprising the plasma cfDNA.

21. The method of claim 20, wherein the sample comprises a biological sample that comprises a plasma sample, a blood sample, or a tissue sample. 164810982.1Docket: 348358.17201 22. The method of any one of claims 1 to 21, wherein the subject has cancer.

23. The method of any one of claims 1 to 22, wherein the subject has a solid tumor.

24. A system of discriminating between clonal hematopoiesis-derived and tumor-derived variants in plasma cell-free DNA (cfDNA) of a subject, comprising one or more processors configured to: determine values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; determine a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and identify a variant in the plasma cfDNA as a clonal hematopoiesis-derived variant if the classification score is equal to or greater than a threshold value.

25. A system of identifying a clonal hematopoiesis-derived variant in plasma cell-free DNA (cfDNA) of a subject, comprising one or more processors configured to: determine values of a set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; determine a classification score by a classifier based on the values of the set of features selected from fragment-level features, variant-level features, and patient-level features associated with the plasma cfDNA; and identify a variant in the plasma cfDNA as a clonal hematopoiesis-derived variant if the classification score is equal to or greater than a threshold value. 164810982.1

Images