Genome-wide repeat landscapes in cancer and cell-free DNA

Abstract

Compositions and methods are provided for analyzing repeat sequences in genomes are utilized in methods for large-scale analyses of these regions for characterization, detection and monitoring of human cancer.

Description

ATTORNEY DOCKET NO.: 348358.16902 GENOME-WIDE REPEAT LANDSCAPES IN CANCER AND CELL-FREE DNA The present application claims the benefit of U.S. provisional application no.63 / 532,642 filed August 14, 2023, which is incorporated herein by reference in its entirety. STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH

[0001] This invention was made with government support under grants CA006973, CA062924, CA121113, CA233259, CA271896 and GM136577 awarded by the National Institutes of Health. The government has certain rights in the invention. FIELD

[0002] Compositions for analyzing repeat sequences in genomes are utilized in methods for large-scale analyses of these regions for characterization, detection and monitoring of human cancer. BACKGROUND

[0003] Genomic repeats comprise more than half the human genome and include a diverse set of elements that vary widely between individuals and exert key influences on genome structure and function1,2. Due to technical limitations of short-read alignment and a reliance on incomplete genome assemblies, repeats have historically been neglected3. Such repetitive sequences include tandem repeats such as human satellites, and interspersed repeats including structural RNAs, long interspersed nuclear elements (LINEs), short interspersed nuclear elements (SINEs), long terminal repeats (LTRs) and other transposable elements. The recent completion of a telomere-to-telomere genome has added nearly 200 Mb to the previous reference genome, revealed the genomic and epigenomic states of repeats, and revitalized study of these integral genomic regions4–6. SUMMARY

[0004] An genome-wide approach for analyzing repeat landscapes in next generation sequencing is provided. The approach, termed herein, ARTEMIS, Analysis of RepeaT EleMents in dISease, can assess over thousands of individual repeat types that occur genome-wide and span a plurality of subfamilies comprising a plurality of families (Satellites, RNA elements, transposable elements, LINEs, SINEs, LTRs). 148264593

[0005] In certain aspects, the read length may be for example less than 800, 700, 600, 500, 400 or 300 bp. In certain aspects, the read length may be for example 50 to 400 bp, or 50 to 300 or 250 bp, or 75 to 300 bp or 75 to 250 bp.

[0006] In certain aspects, the approach and methods can be alignment free.

[0007] In an aspect, a method of identifying repeat element types is provided, comprising extracting repeat sequences and coordinates from known repeat element types; identifying nucleic acid sequences (kmers) in nucleic acid sequences including genomic or cell free DNA; selecting kmers occurring in a single repeat type and identifying unique kmers of repeat element types; wherein the unique kmers identify one or a plurality of repeat element types.

[0008] In an aspect, a method of identifying repeat element types is provided, comprising extracting repeat sequences and coordinates from known repeat element types; identifying nucleic acid sequences (kmers) in an RNA sequence; selecting kmers occurring in a single repeat type and identifying unique kmers of repeat element types; wherein the unique kmers identify one or a plurality of repeat element types.

[0009] In an aspect, a method of identifying repeat element types is provided, comprising a) extracting repeat sequences and genomic coordinates from known repeat element types; b) selecting nucleic acid sequences (kmers) occurring in a single repeat type and identifying unique kmers of repeat element types, wherein the unique kmers identify one or a plurality of repeat element types; and c) identifying the kmers in nucleic acid sequences including genomic or cell free DNA. In certain aspect, the type and / or frequency of kmers in the nucleic acid sequences including genomic or cell free DNA may be identified.

[0010] In an aspect, a method of identifying repeat element types is provided, comprising a) extracting repeat sequences and genomic coordinates from known repeat element types; b) selecting nucleic acid sequences (kmers) occurring in a single repeat type and identifying unique kmers of repeat element types, wherein the unique kmers identify one or a plurality of repeat element types; and c) identifying the kmers in genomic or cell free DNA. In certain aspect, the type and / or frequency of kmers in genomic or cell free DNA may be identified.

[0011] In certain embodiments, repeat element types are excluded from families comprising low complexity, unknown, simple repeats or combinations thereof.2 161564170.1

[0012] In certain embodiments, elements from each family comprising tRNAs, srpRNAs, snRNAs, scRNAs, rRNAs, RNA elements, DNA, retroposons, retrotransposons, or combinations thereof, are aggregated.

[0013] In certain embodiments, the families comprise long interspersed nuclear elements (LINEs), short interspersed nuclear elements (SINEs), long terminal repeats (LTRs), satellites, transposable elements, RNA elements or combinations thereof. In certain embodiments, the kmers occurring in single repeat type elements and not in non-repeat regions are selected.

[0014] In certain embodiments, the kmer comprises up to or about 5, 8, 10 to 40, 50, 60, 70, 80, 90, 100, 120, 140, 160, 180 or 200 nucleotides In certain embodiments, the kmer comprises 10 to 40, 50, , 60, 70, 80, 90 or 100 nucleotides. In certain embodiments, the kmer comprises 15 to 35 nucleotides. In certain embodiments, the kmer comprises 20 to 30 nucleotides. In certain embodiments, the kmer comprises 22 to 28 nucleotides. In certain embodiments, the kmer comprises 23 to 26 nucleotides. In certain embodiments, the kmer comprises about 24 nucleotides.

[0015] In certain embodiments, kmer repeat landscapes are generated, each unique kmer and reverse complement thereof, is counted and kmer counts for each repeat type element are aggregated. In certain embodiments, regions of genes comprising a plurality of transcripts are counted to identify kmers occurring in each gene.

[0016] In certain embodiments, genes are ranked by total number of kmers and a corrected kmer density. In certain embodiments, the corrected kmer density comprises the number of types occurring in a gene divided by kmer density. In certain embodiments, the kmer density is total kmers per Mb.

[0017] In certain embodiments, repeat element types are increased in genes associated with a cancer diagnosis as compared to a non-cancer control. In certain embodiments, kmer-defined repeat element type families are altered genome-wide as compared to a normal control. In certain embodiments, the alterations in the genome as compared to a non-cancer genome comprises focal amplifications, deletions, copy-number changes, rearrangements, changes in repeat elements or combinations thereof.

[0018] In another aspect, a method of diagnosing and treating cancer in a subject is provided and suitably comprises generating kmer repeat landscapes from a subject’s biological sample, assaying for a change in a repeat element comprising selecting normal samples, calculating a ratio3 161564170.1of the kmer repeat landscapes between the samples to diagnose the subject as having cancer and, treating the subject diagnosed with cancer.

[0019] In certain embodiments, the subject is diagnosed as having cancer if the kmer repeat landscape between two samples comprises a tumor / normal ratio less than the 1stpercentile or above the 99thpercentile of a normal / normal ratio, where normal / normal ratios are obtained from subsampling in silico normal samples wherein each subsample comprises about half coverage of an original sample . In certain embodiments, changes in the tumor / normal and normal / normal ratios are correlated with one or more metrics of genomic instability.

[0020] In certain embodiments, the one or more metrics of genomic instability comprise entropy, non-modal ploidy fraction, non-diploid fraction, loss of heterozygosity fraction, breakpoint count, tumor mutation burden, ploidy, modal ploidy or combinations thereof.

[0021] In certain embodiments, the method further comprises repeat element types with overlapping kmers and assaying differences in aggregate kmer counts for the repeat element types between tumors with and without focal amplification.

[0022] In certain embodiments, the kmer repat landscapes are inputted into a machine- learning or artificial intelligence program to generate a cross-validated or externally validated score for distinguishing tumor samples from normal samples.

[0023] In certain embodiments, the biological sample comprises genomic DNA, cell free DNA (cfDNA) or combinations thereof. In certain embodiments, epigenetic features in cfDNA fragments, differences in fragment length and coverage for fragments in regions of histone marks are evaluated. In certain embodiments, localization of histone marks within repeat element types and densities of each histone mark are evaluated. In certain embodiments, ratios of observed kmer counts to expected kmer count variations between repeat element types are evaluated based on histone mark density.

[0024] In certain embodiments, cancer treatment or therapy comprises surgery, adjuvant chemotherapy, neoadjuvant chemotherapy, radiation therapy, hormone therapy, cytotoxic therapy, immunotherapy, adoptive T cell therapy, targeted therapy, and combinations thereof.

[0025] In another aspect, a method of detecting and monitoring of cancer progression in a subject is provided and may suitably comprises: assaying a plurality of features of a subject’s sample to produce a kmer repeat landscape; partitioning further features into families comprising long interspersed nuclear elements (LINEs), short interspersed nuclear elements (SINEs), long terminal4 161564170.1repeats (LTRs), satellites, transposable elements, RNA elements or combinations thereof, defining a plurality of megabase (Mb) bins by epigenetic analyses; calculating aligned fragment coverage of each bin, wherein, bins with mappability of less than 0.9 or GC content of less than 0.3 are excluded; training of regression models using repeat landscapes and epigenetic profiles, and ensembling the scores together with the regression model to produce a combined score. In certain embodiments, the method further comprises incorporating fragmentation and a score obtained by one or more software models, machine learning models, artificial intelligence or combinations thereof.

[0026] In some aspects, methods and materials described herein also can include machine learning.

[0027] In certain aspects, a method of determining whether a subject is responding to treatment is provide and may suitably comprise any one or more of the methods embodied herein. In certain embodiments, the treatment comprises surgery, adjuvant chemotherapy, neoadjuvant chemotherapy, radiation therapy, hormone therapy, cytotoxic therapy, immunotherapy, adoptive T cell therapy, targeted therapy, and combinations thereof.

[0028] Definitions

[0029] Unless otherwise defined, all terms (including 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. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0030] As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. Furthermore, to the extent that the terms “including”, “includes”, “having”, “has”, “with”, or variants thereof are used in either the detailed description and / or the claims, such terms are intended to be inclusive in a manner similar to the term “comprising.”

[0031] The term “about” or “approximately” means within an acceptable error range for the particular value as determined by one of ordinary skill in the art, which will depend in part on how the value is measured or determined, i.e., the limitations of the measurement system. For example, “about” can mean within 1 or more than 1 standard deviation, per the practice in the art. Alternatively, “about” can mean a range of up to 20%, up to 10%, up to 5%, or up to 1% of a given5 161564170.1value or range. Alternatively, particularly with respect to biological systems or processes, the term can mean within an order of magnitude within 5-fold, and also within 2-fold, of a value. Where particular values are described in the application and claims, unless otherwise stated the term “about” meaning within an acceptable error range for the particular value should be assumed.

[0032] The term “cancer” as used herein is meant, a disease, condition, trait, genotype or phenotype characterized by unregulated cell growth or replication as is known in the art; including liver cancer (including hepatocellular carcinoma (HCC)), lung cancer (including non-small cell lung carcinoma), gastric cancer, colorectal cancer, as well as, for example, leukemias, e.g., acute myelogenous leukemia (AML), chronic myelogenous leukemia (CML), acute lymphocytic leukemia (ALL), and chronic lymphocytic leukemia, AIDS related cancers such as Kaposi's sarcoma; breast cancers; bone cancers such as Osteosarcoma, Chondrosarcomas, Ewing's sarcoma, Fibrosarcomas, Giant cell tumors, Adamantinomas, and Chordomas; Brain cancers such as Meningiomas, Glioblastomas, Lower- Grade Astrocytomas, Oligodendrocytomas, Pituitary Tumors, Schwannomas, and Metastatic brain cancers; cancers of the head and neck including various lymphomas such as mantle cell lymphoma, non-Hodgkins lymphoma, adenoma, squamous cell carcinoma, laryngeal carcinoma, gallbladder and bile duct cancers, cancers of the retina such as retinoblastoma, cancers of the esophagus, gastric cancers, multiple myeloma, ovarian cancer, uterine cancer, thyroid cancer, testicular cancer, endometrial cancer, melanoma, bladder cancer, prostate cancer, pancreatic cancer, sarcomas, Wilms' tumor, cervical cancer, head and neck cancer, skin cancers, nasopharyngeal carcinoma, liposarcoma, epithelial carcinoma, renal cell carcinoma, gallbladder adeno carcinoma, parotid adenocarcinoma, endometrial sarcoma, multidrug resistant cancers; and proliferative diseases and conditions, such as neovascularization associated with tumor angiogenesis.

[0033] The term “cell free nucleic acid,” “cell free DNA,” or “cfDNA” refers to nucleic acid fragments that circulate in an individual's body (e.g., bloodstream) and originate from one or more healthy cells and / or from one or more cancer cells. Additionally, cfDNA may come from other sources such as viruses, fetuses, etc.

[0034] The term “cfDNA sequence coverage” refers to the average number of cfDNA molecules overlapping a specific position.6 161564170.1

[0035] The term “circulating tumor DNA” or “ctDNA” refers to nucleic acid fragments that originate from tumor cells or other types of cancer cells, which may be released into an individual's bloodstream as result of biological processes such as apoptosis or necrosis of dying cells or actively released by viable tumor cells.

[0036] As used herein, the terms “comprising,” “comprise” or “comprised,” and variations thereof, in reference to defined or described elements of an item, composition, apparatus, method, process, system, etc. are meant to be inclusive or open ended, permitting additional elements, thereby indicating that the defined or described item, composition, apparatus, method, process, system, etc. includes those specified elements--or, as appropriate, equivalents thereof--and that other elements can be included and still fall within the scope / definition of the defined item, composition, apparatus, method, process, system, etc.

[0037] “Diagnostic” or “diagnosed” means identifying the presence or nature of a pathologic condition. Diagnostic methods differ in their sensitivity and specificity. The “sensitivity” of a diagnostic assay is the percentage of diseased individuals who test positive (percent of “true positives”). Diseased individuals not detected by the assay are “false negatives.” Subjects who are not diseased and who test negative in the assay, are termed “true negatives.” The “specificity” of a diagnostic assay is 1 minus the false positive rate, where the “false positive” rate is defined as the proportion of those without the disease who test positive. While a particular diagnostic method may not provide a definitive diagnosis of a condition, it suffices if the method provides a positive indication that aids in diagnosis.

[0038] An “effective amount” as used herein, means an amount which provides a therapeutic or prophylactic benefit.

[0039] As used herein, the terms “fragmentation profile,” “position dependent differences in fragmentation patterns,” and “differences in fragment size and coverage in a position dependent manner across the genome” are equivalent and can be used interchangeably. In some embodiments, determining a cfDNA fragmentation profile in a mammal can be used for identifying a mammal as having cancer. For example, cfDNA fragments obtained from a mammal (e.g., from a sample obtained from a mammal) can be subjected to low coverage whole- genome sequencing, and the sequenced fragments can be mapped to the genome (e.g., in non- overlapping windows) and assessed to determine a cfDNA fragmentation profile. As described herein, a cfDNA fragmentation profile of7 161564170.1a mammal having cancer is more heterogeneous (e.g., in fragment lengths) than a cfDNA fragmentation profile of a healthy mammal (e.g., a mammal not having cancer). As such, this disclosure also provides methods and materials for assessing, monitoring, and / or treating mammals (e.g., humans) having, or suspected of having, cancer. In some embodiments, this document provides methods and materials for identifying a mammal as having cancer. For example, a sample (e.g., a blood sample) obtained from a mammal can be assessed to determine the presence and, optionally, the tissue of origin of the cancer in the mammal based, at least in part, on the cfDNA fragmentation profile of the mammal. In some embodiments, methods and materials for monitoring a mammal as having cancer are provided. For example, a sample (e.g., a blood sample) obtained from a mammal can be assessed to determine the presence of the cancer in the mammal based, at least in part, on the cfDNA fragmentation profile of the mammal. In some embodiments, methods and materials for identifying a mammal as having cancer and administering one or more cancer treatments to the mammal to treat the mammal are provided. For example, a sample (e.g., a blood sample) obtained from a mammal can be assessed to determine if the mammal has cancer based, at least in part, on the cfDNA fragmentation profile of the mammal, and one or more cancer treatments can be administered to the mammal.

[0040] The term “ensemble learning” refers to algorithms that combine the predictions from two or more models. “Ensemble methods” are a machine learning technique that combines several base models in order to produce one optimal predictive model. Widely used ensemble learning strategies include bagging, stacking, boosting. Bootstrap aggregation, or bagging for short, is an ensemble learning method that seeks a diverse group of ensemble members by varying the training data. The name Bagging came from the abbreviation of Bootstrap AGGregatING. As the name implies, the two key ingredients of Bagging are bootstrap and aggregation.

[0041] Bootstrap aggregation (bagging) involves training an ensemble on bootstrapped data sets. A bootstrapped set is created by selecting from original training data set with replacement. Thus, a bootstrap set may contain a given example zero, one, or multiple times. Ensemble members can also have limits on the features (e.g., nodes of a decision tree), to encourage exploring of diverse features. The variance of local information in the bootstrap sets and feature considerations promote diversity in the ensemble and can strengthen the ensemble. To reduce overfitting, a member can be validated using the out-of-bag set (the examples that are not in its bootstrap set).8 161564170.1

[0042] Stacking is a general procedure where a learner is trained to combine the individual learners. Here, the individual learners are called the first-level learners, while the combiner is called the second-level learner, or meta-learner. Stacking (sometimes called stacked generalization) involves training a model to combine the predictions of several other learning algorithms. First, all of the other algorithms are trained using the available data, then a combiner algorithm (final estimator) is trained to make a final prediction using all the predictions of the other algorithms (base estimators) as additional inputs or using cross-validated predictions from the base estimators which can prevent overfitting. If an arbitrary combiner algorithm is used, then stacking can theoretically represent any of the ensemble techniques, although, in practice, a logistic regression model is often used as the combiner. Stacking typically yields performance better than any single one of the trained models. It has been successfully used on both supervised learning tasks (regression, classification and distance learning) and unsupervised learning (density estimation). The key elements of stacking are as follows: unchanged training dataset, different machine learning algorithms for each ensemble member, machine learning model to learn how to best combine predictions.

[0043] Boosting Ensemble Learning is an ensemble method that seeks to change the training data to focus attention on examples that previous fit models on the training dataset have gotten wrong. In boosting, the training dataset for each subsequent classifier increasingly focuses on instances misclassified by previously generated classifiers. The key property of boosting ensembles is the idea of correcting prediction errors. The models are fit and added to the ensemble sequentially such that the second model attempts to correct the predictions of the first model, the third corrects the second model, and so on. This typically involves the use of very simple decision trees that only make a single or a few decisions, referred to in boosting as weak learners. The predictions of the weak learners are combined using simple voting or averaging, although the contributions are weighed proportional to their performance or capability. The objective is to develop a so-called “strong-learner” from many purpose-built “weak-learners.” Typically, the training dataset is left unchanged and instead, the learning algorithm is modified to pay more or less attention to specific examples (rows of data) based on whether they have been predicted correctly or incorrectly by previously added ensemble members. For example, the rows of data can be weighed to indicate the amount of focus a learning algorithm must give while learning the model. The key elements of boosting are as follows: Bias training data toward those examples that are hard to predict, Iteratively9 161564170.1add ensemble members to correct predictions of prior models, Combine predictions using a weighted average of models.

[0044] The term “genomic nucleic acid,” or “genomic DNA,” refers to nucleic acid including chromosomal DNA that originates from one or more healthy (e.g., non-tumor) cells or tumor cells. In various embodiments, genomic DNA can be extracted from a cell derived from a blood cell lineage, such as a white blood cell (WBC).

[0045] As used herein, a “kmer” or, “k-mer” is a sequence of DNA of k consecutive nucleotides (in the forward or reverse strand of the DNA molecule), where k is a natural integer higher than 1. Any sequence of length L will contain L - k + 1 k-mers. In certain aspects, a kmer also can be described as subsequence of a polynucleotide of length k, including short sequences such as less than 200, 150, 100, 90, 80, 70, 60 or 40 bases.

[0046] As used herein, “LINEs” (long interspersed nuclear elements) are longer non-LTR retrotransposons. They are widespread in the genomes of eukaryotes. They usually make up 21.1% of the human genome. Each LINE is around 7000 base pairs long. LINEs can transcribe into mRNA and translate into a protein that can function as a reverse transcriptase enzyme. This reverse transcriptase produces DNA copies of the LINEs RNA. These DNA copies can be integrated into the genome at a new site. The human genome has only one abundant LINE called LINE-1. LINE-1 element is around 6000 base pairs long. There are around 100,000 truncated LINE-1 elements in the human genome. Random mutation can occur in LINEs. Due to random mutations, the LINEs can degenerate. They are no longer transcribed or translated. Furthermore, LINEs are grouped into five main groups such as L1, RTE, R2, I and jockey. These five groups further subdivide into another 28 clades. LINEs are normally propagated through a mechanism called target primed reverse transcription mechanism (TPRT). Insertion of LINEs causes human diseases like hemophilia A, cancer, Mendelian disorders, etc. Hypomethylation of LINEs also triggers certain types of cancer.

[0047] “Optional” or “optionally” means that the subsequently described event or circumstance can or cannot occur, and that the description includes instances where the event or circumstance occurs and instances where it does not.

[0048] As used in this specification and the appended claims, the term “or” is generally employed in its sense including “and / or” unless the content clearly dictates otherwise.10 161564170.1

[0049] “Parenteral” administration of an immunogenic composition includes, e.g., subcutaneous (s.c.), intravenous (i.v.), intramuscular (i.m.), or intrasternal injection, or infusion techniques.

[0050] The terms “patient” or “individual” or “subject” are used interchangeably herein, and refers to a mammalian subject to be treated, with human patients being preferred. In some embodiments, the methods of the disclosure find use in experimental animals, in veterinary application, and in the development of animal models for disease, including, but not limited to, rodents including mice, rats, and hamsters, and primates.

[0051] The term “reference genome” as used herein may refer to a digital or previously identified nucleic acid sequence database, assembled as a representative example of a species or subject. Reference genomes may be assembled from the nucleic acid sequences from multiple subjects, sample or organisms and does not necessarily represent the nucleic acid makeup of a single person. Reference genomes may be used to for mapping of sequencing reads from a sample to chromosomal positions. For example, a reference genome used for human subjects as well as many other organisms is found at the National Center for Biotechnology Information at ncbi.nlm.nih.gov.

[0052] The term “read segment” or “read” refers to any nucleotide sequences including sequence reads obtained from an individual and / or nucleotide sequences derived from the initial sequence read from a sample obtained from an individual.

[0053] The terms “sample,” “patient sample,” “biological sample,” and the like, encompass a variety of sample types obtained from a patient, individual, or subject and can be used in a diagnostic, prognostic and / or monitoring assay. The patient sample may be obtained from a healthy subject, a diseased patient, or a patient with lung cancer. In certain embodiments, a sample that is “provided” can be obtained by the person (or machine) conducting the assay, or it can have been obtained by another, and transferred to the person (or machine) carrying out the assay. Moreover, a sample obtained from a patient can be divided and only a portion may be used for diagnosis. Further, the sample, or a portion thereof, can be stored under conditions to maintain sample for later analysis. The definition specifically encompasses blood and other liquid samples of biological origin (including, but not limited to, peripheral blood, serum, plasma, cord blood, amniotic fluid, cerebrospinal fluid, urine, saliva, stool and synovial fluid), solid tissue samples such as a biopsy specimen or tissue cultures or cells derived therefrom and the progeny thereof. In certain11 161564170.1embodiment, a sample comprises cerebrospinal fluid. In a specific embodiment, a sample comprises a blood sample. In another embodiment, a sample comprises a plasma sample. In yet another embodiment, a serum sample is used. The definition of “sample” also includes samples that have been manipulated in any way after their procurement, such as by centrifugation, filtration, precipitation, dialysis, chromatography, treatment with reagents, washed, or enriched for certain cell populations. The terms further encompass a clinical sample, and also include cells in culture, cell supernatants, tissue samples, organs, and the like. Samples may also comprise fresh-frozen and / or formalin-fixed, paraffin-embedded tissue blocks, such as blocks prepared from clinical or pathological biopsies, prepared for pathological analysis or study by immunohistochemistry.

[0054] The term “sequence reads” refers to nucleotide sequences read from a sample obtained from an individual. Sequence reads can be obtained through various methods known in the art.

[0055] As used herein, “SINEs” (short interspersed nuclear elements) are a type of much shorter non-LTR retrotransposons. They are about 100 to 700 base pairs in length. SINEs are also DNA elements that amplify themselves throughout eukaryotic genomes through RNA intermediates. SINEs make up about 13% of the mammalian genome. The internal regions of SINEs originate from tRNA. It remains highly conserved. They are often present in many species of vertebrates and invertebrates. The copy number variation and mutations in the SINEs can be incorporated to construct the phylogeny-based classification of species. SINEs can be grouped into three main types: CORE-SINEs, V-SINEs, and AmnSINEs. Alu element is the most common SINE in primates. Moreover, there are more than 50 human diseases associated with the insertion of SINEs. When they insert within or near exons, they can cause improper splicing or change the reading frame. This leads to disease phenotypes such breast cancer, colon cancer, leukemia, hemophilia, cystic fibrosis, colon cancer, Dent’s disease, neurofibromatosis, etc.

[0056] As used herein, a “therapeutically effective” amount of a compound or agent (i.e., an effective dosage) means an amount sufficient to produce a therapeutically (e.g., clinically) desirable result. The compositions can be administered from one or more times per day to one or more times per week, including once every other day. The skilled artisan will appreciate that certain factors can influence the dosage and timing required to effectively treat a subject, including but not limited to the severity of the disease or disorder, previous treatments, the general health and / or age of the12 161564170.1subject, and other diseases present. Moreover, treatment of a subject with a therapeutically effective amount of the compounds of the disclosure can include a single treatment or a series of treatments.

[0057] As used herein, the terms “treat,” treating,” “treatment,” and the like refer to reducing or ameliorating a disorder and / or symptoms associated therewith. It will be appreciated that, although not precluded, treating a disorder or condition does not require that the disorder, condition or symptoms associated therewith be completely eliminated.

[0058] Genes: All genes, gene names, and gene products disclosed herein are intended to correspond to homologs from any species for which the compositions and methods disclosed herein are applicable. It is understood that when a gene or gene product from a particular species is disclosed, this disclosure is intended to be exemplary only, and is not to be interpreted as a limitation unless the context in which it appears clearly indicates. Thus, for example, for the genes or gene products disclosed herein, are intended to encompass homologous and / or orthologous genes and gene products from other species.

[0059] Ranges: throughout this disclosure, various aspects of the disclosure can be presented in a range format. It should be understood that the description in range format is merely for convenience and brevity and should not be construed as an inflexible limitation on the scope of the disclosure. Accordingly, the description of a range should be considered to have specifically disclosed all the possible subranges as well as individual numerical values within that range. For example, description of a range such as from 1 to 6 should be considered to have specifically disclosed subranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6 etc., as well as individual numbers within that range, for example, 1, 2, 2.7, 3, 4, 5, 5.3, and 6. This applies regardless of the breadth of the range.

[0060] Any compositions or methods provided herein can be combined with one or more of any of the other compositions and methods provided herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] 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.

[0062] FIG. 1 is a diagram of an embodiment of the ARTEMIS method. De novo identification of kmers revealed ~1.1 billion unique kmers spanning 6 families and 1280 distinct13 161564170.1repeat elements. The kmer repeat landscape is defined as the sum of the counts of all kmers comprising each repeat type identified in all sequence reads, normalized by coverage. These landscapes are used in machine learning to generate an ARTEMIS score for disease prediction.

[0063] FIG. 2 (includes FIGS. 2A-2E): shows kmer repeat landscapes across human cancers reveal widespread differences from normal tissues The heatmap (FIG.2A) shows the ratio of kmer repeat landscapes for each PCAWG tumor as compared to its matched normal, revealing high numbers of tumor-specific changes which can be correlated to genomic instability metrics (n=469 tumor / normal pairs representing all PCAWG samples with genomic instability metrics available). Each PCAWG tumor is listed along the y-axis, and each individual repeat element type is along the x-axis. Ratios greater than one (red) indicate an increase of the element in the tumor, whereas ratios less than one (blue) indicate a decrease in the tumor. The majority of such identified changes are in elements (820 of 1280) with no prior evidence for changes in cancer, as shown in yellow in the evidence bar along the x-axis. FIG.2B: The plot shows elements from all six repeat element families ordered by the Benjamini-Hochberg corrected p-value of the Wilcoxon signed-rank test comparing the overlap of repeat elements with tumor-specific structural breakpoints versus their overlap with randomly selected genomic regions. Filled circles indicate elements newly implicated in cancer through this study while open circles indicate elements with prior evidence for involvement in cancer. Red circles indicate elements depleted of breakpoints, and yellow circles indicate elements enriched for breakpoints. FIG. 2C: Boxplots show the distribution of Tumor:Normal kmer count ratios for repeat element types overlapping the ERBB2 region (1 Mb) in PCAWG breast cancers (n=91 tumor / normal pairs). Ratios for each patient for all elements with >0.5% of kmers found in the region are shown (left), and the Benjamini-Hochberg corrected p-value from the Wilcoxon signed-rank test is plotted for each comparison, with points in red indicating p<0.05 (right). Element names in bold indicate those newly implicated in cancer through this study. FIG. 2D: Boxplots show the ratios of Tumor:Normal kmer counts for kmers occurring within LINE-1 mediated deletions in PCAWG lung tumors containing at least one LINE-1 mediated deletion (n=5, Supplemental Data File 2). FIG.2E: Kaplan-Meir plots of Overall Survival and Progression-free Survival of PCAWG tumors of AJCC Stages III or IV (n=167) stratified into two groups based on predicted ARTEMIS Scores. The group shown in blue had ARTEMIS scores below the median value, and the group shown in red had ARTEMIS scores above the median value.14 161564170.1

[0064] FIGS. 3A-3B: Kmer repeat landscapes capture tumor-specific changes in the plasma. FIG.3A: Top panel, Each bar plot shows for a given Human Satellite 2 or 3 element type, the percentage of its kmer occurrences found on chrY (dark blue) and on all other chromosomes (light blue) in the chm13 reference. Bottom panel, in individuals without cancer (n=158), the distribution of coverage-normalized kmer counts in cfDNA for these satellite types in males (n=87) and females (n=71). P-values for the Wilcoxon signed-rank test are shown at the top of each plot. FIG.3B: Kmer counts for PCAWG Tissue (top panel, n= 54 liver, n=48 lung squamous and n=38 lung adenocarcinoma tumor / normal pairs) and plasma cfDNA (bottom panel; n=75 patients with liver cancer and n=133 patients without cancer; n=29 patients with lung squamous cell cancer and n=158 patients without cancer; n=62 patients with lung adenocarcinoma and n=158 patients without cancer). The top five features with significant differences in both tissue and plasma, and at least 1000 expected kmers per million aligned reads are shown for each cancer type as separate plots. P- values are shown at the top of each plot and were calculated by the Wilcoxon signed-rank test.

[0065] FIG.4 demonstrates the impact of epigenetic state on repeat element representation in cfDNA. Top panel (Panel 1), summary of peaks per Mb of each chromatin state for each histone type. The peak density is scaled within each CHIP-Seq experiment to account for different numbers of peaks in each experiment. Panel 2, the proportion of histone peaks of each type in each of 1280 repeat elements organized in six families shows variation in epigenetic state between repeat types. Panel 3, in plasma from patients without cancer in the LUCAS Cohort, the distribution of aligned fragment sizes for fragments overlapping each histone mark and all fragments shows an increase in shorter fragments within regions of histone marks associated with activation. The line is the median and the shading indicates + / - 1 SD, plotted as a difference in distributions. Panel 4, in plasma from patients without cancer in the LUCAS cohort, plots of coverage genome-wide vs. within regions of each histone mark shows a decrease in coverage within regions of histone marks associated with activation. The x-axis represents the log average coverage and the y-axis represents the log differences in count. Panel 5, in plasma from patients without cancer in the LUCAS cohort, the ratio of average observed to expected kmer counts for the features in the top and bottom deciles of histone mark density shows reduced representation of elements with higher amounts of activating histone marks.

[0066] FIGS.5A-5C demonstrate ARTEMIS and ARTEMIS-DELFI for early detection of lung cancer using cfDNA. (FIG.5A) Distribution of ARTEMIS and joint ARTEMIS-DELFI scores15 161564170.1for cancer and non-cancer patients in the cross-validated LUCAS cohort show lower scores in non- cancer patients than cancer patients. (FIG.5B) ROC analyses of ARTEMIS and ARTEMIS-DELFI scores show high performance for classifying individuals with and without lung cancer in the full LUCAS cohort, and in subgroups by cancer stage. (FIG.5C) The sensitivity and specificity achieved by ARTEMIS and ARTEMIS-DELFI in the external validation cohort at score thresholds selected to achieve 50% - 80% specificity in the cross-validated cohort demonstrates the generalizability of these approaches for early detection in a high-risk population.

[0067] FIG.6 is a detailed schematic of ARTEMIS Method.

[0068] FIG. 7 shows the relationship between size of repeat element genome wide and number of unique kmers identified. The x-axis shows the number of unique kmers identified in the reference genome and the y-axis shows the size of the repeat element genome wide. Each point is one of 1280 repeat element types, grouped into 6 families.

[0069] FIGS.8A-8D are a characterization of genome-wide ARTEMIS kmers and enrichment in cancer-related genes. (FIG.8A) Number of chromosomes containing at least one kmer for each of 1280 repeat element features. Each boxplot represents this distribution for a subfamily of elements. The color scale represents the number (log10) of unique kmers that define this element. (FIG. 8B) the distribution of genome wide occurrences of the 1.1 billion kmers defining these elements reveals that the vast majority of kmers occur once in the genome. (FIG.8C) The number of repeat element features (of 1280) found on each chromosome demonstrates their genome-wide distribution. (FIG.8D) Leading edge analysis for gene set enrichment (GSEA) of the COSMIC Cancer Gene Census as ranked by total kmers per gene (top left panel) and kmer density adjusted for the number of unique types found per gene (top right panel) shows the enrichment of repeat kmers among cancer related genes. As a control, a random selection of the same number of genes is shown in the bottom panels. On each panel, the gray lines represent 100 permutations of the gene ranks as a simulated null distribution while the blue line represents the analyzed gene set.

[0070] FIG.9 is a gene set enrichment analysis for total number of kmers and kmer density. All KEGG gene sets enriched with FDR < 0.05 are shown, including many that are related to cancer cell processes or pathways.

[0071] FIG.10 is a simulation of kmer counts in short-read next generation sequencing.50 million 100bp paired-end reads were simulated from the chm13 reference genome incorporating a realistic sequencing error rate. Reads that truly originated from a repeat region were selected and16 161564170.1repeat element kmers were counted. The percentage of kmers counted from a given element type is plotted for each true element of origin (i.e., rows sum to 1) showing that 98% of counted kmers are found in reads originating from the correct repeat element. The inset shows only the SINE element types.

[0072] FIG.11 shows the localizing the Human Satellite 2 and 3 kmers within chm13. The subgroups of Human Satellite 2 and 3 described in Altemose et al.1are not annotated in the RepeatMasker track. In simulation the kmers from these element types largely overlap the broader satellite regions defined in the RepeatMasker track (left panel) and are counted genome-wide (right panel).

[0073] FIG.12 demonstrates the correlation between kmer counts and copy-number. The ratio of Tumor to Normal kmer counts for PCAWG samples by chromosome arm shows a correlation between kmer counts and increasing tumor copy number. The dashed gray lines indicate the expected kmer count ratio for copy neutral regions (diploid). The observed slope of the fitted line is less than 1, indicating that less kmers are found per chromosome arm than expected by copy number, consistent with the concept that such repeat sequences may undergo deletion as they facilitate gains in nearby genomic content.

[0074] FIGS. 13A-13C show the distribution of ERBB2 and SOX2 / PIK3CA gene copy number and repeat-related changes in PCAWG tumors. Copy number of ERBB2 in breast cancers (FIG.13A), SOX2 in lung cancers (FIG.13B, left), PIK3CA in lung cancers (FIG.13B, middle), and the number of samples with and without SOX2 / PIK3CA amplification showing a copy number > 5 along chr3, with the region of SOX2 and PIK3CA outlined by the black box (FIG.13B, right). (FIG. 13C) The distribution of Tumor:Normal kmer count ratios for repeat element types overlapping the SOX2 / PIK3CA region (31 Mb) in lung cancers shows increases in counts for tumors carrying a focal amplification. All elements with >10.0% of kmers found in the region are plotted. The Benjamini-Hochberg corrected p-value from the Wilcoxon signed-rank test is plotted for each comparison. Several of these elements represent novel changes not previously implicated in cancer (bold names).

[0075] FIG.14 demonstrates that the kmer repeat landscapes across human cancers reveal numerous differences from normal tissue with magnitude exceeding that expected by chromosomal arm gains and losses alone. The ratio of kmer repeat landscapes for each PCAWG tumor as compared to its matched normal is shown, with all non-significant changes or changes on the scale of that17 161564170.1expected by chromosomal arm gains and losses alone colored white. After this copy-number normalization, numerous changes to the kmer repeat landscapes remain evident, including in novel repeat elements.329 of 333 tumors are shown in the heatmap, comprising those with available copy number data from TCGA.

[0076] FIG. 15 demonstrates the impact of LINE-1 mediated deletions on kmer counts in the PCAWG Lung Cohort. Left panel, regardless of surrounding copy number, tumors with a deletion have lower kmer counts. Right panel, repeat element families represented by kmers in each deletion region. As expected, most kmers are from LINE repeat element types.

[0077] FIGS. 16A-16C demonstrate the impact of SINE element changes on survival in PCAWG tumors. (FIGS. 16A, 16B) To adjust for tumor-type, tumors are stratified by the type- specific median number of SINE changes ensuring that half of each tumor type is in each group. This separates tumors by overall survival and shows a trend towards significance for progression- free survival. (FIG. 16C) Distribution of SINE element changes among PCAWG tumors shows tissue-type specific patterns of change.

[0078] FIGS.17A, 17B demonstrate the impact of genomic stability metrics on survival in PCAWG tumors. (FIG.17A) Overall survival and (FIG.17B) Progression-free survival do not show significant associations with genomic stability metrics.

[0079] FIG.18 demonstrates the germline variability in repeat elements in PCAWG normal tissue samples. Variability in the ratio of observed to expected counts for each repeat element type among all PCAWG normal samples. The top panel shows the coefficient of variation for each of 1275 repeat types which have <75% of their kmers occurring on chrY (we excluded five repeat types that have high percentages of kmers on chrY as variability between sexes may affect the coefficient of variation for these repeat types). The bottom panel shows the distributions of kmer counts, with an inset highlighting the 10 features with the largest coefficient of variation, most of which are satellite elements.

[0080] FIG.19 demonstrates that ARTEMIS can distinguish PCAWG tumor from normal samples with high performance across all cancer types. The ARTEMIS model was trained with penalized logistic regression on the kmer repeat landscapes for 333 tumor and 333 normal samples and evaluated by mean score from 5-fold cross-validation with 10 repeats. ARTEMIS shows high overall performance for distinguishing PCAWG tumors from normal tissues across all tumor types and among individual tumor types.18 161564170.1

[0081] FIG. 20 demonstrates that kmer repeat landscapes in PCAWG samples show consistency across subsampled coverages. Correlation coefficients for 42 PCAWG lung cancers (top) and Bland-Altman plots for a random sample of five samples (bottom) between the original 40-80X samples, and the subsampled 30X and 1-2X versions show concordance. In the Bland- Altman plots, the x-axis is the log average kmer count and the y-axis is the log difference in counts.

[0082] FIG.21 demonstrates that kmer repeat landscapes in PCAWG subsamples of normal tissue reveal minimal differences. Heatmap showing ratio of kmer repeat landscapes (left) and ROC analysis (right) between two half-coverage subsamples of 100 random Normal samples yields minimal differences, suggesting that observed changes between Tumor and Normal are due to biological changes rather than technical variation.

[0083] FIG.22 shows a comparison of kmer repeat landscapes in plasma sequenced at 1-2x coverage show concordance across platforms and sequencing batches. Top panels, PCA analysis of the LUCAS cohort healthy samples by genomic library batch and sex shows that while sex differences are evident in kmer counts, technical variation between library batches is not evident. For kmer repeat landscapes from the LUCAS Cohort sequenced on the HiSeq and NovaSeq platforms, correlations (bottom left) and Bland-Altman plots for 5 random cancer and healthy patients (bottom right) show that concordance between sequencing replicates is high. On the Bland- Altman plots, the x-axis shows the log average count, and the y-axis shows the difference in the log counts. The dotted line separates repeat element types with an expected count of less than or greater than 1000 kmers / million aligned reads – elements to the right of this line show greater concordance between platforms in low coverage sequencing.

[0084] FIG.23 shows a diagram of cohorts and datasets used for plasma analyses.

[0085] FIG.24 shows the locations of 1 Mb bins with high density of epigenetic marks used in the ARTEMIS analyses in plasma.

[0086] FIG.25 (includes FIGS.25A-25C): . Feature importance and stability in the locked lung cancer ARTEMIS and ARTEMIS-DELFI models. FIG.25A: Normalized coefficients of the individual features retained by the LASSO regression model for lung cancer, scaled by their ensemble weights in the ARTEMIS-DELFI model FIG.25B: Stability analysis for scores generated by these models for individual patients using different cross-validation fold sizes. FIG. 25C: Stability analysis for scores across the population generated using different cross-validation fold sizes.19 161564170.1

[0087] FIG. 26 demonstrates ARTEMIS for early detection of liver cancer using cfDNA. Top panel, Distribution of ARTEMIS and joint ARTEMIS-DELFI scores for cancer and non-cancer patients in the cross-validated liver cohort show lower scores in non-cancer patients than cancer patients. Bottom panel, ROC analyses of ARTEMIS scores and a joint ARTEMIS-DELFI model show high performance for classifying individuals with and without liver cancer in the full cohort, and in subgroups by cancer stage.

[0088] FIG 27 demonstrates the application of the locked ARTEMIS cfDNA model in a cohort of patients undergoing treatment for lung cancer. From top to bottom, panels show the similar dynamics of maximum mutant allele fractions, ARTEMIS-DELFI scores, DELFI scores, and ARTEMIS scores at each available time point. Patients are plotted from left to right in order of increasing progression free survival, demonstrating that MAF trajectories related to tumor burden can be represented by kmer repeat landscape and fragmentation profile derived scores.

[0089] FIG.28: Germline variability in repeat elements in PCAWG Normal Tissue samples. Variability in the ratio of observed to expected counts for each repeat element type among all PCAWG normal samples. The top panel shows the coefficient of variation for each of 1275 repeat types which have <75% of their kmers occurring on chrY (we excluded five repeat types that have high percentages of kmers on chrY as variability between sexes may affect the coefficient of variation for these repeat types). The bottom panel shows the distributions of kmer counts, with an inset highlighting the 10 features with the largest coefficient of variation, many of which are satellite elements.

[0090] FIG.28: ARTEMIS can distinguish PCAWG tumor from normal samples with high performance across all cancer types. ROC Curves for the ARTEMIS model was trained with penalized logistic regression on the kmer repeat landscapes for 525 tumor and 525 normal samples and evaluated by mean score from 5-fold cross-validation with 10 repeats. The overall ROC curve and ROC curves separated by race or tumor type are shown.

[0091] FIG.29 (includes FIGS.29A and 29B): Impact of ARTEMIS Score on survival in PCAWG tumors. FIG. 29A: ARTEMIS scores across all tumor tissues as compared to normal samples. All Stage III or IV tumors and their matched normal are shown (n=167). FIG.29B: Kaplan- Meier plots of overall survival and progression-free survival for these patients with cancer stratified by high or low ARTEMIS score. To adjust for tumor-type, tumors are stratified by the type-specific median ARTEMIS score ensuring that half of each tumor type is in each group. P-values were20 161564170.1calculated using the log-rank test. In FIG.29, ARTEMIS Score High values are lower on the y axis relative to the ARTEMIS Score Low values.

[0092] FIG.30: Simulations of combined impact of tumor-specific structural and epigenetic changes on cfDNA representation. Bar plots show the percentage of simulations of structural and epigenetic changes to repeat elements in cfDNA from cancer patients where tumor-derived changes showed concordant or discordant directionality in the plasma. Red bars show simulations where plasma coverage was impacted by both tumor-derived structural and epigenetic changes, whereas gray bars show simulations where only structural changes were considered. In the pairs of bars, on the left is depicted epigenetic changes impact plasma coverage, and on the right is depicted no epigenetic impacts on plasma coverage.

[0093] FIG.31 (includes FIGS.31A-31B): ARTEMIS for detection of liver cancer using cfDNA. FIG.31A: Distribution of ARTEMIS and joint ARTEMIS-DELFI scores for cancer and non-cancer patients in the cross-validated liver cohort. FIG. 31B: ROC analyses of ARTEMIS scores and a joint ARTEMIS-DELFI model classifying individuals with and without liver cancer in the full cohort, and in subgroups by cancer stage.

[0094] FIG.32: External validation of the locked ARTEMIS and ARTEMIS-DELFI models for lung cancer detection. ROC analyses by stage for external validation of the locked ARTEMIS and ARTEMIS-DELFI models for lung cancer detection

[0095] FIG. 33. Validation of the locked ARTEMIS and ARTEMIS-DELFI models for detection of lung cancer recurrence. The locked ARTEMIS and ARTEMIS-DELFI models for lung cancer detection for samples from individuals with a prior history of cancer who experienced cancer recurrence as compared to scores for samples from individuals who did not experience a recurrence. P-values were calculated using the Wilcoxon signed-rank test.

[0096] FIG. 34 (includes FIGS. 34A-34B) Validation of the locked ARTEMIS and ARTEMIS-DELFI models for detection of lung cancer recurrence. The locked ARTEMIS and ARTEMIS-DELFI models for lung cancer detection for samples from individuals with a prior history of cancer who experienced cancer recurrence as compared to scores for samples from individuals who did not experience a recurrence. P-values were calculated using the Wilcoxon signed-rank test. DETAILED DESCRIPTION21 161564170.1

[0097] In one aspect, an alignment-free de novo kmer finding is provided for identifying repeat elements from whole genome sequencing. Using the present methods, it was demonstrated in the recently characterized long-read reference genome that repeat elements are enriched in genes and pathways commonly altered in human cancer. I n t h e e x a m p l e s s e c t i o n w h i c h f o l l o w s analysis of 2428 samples from 1758 individuals, revealed tumor-specific changes in repeat element representation could be detected in tissue and circulating cell-free DNA (cfDNA) of cancer patients. A machine learning model trained using genome-wide repeat landscapes and fragmentation profiles in cfDNA detected patients with early-stage lung cancer and was validated in a separate diagnostic population. The kmer repeat landscape approach allows the reconstruction of repeat landscapes using low-coverage whole genome sequencing and facilitates large-scale analyses of these regions for characterization and detection of human cancer.

[0098] ARTEMIS

[0099] ARTEMIS (Analysis of RepeaT EleMents in dISease) was developed as an alignment-free, genome-wide approach for analyzing repeat landscapes in short read sequencing. ARTEMIS assesses over 1200 individual repeat types that occur genome-wide and span 57 subfamilies comprising 6 families (Satellites, RNA elements, transposable elements, LINEs, SINEs, LTRs). In this study, we use ARTEMIS to show that repeat landscapes are enriched in genes commonly altered in human cancer and tumor-specific changes in repeats reflect a combination of structural and epigenetic changes in the cancer genome. Genome-wide repeat landscape analyses with ARTEMIS can be performed using low-coverage whole genome sequencing, permitting analysis of repeat landscapes in cfDNA for detection of human cancer.

[0100] Accordingly, in certain aspects, a method of identifying repeat element types, comprises extracting repeat sequences and coordinates from known repeat element types; identifying short nucleic acid sequences (kmers) in genomic or cell free DNA; selecting kmers occurring in a single repeat type and identifying unique kmers of repeat element types; wherein the unique kmers identify one or a plurality of repeat element types. In certain embodiments, repeat element types are excluded from families comprising low complexity, unknown, simple repeats or combinations thereof. In certain embodiments, elements from each family comprises tRNAs, srpRNAs, snRNAs, scRNAs, rRNAs, RNA elements, DNA, retroposons, or combinations thereof, are aggregated.

[0101] In certain embodiments, one or a plurality of kmers are generated from one or more nucleic acid sequences. The nucleic acid sequence may be any sequence, such as a genomic sequence22 161564170.1or a cfDNA. In one embodiment, kmer generation is performed by an automated algorithm that receives a nucleic acid sequence and generates multiple kmers from that sequence. The kmer may be produced from a single organism or species, or from multiple organisms or species. For example, kmer can be generated from the genomic sequences of common, possible or known contaminants to detect these contaminants during quality control analysis. In addition, k-mers can be generated from the genomic sequences of different organisms that may be in the composite sample to detect and analyze sequence data from multiple different organisms during quality control analysis.

[0102] k-mers extracted from one or more nucleic acid sequences may have the same length or various different lengths. As just one example, the generated kmer may be about 20, 24, 28 or 32 bases, but longer and shorter k-mers also are suitable. Larger ks are less likely to find multiple hits in the reference genome, which can result in multiple types of k-mer annotations. For example, a small k maps to many parts of the genome and does not provide useful and unique annotation information. However, a large k also increases the amount of memory required to store the generated k-mer. A short k reduces the amount of memory required to store the generated kmer. Therefore, the determination of optimal k depends in particular on various factors such as computational power and memory size.

[0103] In certain embodiments, the kmers occurring in single repeat type elements and not in non-repeat regions are selected. In certain embodiments, the kmer comprises 10 to 40, 50, 60 or 70 nucleotides. In certain embodiments, the kmer comprises 10 to 40 nucleotides. In certain embodiments, the kmer comprises 15 to 35 nucleotides. In certain embodiments, the kmer comprises 20 to 30 nucleotides. In certain embodiments, the kmer comprises 22 to 28 nucleotides. In certain embodiments, the kmer comprises 23 to 26 nucleotides. In certain embodiments, the kmer comprises about 24 nucleotides.

[0104] Systems

[0105] In some examples, the present disclosure provides systems, methods, or kits that can include data analysis realized in measurement devices (e.g., laboratory instruments, such as a sequencing machine), software code that executes on computing hardware. The software can be stored in memory and execute on one or more hardware processors. The software can be organized into routines or packages that can communicate with each other. A module can comprise one or more devices / computers, and potentially one or more software routines / packages that execute on the one or more devices / computers. For example, an analysis application or system can include at least23 161564170.1a data receiving module, a data pre-processing module, a data analysis module (which can operate on one or more types of genomic data), a data interpretation module, or a data visualization module.

[0106] The data receiving module can connect laboratory hardware or instrumentation with computer systems that process laboratory data. The data pre-processing module can perform operations on the data in preparation for analysis. Examples of operations that can be applied to the data in the pre-processing module include affine transformations, denoising operations, data cleaning, reformatting, or subsampling. The data analysis module, which can be specialized for analyzing genomic data from one or more genomic materials, can, for example, take assembled genomic sequences and perform probabilistic and statistical analysis to identify abnormal patterns related to a disease, pathology, state, risk, condition, or phenotype. The data interpretation module can use analysis methods, for example, drawn from statistics, mathematics, or biology, to support understanding of the relation between the identified abnormal patterns and health conditions, functional states, prognoses, or risks. The data analysis module and / or the data interpretation module can include one or more machine learning models, which can be implemented in hardware, e.g., which executes software that embodies a machine learning model. The data visualization module can use methods of mathematical modeling, computer graphics, or rendering to create visual representations of data that can facilitate the understanding or interpretation of results. The present disclosure provides computer systems that are programmed to implement methods of the disclosure.

[0107] In some embodiments, the methods disclosed herein can include computational analysis on nucleic acid sequencing data of samples from an individual or from a plurality of individuals. An analysis can identify a variant inferred from sequence data to identify sequence variants based on probabilistic modeling, statistical modeling, mechanistic modeling, network modeling, or statistical inferences. Non-limiting examples of analysis methods include principal component analysis, autoencoders, singular value decomposition, Fourier bases, wavelets, discriminant analysis, regression, support vector machines, tree-based methods, networks, matrix factorization, and clustering. Non-limiting examples of variants include a germline variation or a somatic mutation. In some examples, a variant can refer to an already-known variant. The already- known variant can be scientifically confirmed or reported in literature. In some examples, a variant can refer to a putative variant associated with a biological change. A biological change can be known or unknown. In some examples, a putative variant can be reported in literature, but not yet biologically confirmed. Alternatively, a putative variant is never reported in literature, but can be24 161564170.1inferred based on a computational analysis disclosed herein. In some examples, germline variants can refer to nucleic acids that induce natural or normal variations.

[0108] In certain embodiments, the computer system includes a central processing unit (CPU, also “processor” and “computer processor” herein), which can be a single core or multi core processor, or a plurality of processors for parallel processing; memory (e.g., cache, random-access memory, read-only memory, flash memory, or other memory); electronic storage unit (e.g., hard disk), communication interface (e.g., network adapter) for communicating with one or more other systems; and peripheral devices, such as adapters for cache, other memory, data storage and / or electronic display. The memory, storage unit, interface and peripheral devices may be in communication with the CPU through a communication bus (solid lines), such as a motherboard. The storage unit can be a data storage unit (or data repository) for storing data. One or more analyte feature inputs can be entered from the one or more measurement devices. Example analytes and measurement devices are described herein.

[0109] The computer system can be operatively coupled to a computer network (“network”) with the aid of the communication interface. The network can be the Internet, an internet and / or extranet, or an intranet and / or extranet that is in communication with the Internet. The network in some cases is a telecommunication and / or data network. The network can include one or more computer servers, which can enable distributed computing, such as cloud computing over the network (“the cloud”) to perform various aspects of analysis, calculation, and generation of the present disclosure, such as, for example, activation of a valve or pump to transfer a reagent or sample from one chamber to another or application of heat to a sample (e.g., during an amplification reaction), other aspects of processing and / or assaying a sample, performing sequencing analysis, measuring sets of values representative of classes of molecules, identifying sets of features and feature vectors from assay data, processing feature vectors using a machine learning model to obtain output classifications, and training a machine learning model (e.g., iteratively searching for optimal values of parameters of the machine learning model). Such cloud computing may be provided by cloud computing platforms such as, for example, Amazon Web Services (AWS), Microsoft Azure, Google Cloud Platform, and IBM cloud. The network, in some cases with the aid of the computer system, can implement a peer-to-peer network, which may enable devices coupled to the computer system to behave as a client or a server.25 161564170.1

[0110] The CPU can execute a sequence of machine-readable instructions, which can be embodied in a program or software. The instructions can be stored in a memory location, such as the memory. The instructions can be directed to the CPU, which can subsequently program or otherwise configure the CPU to implement methods of the present disclosure. The CPU can be part of a circuit, such as an integrated circuit. One or more other components of the system can be included in the circuit. In some cases, the circuit is an application specific integrated circuit (ASIC).

[0111] The storage unit can store files, such as drivers, libraries and saved programs. The storage unit can store user data, e.g., user preferences and user programs. The computer system in some cases can include one or more additional data storage units that are external to the computer system, such as located on a remote server that is in communication with the computer system through an intranet or the Internet.

[0112] The computer system can communicate with one or more remote computer systems through the network. For instance, the computer system can communicate with a remote computer system of a user. Examples of remote computer systems include personal computers (e.g., portable PC), slate or tablet PC's (e.g., Apple® iPad, Samsung® Galaxy Tab), telephones, Smart phones (e.g., Apple® iPhone, Android-enabled device, Blackberry®), or personal digital assistants. The user can access the computer system via the network.

[0113] Methods as described herein can be implemented by way of machine (e.g., computer processor) executable code stored on an electronic storage location of the computer system such as, for example, on the memory or electronic storage unit. The machine executable or machine-readable code can be provided in the form of software. During use, the code can be executed by the CPU. In some cases, the code can be retrieved from the storage unit and stored on the memory for ready access by the CPU. In some situations, the electronic storage unit can be precluded, and machine- executable instructions are stored on memory.

[0114] The code can be pre-compiled and configured for use with a machine having a processer adapted to execute the code or can be compiled during runtime. The code can be supplied in a programming language that can be selected to enable the code to execute in a pre-compiled or as compiled fashion.

[0115] Aspects of the systems and methods provided herein, such as the computer system, can be embodied in programming. Various aspects of the technology can be thought of as “products” or “articles of manufacture” typically in the form of machine (or processor) executable code and / or26 161564170.1associated data that is carried on or embodied in a type of machine readable medium. Machine- executable code can be stored on an electronic storage unit, such as memory (e.g., read-only memory, random-access memory, flash memory) or a hard disk. “Storage” type media can include any or all of the tangible memory of the computers, processors or the like, or associated modules thereof, such as various semiconductor memories, tape drives, disk drives and the like, which may provide non-transitory storage at any time for the software programming. All or portions of the software may at times be communicated through the Internet or various other telecommunication networks. Such communications, for example, may enable loading of the software from one computer or processor into another, for example, from a management server or host computer into the computer platform of an application server. Thus, another type of media that can bear the software elements includes optical, electrical and electromagnetic waves, such as used across physical interfaces between local devices, through wired and optical landline networks and over various air-links. The physical elements that carry such waves, such as wired or wireless links, optical links or the like, also can be considered as media bearing the software. As used herein, unless restricted to non-transitory, tangible “storage” media, terms such as computer or machine “readable medium” refer to any medium that participates in providing instructions to a processor for execution.

[0116] Hence, a machine readable medium, such as computer-executable code, may take many forms, including but not limited to, a tangible storage medium, a carrier wave medium, or physical transmission medium. Non-volatile storage media include, for example, optical or magnetic disks, such as any of the storage devices in any computer(s) or the like, such as can be used to implement the databases, etc. shown in the drawings. Volatile storage media include dynamic memory, such as main memory of such a computer platform. Tangible transmission media include coaxial cables; copper wire and fiber optics, including the wires that comprise a bus within a computer system.

[0117] Carrier-wave transmission media may take the form of electric or electromagnetic signals, or acoustic or light waves such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer-readable media therefore include for example: a floppy disk, a flexible disk, hard disk, magnetic tape, any other magnetic medium, a CD- ROM, DVD or DVD-ROM, any other optical medium, punch cards paper tape, any other physical storage medium with patterns of holes, a RAM, a ROM, a PROM and EPROM, a FLASH-EPROM, any other memory chip or cartridge, a carrier wave transporting data or instructions, cables or links27 161564170.1transporting such a carrier wave, or any other medium from which a computer may read programming code and / or data. Many of these forms of computer readable media can be involved in carrying one or more sequences of one or more instructions to a processor for execution.

[0118] The computer system can include or be in communication with an electronic display that comprises a user interface (UI) for providing, for example, a current stage of processing or assaying of a sample (e.g., a particular step, such as a lysis step, or sequencing step that is being performed). Inputs are received by the computer system from one or more measurement. Examples of UIs include, without limitation, a graphical user interface (GUI) and web-based user interface. The algorithm can, for example, process and / or assay a sample, perform sequencing analysis, identify kmers comprising each repeat type, measure sets of values representative of classes of molecules, identify sets of features and feature vectors from assay data, process feature vectors using a machine learning model to obtain output classifications, and train a machine learning model (e.g., iteratively search for optimal values of parameters of the machine learning model).

[0119] In some embodiments, systems capable of executing one or more algorithms, e.g., laptops, desktops, iPads, mobile devices etc., for determining changes in cfDNA mutation profiles, frequency of mutations and / or fragmentation profiles classifies the subject as a cancer patient based on the cfDNA mutation profiles, frequency of mutations and / or fragmentation for the subject.54These systems further execute machine learning algorithms that can be used to generate models such as, for example, high-risk populations and low-risk general populations (a penalized logistic regression with the Mathios et al.27features as well as coverage from transcription factor binding sites.45These models can be trained on the subject cohort with 5-fold cross validation with 10 repeats, and scores for each sample ae calculated by the mean across repeats and evaluated using AUC-ROC. For example, the first model used the high-risk non-cancer and HCC patients while the second used the non-cancer individuals without liver pathology. The locked high-risk model trained on the cohort was applied to a second and different cohort to generate cancer predictions on an external validation set. A “class label” can be applied to each sample indicating the classification of the sample for any number of input features. For example, the class labels for the set of cohorts could identify kmer repeat landscapes, indicate the identity of cfDNA mutation profiles, frequency of mutations and / or fragmentation profiles based on genomic location etc. The resulting training sets are provided to machine learning unit, such as a neural network or a support vector machine. Using the training set, the machine learning unit may generate a model to classify the sample according to28 161564170.1the kmer repeat landscape to generate an ARTEMIS score for disease prediction, cfDNA mutation profiles, frequency of mutations and / or fragmentation profile.

[0120] In some embodiments, a method is provided for creating a trained classifier, comprising the steps of: extracting repeat sequences and coordinates from known repeat element types; identifying short nucleic acid sequences (kmers) in genomic or cell free DNAs; electing kmers occurring in a single repeat type and identifying unique kmers of repeat element types; wherein the unique kmers identify one or a plurality of repeat element types. For analysis of an individual sample, the kmer repeat landscape was defined as the count of all kmers in a sequenced sample that matched each of the 1280 repeat types divided by the number of aligned sequence reads. As changes in repeat sequences may occur during initiation of cancer and other diseases, this comprehensive compendium of repeat features can be used to train machine learning models to distinguish genomes from normal and disease states.

[0121] As an example, a trained classifier may use a learning algorithm selected from the group consisting of: a random forest, a neural network, a support vector machine, and a linear classifier. Each of the plurality of different classes may be selected from the group consisting of healthy, breast cancer, colon cancer, lung cancer, pancreatic cancer, prostate cancer, ovarian cancer, melanoma, and liver cancer.

[0122] A trained classifier may be applied to a method of classifying a sample from a subject. This method of classifying may comprise: (a) providing a multi-parametric model representative of the kmer repeat from a test sample from the subject; and (b) classifying the test sample using a trained classifier. After the test sample is classified into one or more classes, a therapeutic intervention on the subject can be performed based on the classification of the sample.

[0123] In some embodiments, training sets are provided to a machine learning unit, such as a neural network or a support vector machine. Using the training set, the machine learning unit may generate a model to classify the sample according to a treatment response to one or more therapeutic disclosures. This is also referred to as “calling”. The model developed may employ information from any part of a test vector.

[0124] In general, machine learning can be used to reduce a set of data generated from all (primary sample / analytes / test) combinations into an optimal predictive set of features, e.g., which satisfy specified criteria. In various examples statistical learning, and / or regression analysis can be applied. Simple to complex and small to large models making a variety of modeling assumptions29 161564170.1can be applied to the data in a cross-validation paradigm. Simple to complex includes considerations of linearity to non-linearity and non-hierarchical to hierarchical representations of the features. Small to large models includes considerations of the size of basis vector space to project the data onto as well as the number of interactions between features that are included in the modelling process.

[0125] Machine learning techniques can be used to assess the commercial testing modalities most optimal for cost / performance / commercial reach as defined in the initial question. A threshold check can be performed: If the method applied to a hold-out dataset that was not used in cross validation surpasses the initialized constraints, then the assay is locked, and production initiated. For example, a threshold for assay performance may include a desired minimum accuracy, positive predictive value (PPV), negative predictive value (NPV), clinical sensitivity, clinical specificity, area under the curve (AUC), or a combination thereof. For example, a desired minimum accuracy, PPV, NPV, clinical sensitivity, clinical specificity, or combination thereof may be at least about 50%, at least about 55%, at least about 60%, at least about 65%, at least about 70%, at least about 75%, at least about 80%, at least about 81%, at least about 82%, at least about 83%, at least about 84%, at least about 85%, at least about 86%, at least about 87%, at least about 88%, at least about 89%, at least about 90%, at least about 91%, at least about 92%, at least about 93%, at least about 94%, at least about 95%, at least about 96%, at least about 97%, at least about 98%, or at least about 99%. As another example, a desired minimum AUC may be at least about 0.50, at least about 0.55, at least about 0.60, at least about 0.65, at least about 0.70, at least about 0.75, at least about 0.80, at least about 0.81, at least about 0.82, at least about 0.83, at least about 0.84, at least about 0.85, at least about 0.86, at least about 0.87, at least about 0.88, at least about 0.89, at least about 0.90, at least about 0.91, at least about 0.92, at least about 0.93, at least about 0.94, at least about 0.95, at least about 0.96, at least about 0.97, at least about 0.98, or at least about 0.99. A subset of assays may be selected from a set of assays to be performed on a given sample based on the total cost of performing the subset of assays, subject to the threshold for assay performance, such as desired minimum accuracy, positive predictive value (PPV), negative predictive value (NPV), clinical sensitivity, clinical specificity, area under the curve (AUC), and a combination thereof. If the thresholds are not met, then the assay engineering procedure can loop back to either the constraint setting for possible relaxation or to the wet lab to change the parameters in which data was acquired. Given the clinical question, biological constraints, budget, lab machines, etc., can constrain the problem.30 161564170.1

[0126] In certain embodiments, the computer processing of a machine learning technique can include method(s) of statistics, mathematics, biology, or any combination thereof. In various examples, any one of the computer processing methods can include a dimension reduction method, logistic regression, dimension reduction, principal component analysis, autoencoders, singular value decomposition, Fourier bases, singular value decomposition, wavelets, discriminant analysis, support vector machine, tree-based methods, random forest, gradient boost tree, logistic regression, matrix factorization, network clustering, statistical testing and neural network.

[0127] In certain embodiments, the computer processing of a machine learning technique can include logistic regression, multiple linear regression (MLR), dimension reduction, partial least squares (PLS) regression, principal component regression, autoencoders, variational autoencoders, singular value decomposition, Fourier bases, wavelets, discriminant analysis, support vector machine, decision tree, classification and regression trees (CART), tree-based methods, random forest, gradient boost tree, logistic regression, matrix factorization, multidimensional scaling (MDS), dimensionality reduction methods, t-distributed stochastic neighbor embedding (t-SNE), multilayer perceptron (MLP), network clustering, neuro-fuzzy, neural networks (shallow and deep), artificial neural networks, Pearson product-moment correlation coefficient, Spearman's rank correlation coefficient, Kendall tau rank correlation coefficient, or any combination thereof. In some examples, the computer processing method is a supervised machine learning method including, for example, a regression, support vector machine, tree-based method, and neural network. In some examples, the computer processing method is an unsupervised machine learning method including, for example, clustering, network, principal component analysis, and matrix factorization.

[0128] For supervised learning, training samples (e.g., in thousands) can include measured data (e.g., of various analytes) and known labels, which may be determined via other time- consuming processes, such as imaging of the subject and analysis by a trained practitioner. Example labels can include classification of a subject, e.g., discrete classification of whether a subject has cancer or not or continuous classifications providing a probability (e.g., a risk or a score) of a discrete value. A learning module can optimize parameters of a model such that a quality metric (e.g., accuracy of prediction to known label) is achieved with one or more specified criteria. Determining a quality metric can be implemented for any arbitrary function including the set of all risk, loss, utility, and decision functions. A gradient can be used in conjunction with a learning step (e.g., a31 161564170.1measure of how much the parameters of the model should be updated for a given time step of the optimization process).

[0129] As described above, examples can be used for a variety of purposes. For example, plasma (or other sample) can be collected from subjects symptomatic with a condition (e.g., known to have the condition) and healthy subjects. Genetic data (e.g., kmer repeats, cfDNA) can be acquired analyzed to obtain a variety of different features, which can include features based on a genome wide analysis. These features can form a feature space that is searched, stretched, rotated, translated, and linearly or non-linearly transformed to generate an accurate machine learning model, which can differentiate between healthy subjects and subjects with the condition (e.g., identify a disease or non- disease status of a subject). Output derived from this data and model (which may include probabilities of the condition, stages (levels) of the condition, or other values), can be used to generate another model that can be used to recommend further procedures, e.g., recommend a biopsy or keep monitoring the subject condition.

[0130] To call a therapeutic response, a calling algorithm may take the genetic information and treatment responses of a plurality of individuals having a disease or condition. The data may first be normalized (using the same procedure as for the clustering algorithm). The calling operation (classification) may be performed using, for example, a Bayesian model. The score for each call's Call Score can be the product of a Training Score and a data-to-model fit score. After scoring all the treatment responses, the application may compute a composite score.

[0131] In some embodiments, a training dataset comprises clinical data selected from the group consisting of cancer stage, type of surgical procedure, age, tumor grading, depth of tumor infiltration, occurrence of post-operative complications, and the presence of venous invasion. In some embodiments, the training dataset is pre-processed, comprising transforming the provided data into class-conditional probabilities.

[0132] Another embodiment uses machine learning techniques to train a statistical classifier, specifically a support vector machine, for each cancer stage category based on word occurrences in a corpus of histology reports for each patient. New reports can then be classified according to the most likely stage, facilitating the collection and analysis of population staging data.

[0133] In some embodiments, a machine learning algorithm is selected from the group consisting of: a supervised or unsupervised learning algorithm selected from support vector machine,32 161564170.1random forest, nearest neighbor analysis, linear regression, binary decision tree, discriminant analyses, logistic classifier, and cluster analysis.

[0134] In general, a system can comprise a report generator for reporting on cancer test results and treatment options. The report generator system can be a central data processing system configured to establish communications directly with: a remote data site or laboratory, a medical practice / healthcare provider (treating professional) and / or a patient / subject through communication links. The laboratory can be medical laboratory, diagnostic laboratory, medical facility, medical practice, point-of-care testing device, or any other remote data site capable of generating subject clinical information. Subject clinical information includes but it is not limited to laboratory test data, X-ray data, examination and diagnosis. The healthcare provider or practice 26 includes medical services providers, such as doctors, nurses, home health aides, technicians and physician's assistants, and the practice is any medical care facility staffed with healthcare providers. In certain instances, the healthcare provider / practice is also a remote data site. In a cancer treatment embodiment, the subject may be afflicted with cancer, among others.

[0135] Other clinical information for a cancer subject includes the results of laboratory tests, imaging or medical procedure directed towards the specific cancer that one of ordinary skill in the art can readily identify. The list of appropriate sources of clinical information for cancer includes but it is not limited to: CT scan, MRI scan, ultrasound scan, bone scan, PET Scan, bone marrow test, barium X-ray, endoscopy, lymphangiogram, IVU (Intravenous urogram) or IVP (IV pyelogram), lumbar puncture, cystoscopy, immunological tests (anti-malignin antibody screen), and cancer marker tests.

[0136] The subject clinical information may be obtained from the laboratory manually or automatically. For simplicity of the system the information is obtained automatically at predetermined or regular time intervals. A regular time interval refers to a time interval at which the collection of the laboratory data is carried out automatically by the methods and systems described herein based on a measurement of time such as hours, days, weeks, months, years etc. In one embodiment, the collection of data and processing is carried out at least once a day. In one embodiment, the transfer and collection of data is carried out once every month, biweekly, or once a week, or once every couple of days. Alternatively, the retrieval of information may be carried out at predetermined but not regular time intervals. For instance, a first retrieval step may occur after one week and a second retrieval step may occur after one month. The transfer and collection of data33 161564170.1can be customized according to the nature of the disorder that is being managed and the frequency of required testing and medical examinations of the subjects.

[0137] In certain embodiments, a genetic report is generated from a subject’s sample, e.g. cfDNA. The polynucleotides in a sample can be sequenced, e.g., whole genome sequencing, NGS sequencing, producing a plurality of sequence reads. In some embodiments, genetic information comprises variables defining the genomic organization of cancer cells or the genomic organization of single disseminated cancer cells. In some embodiments, the genetic information comprises kmer repeats or kmer landscapes generated from the genome of one or more subjects. In some embodiments, the genetic information comprises sequence or abundance data from one or more genetic loci in cell-free DNA from the individuals.

[0138] Genetic variants can also be identified. Genetic variants include sequence variants, copy number variants and nucleotide modification variants. A sequence variant is a variation in a genetic nucleotide sequence. A copy number variant is a deviation from wild type in the number of copies of a portion of a genome. Genetic variants include, for example, single nucleotide variations (SNPs), insertions, deletions, inversions, transversions, translocations, gene fusions, chromosome fusions, gene truncations, copy number variations (e.g., aneuploidy, partial aneuploidy, polyploidy, gene amplification), abnormal changes in nucleic acid chemical modifications, abnormal changes in epigenetic patterns and abnormal changes in nucleic acid methylation. The process then determines the frequency of genetic variants in the sample containing the genetic material. Since this process is noisy, the process separates information from noise. The sensitivity of detecting genetic variants can be increased by increasing read depth of polynucleotides (e.g., by sequencing to a greater read depth at in a sample from a subject at two or more time points).

[0139] To increase the diagnosis confidence, a plurality of measurements can be taken. Or alternatively using measurements at a plurality of time points (e.g., 2, 3, 4, 5, 6, 7, 8, 9, 10, or more time points) to determine whether cancer is advancing, in remission or stabilized. The diagnostic confidence can be used to identify disease states. For example, cell free polynucleotides taken from a subject can include polynucleotides derived from normal cells, as well as polynucleotides derived from diseased cells, such as cancer cells. Polynucleotides from cancer cells may bear genetic variants, such as somatic cell mutations and copy number variants. When cell free polynucleotides from a sample from a subject are sequenced, and cfDNA mutation profiles, frequency of mutations and / or fragmentation profiles can be produced as described in the examples section which follows.34 161564170.1

[0140] In certain embodiments, the diagnostic confidence is based on an ARTEMIS score as detailed in the examples section which follows. Briefly, ensembled penalized logistic regression models were trained using the kmer repeat landscapes and defined the predictions as the ARTEMIS score. First, a kmer repeat landscape was obtained for each sample using the 786 features with more than 1000 kmers per million aligned reads expected. This filtering was employed because at low coverage features with low abundance have greater technical variation. The remaining features were partitioned into 6 families (LINEs, SINEs, Satellites, LTRs, RNA Elements, DNA Elements) and centered and scaled features within their family in each sample. RNA Elements and DNA Elements were combined because after filtering for the 1000 kmers / million aligned reads threshold, only two RNA Element features remained.5611 Mb bins were further defined using the epigenetic analyses above that had either >90% of their bases covered by peaks from one of the Histone CHIP-Seq experiments above, or >30% of their bases covered by one of the three groups of chromatin states defined above. Aligned fragment coverage was ca lcula ted in these bins. Bins with mappability < 0.9 or GC content < 0.3 were excluded from downstream analysis. Six penalized logistic regression (PLR) models were then trained using repeat landscapes for the five repeat families above and the epigenetic profiles and ensembled their scores together with a PLR model using leave-one-out cross validation with nested 5-fold cross validation to train each learner. The combined score was defined as the ARTEMIS Score.

[0141] Numerous cancers may be detected using the methods and systems described herein. Cancers cells, as most cells, can be characterized by a rate of turnover, in which old cells die and replaced by newer cells. Generally dead cells, in contact with vasculature in a given subject, may release DNA or fragments of DNA into the blood stream. This is also true of cancer cells during various stages of the disease. Cancer cells may also be characterized, dependent on the stage of the disease, by various genetic aberrations such as copy number variation as well as mutations. This phenomenon may be used to detect the presence or absence of cancers individuals using the methods and systems described herein.

[0142] In the early detection of cancers, any of the systems or methods herein described, including mutation detection or copy number variation detection may be utilized to detect cancers. These system and methods may be used to detect any number of genetic aberrations that may cause or result from cancers. These may include but are not limited to kmer landscapes, kmer repeats, cfDNA mutation profiles, frequency of mutations, cfDNA fragmentation profiles, mutations,35 161564170.1mutations, indels, copy number variations, transversions, translocations, inversion, deletions, aneuploidy, partial aneuploidy, polyploidy, chromosomal instability, chromosomal structure alterations, gene fusions, chromosome fusions, gene truncations, gene amplification, gene duplications, chromosomal lesions, DNA lesions, abnormal changes in nucleic acid chemical modifications, abnormal changes in epigenetic patterns, abnormal changes in nucleic acid methylation infection and cancer.

[0143] Additionally, the systems and methods described herein may also be used to help characterize certain cancers. Genetic data, e.g. kmer landscapes, produced from the system and methods of this disclosure may allow practitioners to help better characterize a specific form of cancer. Often times, cancers are heterogeneous in both composition and staging. Genetic profile data may allow characterization of specific sub-types of cancer that may be important in the diagnosis or treatment of that specific sub-type. This information may also provide a subject or practitioner clues regarding the prognosis of a specific type of cancer.

[0144] The systems and methods provided herein may be used to monitor already known cancers, or other diseases in a particular subject. This may allow either a subject or practitioner to adapt treatment options in accord with the progress of the disease. In this example, the systems and methods described herein may be used to construct kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, cfDNA mutation profiles, frequency of mutations and / or fragmentation profiles of a particular subject of the course of the disease. In some instances, cancers can progress, becoming more aggressive and genetically unstable. In other examples, cancers may remain benign, inactive or dormant. The system and methods of this disclosure may be useful in determining disease progression.

[0145] Further, the systems and methods described herein may be useful in determining the efficacy of a particular treatment option. In one example, certain treatment options may be correlated with kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, cfDNA mutation profiles, frequency of mutations and / or fragmentation profiles of cancers over time. This correlation may be useful in selecting a therapy. Additionally, if a cancer is observed to be in remission after treatment, the systems and methods described herein may be useful in monitoring residual disease or recurrence of disease.

[0146] Further, the methods of the disclosure may be used to characterize the heterogeneity of an abnormal condition in a subject, the method comprising generating kmer repeat landscapes,36 161564170.1kmer repeat landscapes combined with cfDNA mutation profiles, a cfDNA mutation profile, frequency of mutations and / or fragmentation profile of extracellular polynucleotides in the subject, wherein the cfDNA mutation profile comprises a plurality of data resulting from profile variation and mutation analyses. In some cases, including but not limited to cancer, a disease may be heterogeneous. Disease cells may not be identical. In the example of cancer, some tumors are known to comprise different types of tumor cells, some cells in different stages of the cancer. In other examples, heterogeneity may comprise multiple foci of disease. Again, in the example of cancer, there may be multiple tumor foci, perhaps where one or more foci are the result of metastases that have spread from a primary site (also known as distant metastases).

[0147] The methods of this disclosure may be used to generate kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, a profile, fingerprint, or set of data that is a summation of genetic information derived from different cells in a heterogeneous disease. This set of data may comprise copy number variation and mutation analyses alone or in combination.

[0148] Further, these reports are submitted and accessed electronically via the internet. Analysis of data occurs at a site other than the location of the subject. The report is generated and transmitted to the subject's location. Via an internet enabled computer, the subject accesses the reports reflecting his tumor burden.

[0149] The annotated information can be used by a health care provider to select other drug treatment options and / or provide information about drug treatment options to an insurance company. The method can include annotating the drug treatment options for a condition in, for example, the NCCN Clinical Practice Guidelines in Oncology™ or the American Society of Clinical Oncology (ASCO) clinical practice guidelines.

[0150] Reports are generated for kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, mapping genome positions and cfDNA mutation profile variation for the subject with cancer. These reports, in comparison to other profiles of subjects with known outcomes, can indicate that a particular cancer is aggressive and resistant to treatment. The subject is monitored for a period and retested. If at the end of the period, the kmer repeat landscapes, the kmer repeat landscapes combined with cfDNA mutation profiles, the cfDNA mutation profiles, frequency of mutations and / or fragmentation variation profile do not vary, this may indicate that the current treatment is not working. A comparison is done with kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, cfDNA mutation profiles of other subjects. For37 161564170.1example, if it is determined that a change in kmer repeat landscapes or kmer repeats indicates that the cancer is advancing, then the original treatment regimen as prescribed is no longer treating the cancer and a new treatment is prescribed.

[0151] In certain embodiments, the system receives genetic information from a DNA sequencer. The process then determines specific kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, or cfDNA alterations and frequencies thereof. These reports are submitted and accessed electronically via the internet. Analysis of data occurs at a site other than the location of the subject. The report is generated and transmitted to the subject's location. Via an internet enabled computer, the subject accesses the reports reflecting his tumor burden.

[0152] While temporal information can be used to enhance the information for kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, cfDNA mutation profiles or frequency of mutations, other consensus methods can be applied. In other embodiments, the historical comparison can be used in conjunction with other consensus kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, cfDNA mutation profiles, frequency of mutations and / or fragmentation profiles. Consensus kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, cfDNA mutation profiles and frequency of mutations can be normalized against control samples. Measures of molecules mapping to reference sequences can also be compared across a genome to identify areas in the genome in which kmer repeat landscapes, kmer repeat landscapes combined with cfDNA mutation profiles, cfDNA mutation profiles and frequency of mutations varies, or remains the same. Consensus methods include, for example, linear or non-linear methods of building consensus kmer repeats, kmer repeat landscapes, cfDNA mutation profiles and frequency of mutations (such as voting, averaging, statistical, maximum a posteriori or maximum likelihood detection, dynamic programming, Bayesian, hidden Markov or support vector machine methods, etc.) derived from digital communication theory, information theory, or bioinformatics. After the sequence read coverage has been determined, a stochastic modeling algorithm is applied to convert the normalized nucleic acid sequence read coverage for each window region to the discrete copy number states. In some cases, this algorithm may comprise one or more of the following: Hidden Markov Model, dynamic programming, support vector machine, Bayesian network, trellis decoding, Viterbi decoding, expectation maximization, Kalman filtering methodologies and neural networks.38 161564170.1

[0153] Artificial neural networks (NNets) mimic networks of “neurons” based on the neural structure of the brain. They process records one at a time, or in a batch mode, and “learn” by comparing their classification of the record (which, at the outset, is largely arbitrary) with the known actual classification of the record. In MLP-NNets, the errors from the initial classification of the first record is fed back into the network, and are used to modify the network's algorithm the second time around, and so on for many iterations. The neural networks use an iterative learning process in which data cases (rows) are presented to the network one at a time, and the weights associated with the input values are adjusted each time.

[0154] After all cases are presented, the process often starts over again. During this learning phase, the network learns by adjusting the weights so as to be able to predict the correct class label of input samples. Neural network learning is also referred to as “connectionist learning,” due to connections between the units. Advantages of neural networks include their high tolerance to noisy data, as well as their ability to classify patterns on which they have not been trained. One neural network algorithm is back-propagation algorithm, such as Levenberg-Marquadt. Once a network has been structured for a particular application, that network is ready to be trained. To start this process, the initial weights are chosen randomly. Then the training, or learning, begins.

[0155] The network processes the records in the training data one at a time, using the weights and functions in the hidden layers, then compares the resulting outputs against the desired outputs. Errors are then propagated back through the system, causing the system to adjust the weights for application to the next record to be processed. This process occurs over and over as the weights are continually tweaked. During the training of a network the same set of data is processed many times as the connection weights are continually refined.

[0156] In an embodiment, the training step of the machine learning unit on the training data set may generate one or more classification models for applying to a test sample. These classification models may be applied to a test sample to predict the response of a subject to a therapeutic intervention.

[0157] Comparison of sequence coverage to a control sample or reference sequence may aid in normalization across windows. In this embodiment, genome DNA and / or cell free DNAs are extracted and isolated from a readily accessible bodily fluid such as blood. For example, cell free DNAs can be extracted using a variety of methods known in the art, including but not limited to isopropanol precipitation and / or silica based purification. DNAs may be extracted from any number39 161564170.1of subjects, such as subjects without cancer, subjects at risk for cancer, or subjects known to have cancer (e.g. through other means).

[0158] Following the isolation / extraction step, any of a number of different sequencing operations may be performed on the cell free polynucleotide sample. Samples may be processed before sequencing with one or more reagents (e.g., enzymes, unique identifiers (e.g., barcodes), probes, etc.). In some cases, if the sample is processed with a unique identifier such as a barcode, the samples or fragments of samples may be tagged individually or in subgroups with the unique identifier. The tagged sample may then be used in a downstream application such as a sequencing reaction by which individual molecules may be tracked to parent molecules.

[0159] The kmer repeats and / or cell free polynucleotides can be tagged or tracked in order to permit subsequent identification and origin of the particular polynucleotide. The assignment of an identifier (e.g., a barcode) to individual or subgroups of polynucleotides may allow for a unique identity to be assigned to individual sequences or fragments of sequences. This may allow acquisition of data from individual samples and is not limited to averages of samples. In some examples, nucleic acids or other molecules derived from a single strand may share a common tag or identifier and therefore may be later identified as being derived from that strand. Similarly, all of the fragments from a single strand of nucleic acid may be tagged with the same identifier or tag, thereby permitting subsequent identification of fragments from the parent strand. In other cases, gene expression products (e.g., mRNA) may be tagged in order to quantify expression, by which the barcode, or the barcode in combination with sequence to which it is attached can be counted. In still other cases, the systems and methods can be used as a PCR amplification control. In such cases, multiple amplification products from a PCR reaction can be tagged with the same tag or identifier. If the products are later sequenced and demonstrate sequence differences, differences among products with the same identifier can then be attributed to PCR error. Additionally, individual sequences may be identified based upon characteristics of sequence data for the read themselves. For example, the detection of unique sequence data at the beginning (start) and end (stop) portions of individual sequencing reads may be used, alone or in combination, with the length, or number of base pairs of each sequence read unique sequence to assign unique identities to individual molecules. Fragments from a single strand of nucleic acid, having been assigned a unique identity, may thereby permit subsequent identification of fragments from the parent strand. This can be used in conjunction with bottlenecking the initial starting genetic material to limit diversity.40 161564170.1

[0160] Generally, the methods and systems provided herein are useful for preparation of genomic and / or cell free polynucleotide sequences to a down-stream application sequencing reaction. Often, a sequencing method is next generation sequencing (NGS), classic Sanger sequencing, whole-genome bisulfite sequencing (WGSB), small-RNA sequencing, low-coverage Whole-Genome Sequencing (lcWGS), etc.

[0161] As used herein, the term “sequencing” refers to any of a number of technologies used to determine the sequence of a biomolecule, e.g., a nucleic acid such as DNA or RNA. Exemplary sequencing methods include, but are not limited to, targeted sequencing, single molecule real-time sequencing, exon sequencing, RNA-seq, electron microscopy-based sequencing, panel sequencing, transistor-mediated sequencing, direct sequencing, random shotgun sequencing, Sanger dideoxy termination sequencing, whole-genome sequencing, sequencing by hybridization, pyrosequencing, capillary electrophoresis, gel electrophoresis, duplex sequencing, cycle sequencing, single-base extension sequencing, solid-phase sequencing, high-throughput sequencing, massively parallel signature sequencing, emulsion PCR, co-amplification at lower denaturation temperature-PCR (COLD-PCR), multiplex PCR, sequencing by reversible dye terminator, paired-end sequencing, near-term sequencing, exonuclease sequencing, sequencing by ligation, short-read sequencing, single-molecule sequencing, sequencing-by-synthesis, real-time sequencing, reverse-terminator sequencing, nanopore sequencing, 454 sequencing, Solexa Genome Analyzer sequencing, SOLiD™ sequencing, MS-PET sequencing, and a combination thereof. In some embodiments, sequencing can be performer by a gene analyzer such as, for example, gene analyzers commercially available from Illumina or Applied Biosystems. In some embodiments, the sequencing method can be massively parallel sequencing, that is, simultaneously (or in rapid succession) sequencing any of at least 100, 1000, 10,000, 100,000, 1 million, 10 million, 100 million, or 1 billion polynucleotide molecules.

[0162] After sequencing, reads are assigned a quality score. A quality score may be a representation of reads that indicates whether those reads may be useful in subsequent analysis based on a threshold. In some cases, some reads are not of sufficient quality or length to perform the subsequent mapping step. Sequencing reads with a quality score at least 90%, 95%, 99%, 99.9%, 99.99% or 99.999% may be filtered out of the data set. In other cases, sequencing reads assigned a quality scored at least 90%, 95%, 99%, 99.9%, 99.99% or 99.999% may be filtered out of the data set. The genomic fragment reads that meet a specified quality score threshold are mapped to a reference genome, or a reference sequence that is known not to contain mutations. After mapping41 161564170.1alignment, sequence reads are assigned a mapping score. A mapping score may be a representation or reads mapped back to the reference sequence indicating whether each position is or is not uniquely mappable. In instances, reads may be sequences unrelated to mutation analysis. For example, some sequence reads may originate from contaminant polynucleotides. Sequencing reads with a mapping score at least 90%, 95%, 99%, 99.9%, 99.99% or 99.999% may be filtered out of the data set. In other cases, sequencing reads assigned a mapping scored less than 90%, 95%, 99%, 99.9%, 99.99% or 99.999% may be filtered out of the data set. For each mappable base, bases that do not meet the minimum threshold for mappability, or low quality bases, may be replaced by the corresponding bases as found in the reference sequence.

[0163] Numerous cancers may be detected using the methods and systems described herein. Cancers cells, as most cells, can be characterized by a rate of turnover, in which old cells die and replaced by newer cells. Generally dead cells, in contact with vasculature in a given subject, may release DNA or fragments of DNA into the blood stream. This is also true of cancer cells during various stages of the disease. Cancer cells may also be characterized, dependent on the stage of the disease, by various genetic aberrations such as copy number variation as well as mutations. This phenomenon may be used to detect the presence or absence of cancers individuals using the methods and systems described herein.

[0164] The types and number of cancers that may be detected may include but are not limited to blood cancers, brain cancers, lung cancers, skin cancers, nose cancers, throat cancers, liver cancers, bone cancers, lymphomas, pancreatic cancers, skin cancers, bowel cancers, rectal cancers, thyroid cancers, bladder cancers, kidney cancers, mouth cancers, stomach cancers, solid state tumors, heterogeneous tumors, homogenous tumors and the like.

[0165] Additionally, the systems and methods described herein may also be used to help characterize certain cancers. Genetic data produced from the system and methods of this disclosure may allow practitioners to help better characterize a specific form of cancer. Often times, cancers are heterogeneous in both composition and staging. Genetic profile data may allow characterization of specific sub-types of cancer that may be important in the diagnosis or treatment of that specific sub-type. This information may also provide a subject or practitioner clues regarding the prognosis of a specific type of cancer.

[0166] The systems and methods provided herein may be used to monitor already known cancers, or other diseases in a particular subject. This may allow either a subject or practitioner to42 161564170.1adapt treatment options in accord with the progress of the disease. In this example, the systems and methods described herein may be used to construct genetic profiles of a particular subject of the course of the disease. In some instances, cancers can progress, becoming more aggressive and genetically unstable. In other examples, cancers may remain benign, inactive or dormant. The system and methods of this disclosure may be useful in determining disease progression.

[0167] Further, the systems and methods described herein may be useful in determining the efficacy of a particular treatment option. In one example, successful treatment options may actually increase the amount of copy number variation or mutations detected in subject's blood if the treatment is successful as more cancers may die and shed DNA. In other examples, this may not occur. In another example, perhaps certain treatment options may be correlated with genetic profiles of cancers over time. This correlation may be useful in selecting a therapy. Additionally, if a cancer is observed to be in remission after treatment, the systems and methods described herein may be useful in monitoring residual disease or recurrence of disease.

[0168] The data is sent over a direct connection or over the internet to a computer for processing. The data processing aspects of the system can be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or in combinations of them. Data processing apparatus of the disclosure can be implemented in a computer program product tangibly embodied in a machine-readable storage device for execution by a programmable processor; and data processing method steps of the disclosure can be performed by a programmable processor executing a program of instructions to perform functions of the disclosure by operating on input data and generating output. The data processing aspects of the disclosure can be implemented advantageously in one or more computer programs that are executable on a programmable system including at least one programmable processor coupled to receive data and instructions from and to transmit data and instructions to a data storage system, at least one input device, and at least one output device. Each computer program can be implemented in a high-level procedural or object-oriented programming language, or in assembly or machine language, if desired; and, in any case, the language can be a compiled or interpreted language. Suitable processors include, by way of example, both general and special purpose microprocessors. Generally, a processor will receive instructions and data from a read-only memory and / or a random access memory. Storage devices suitable for tangibly embodying computer program instructions and data include all forms of nonvolatile memory, including by way of example semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices;43 161564170.1magnetic disks such as internal hard disks and removable disks; magneto-optical disks; and CD- ROM disks. Any of the foregoing can be supplemented by, or incorporated in, ASICs (application- specific integrated circuits).

[0169] To provide for interaction with a user, the methods can be implemented using a computer system having a display device such as a monitor or LCD (liquid crystal display) screen for displaying information to the user and input devices by which the user can provide input to the computer system such as a keyboard, a two-dimensional pointing device such as a mouse or a trackball, or a three-dimensional pointing device such as a data glove or a gyroscopic mouse. The computer system can be programmed to provide a graphical user interface through which computer programs interact with users. The computer system can be programmed to provide a virtual reality, three-dimensional display interface. EXAMPLES

[0170] Example 1: Genome-Wide Repeat Landscapes in Cancer and Cell-Free DNA

[0171] Changes to repeat sequences have long been implicated in the development of cancer. Transposable elements are thought to modulate gene expression and loss of their silencing via global hypomethylation in cancer may drive their movement7. This movement can cause oncogene activation and genomic instability8. Repeat types show differential enrichment in structural breakpoints. For example, tandem repeats are enriched in copy number variant (CNV) breakpoints while Alu repeats are enriched in deletion and duplication breakpoints9. The above changes to repeat elements are broadly characteristic of cancer genomes, but changes to individual element types have been observed in different cancer types10,11. Transposable elements serve as active enhancers for tissue-specific transcription factors dysregulated in cancers, and canonical tandem repeat expansions are associated with gene regulation and vary substantially between tumor sites of origin12,13. Instability and expansion of repeats in peri-centromeric and centromeric regions in cancer patients may drive chromosomal mis-segregation and other structural changes14–18which have been associated with lower overall survival19.

[0172] With the development of liquid biopsies for detection and genome-wide characterization of human cancer, analyses of repeat sequences have begun to be performed in cfDNA. Retrotransposable elements and non-telomeric satellite DNA have been shown to be highly represented in cfDNA20but liquid biopsy efforts to incorporate repeats have been limited to determination of overall cfDNA levels or assessment of aneuploidy21–24. Despite the above44 161564170.1advances, no systematic analysis of the entire compendium of repeat sequences has been performed in any human cancer, largely due to the inability to identify and quantify repeat sequences in a genome-wide fashion.

[0173] To address these challenges, ARTEMIS, Analysis of RepeaT EleMents in dISease, was developed as an alignment-free, genome-wide approach for analyzing repeat landscapes in short read sequencing. ARTEMIS assesses over 1200 individual repeat types that occur genome- wide and span 57 subfamilies comprising 6 families (Satellites, RNA elements, transposable elements, LINEs, SINEs, LTRs). In this study, ARTEMIS was used to show that repeat landscapes are enriched in genes commonly altered in human cancer and tumor-specific changes in repeats reflect a combination of structural and epigenetic changes in the cancer genome. Genome-wide repeat landscape analyses with ARTEMIS can be performed using low-coverage whole genome sequencing, permitting analysis of repeat landscapes in cfDNA for detection of human cancer.

[0174] RESULTS

[0175] De Novo Search of kmers in Genome-Wide Repeat Elements

[0176] To develop ARTEMIS we conducted a de novo search of short sequences (kmers) as we hypothesized that these would have enough complexity to identify the different types of repeat elements in the genome (FIGS. 1, 6). For example, a 24-bp kmer sequence can theoretically distinguish between 281 trillion (424) sequences. Using the recently obtained telomere to telomere (T2T) reference genome (chm13)5, 25assembled from long-read sequencing, we found that 4.73 billion 24-bp kmer sequences were present in the genome and 4.17 billion 24-bp kmers were unique to repeat elements overall. As related repeat elements have diverged in their sequence composition over time, we identified 1.1 billion 24-bp kmers that uniquely defined each of 1266 recently identified repeat types. To be included in this set, a kmer could neither occur in non-repeat regions of the genome, nor occur in multiple repeat types. Each of the 1266 repeat types analyzed were defined by a median of 43,29724-bp kmers spanning an average of 2.6 Mb of genome sequence (FIG. 7). We further included 58k 24-bp kmers from enhanced annotations of 14 human satellite subtypes26. These 1.1 billion kmers representing 1280 repeat types were found on all chromosomes and 98% of utilized kmers were only observed once in the T2T reference genome (FIG.8 A to 8C). These kmers also represented regions of the genome, like human satellites, that could not be aligned with high quality in typical short read next-generation sequencing. This allowed ARTEMIS to45 161564170.1consider the entirety of the genome, rather than only that from the ~60-85% of reads from next generation sequencing that can be aligned with high quality27, 28. To verify that these repeat landscape kmers would not be confounded by human-associated microbial genomes29, 30, we examined 1545 reference genomes representing common microbes, and found a median of 190 ARTEMIS kmers per microbial genome (range 0-2550), in all cases comprising <0.0003% of the 1.1 billion possible kmers counted in ARTEMIS. For analyses of an individual sample, we defined the kmer repeat landscape as the count of all kmers in a sequenced sample that matched each of the 1280 repeat types divided by the number of aligned sequence reads. As changes in repeat sequences may occur during initiation of cancer and other diseases, this comprehensive compendium of repeat features can be used to train machine learning models to distinguish genomes from normal and disease states.

[0177] Genome-wide enrichment of repeat kmers in cancer related genes and pathways

[0178] We first examined the genome-wide distribution of the 1.1 billion kmers defining unique repeat types and found that repeat elements were enriched in genes commonly altered in human cancer. Of the 736 genes in the COSMIC cancer driver gene census31, we found that 474 of these had a higher than expected number of repeat kmer sequences within their exonic or intronic sequences (Normalized Enrichment Score = 9.12, False Discovery Rate q-value = 0.00), including in genes amplified, deleted and rearranged in cancer (Normalized Enrichment Scores 1.92, 4.13, 6.45 and False Discovery Rate q-values 0.01, 0.00, 0.00 respectively). This enrichment remained significant even after correcting for the size of these genes (FIG.8D) and reflected an average 15- fold increase in repeat kmers in these regions (p<2.2e-16, Wilcoxon signed-rank test). In contrast, an analysis of the same number of randomly chosen genes in the genome did not show an enrichment of repeat kmer sequences (Normalized Enrichment Score = -1.0, False Discovery Rate q-value = 0.97). Repeat kmer sequences were also significantly increased in pathways commonly dysregulated in cancer including in cell adhesion, growth, and signaling, as well as cancer type specific gene sets. Together, these observations of repeat kmer localization suggest that alterations in key genes affecting oncogenic pathways in human cancer may be selected for during tumorigenesis using repeat-related genomic changes.

[0179] Kmer repeat landscapes are altered in cancer genomes

[0180] Given the broad number of genomic changes that occur during tumorigenesis, we evaluated whether kmer repeat landscapes were altered in cancers using short-read next generation46 161564170.1sequencing technologies. Given the challenges of distinguishing highly related repeat sequences, we simulated short-read whole genome sequence (WGS) data incorporating typical sequence error rates and analyzed these in an alignment-free fashion. We found that despite potential sequencing errors, the high complexity of kmer sequences allowed them to remain specific for their defined repeat family (98% of kmers counted were found in reads originating from their true repeat type) (FIGS.10 and 11).

[0181] We analyzed matched tumor and normal tissues of 333 cancer patients, including those with lung (n=86), colorectal (n=60), breast (n=91), liver (n=54), and ovarian cancer (n=42) from the pan-cancer analysis of whole genomes (PCAWG)32and determined whether genome-wide kmer counts for specific repeat element types were altered in the tumors. An average of 24.2 billion total kmers were identified in each sample sequenced at 30-60X, representing 1280 repeat elements. A median of 827 repeat elements (range 249-1246) had significantly increased or decreased kmer counts in cancers compared to their matched normal tissues (FIGS.2A, Methods). Nearly two thirds of altered elements (820 of 1280) had not been previously observed as being altered in human cancer (FIG. 2A, Table 1). Elements from Satellites, LINEs and SINEs were altered at the highest rates, though changes were also frequently observed in elements within LTRs, Transposable Elements, and RNA Elements. Nearly a quarter of the elements studied came from the largest repeat subfamily of LTRs, ERV1s, which are hypothesized to aberrantly activate transcription in cancer cells via onco-exaptation33. On average, more than half of the 300 LTR ERV1 elements were altered in all five types of tumors studied, though the individual altered elements varied across tissue types. While changes to 21 ERV1s have been described previously34, we observed changes in an additional 279 novel ERV1s across the cancer types analyzed (Table S6 and S7). Like other genomic changes in cancer genomes35, 36, changes in kmer repeat landscapes were highly complex, with no two patients studied having exactly the same set of alterations.

[0182] We analyzed matched tumor and normal tissues of 525 patients with cancer (table S4) from the Pan-Cancer Analysis of Whole Genomes (PCAWG) (Nature 578, 82–93 (2020), including those with breast (n=91), lung (n=86), colorectal (n=60), liver (n=54), thyroid (n=48), head and neck squamous cell (n=44), ovarian (n=42), gastric (n=38), bladder (n=23), cervical (n=20) and prostate (n=19) and determined whether genome-wide kmer counts for specific repeat element types were altered in the tumors. An average of 22.4 billion total kmers were identified in each sample sequenced at 30-60X, representing 1280 repeat elements. A median of 807 repeat elements47 161564170.1(range 246-1280) had increased or decreased kmer counts in tumors compared to their matched normal tissues (Fig.2A, tables S5 and S6). Nearly two thirds of altered elements (820 of 1280) had not been previously observed as being altered in human cancer (Fig.2A, Table 1). Elements from Satellites, LINEs and SINEs were altered at the highest rates, though changes were also frequently observed in elements within LTRs, Transposable Elements, and RNA Elements. Nearly a quarter of the elements studied came from the largest repeat subfamily of LTRs, ERV1s (table 1), which are hypothesized to aberrantly activate transcription in cancer cells by onco-exaptation, the process by which reactivated transposable elements can drive oncogene expression33. On average, more than 40% of the 300 LTR ERV1 elements were altered in all 12 types of tumors studied, though the individual altered elements varied across tissue types. Although changes to 21 ERV1s have been described previously34, we observed changes in an additional 279 new ERV1s across the cancer types analyzed. Like other large-scale changes in cancer genomes35, 36, changes in kmer repeat landscapes were highly complex, with no two patients studied having the same set of alterations.

[0183] We hypothesized that changes in kmer repeat landscapes would in part be related to structural changes that arise during tumorigenesis such as chromosomal copy number changes, rearrangements, or focal amplifications or deletions. Accordingly, we found that kmer counts reflected chromosomal arm gains or losses genome-wide in the tumors analyzed (r= 0.81; p<2.2e- 16, Spearman’s correlation) (FIG. 12). Additionally, tumors with more changes in kmer repeat landscapes had higher chromosomal instability as reflected through overall genomic entropy, loss of heterozygosity (LOH), non-modal ploidy fraction of the genome and other measures of genome- wide structural changes (r=0.34,p=1.7e-9; r=0.2, p=3.2e-4; r=0.32, p=2.9e-8, respectively, Spearman’s correlation) (FIG.2A). In contrast, tumor mutation burden (TMB), a measure of single base sequence changes in an individual cancer, was not correlated with genome-wide kmer repeat landscape changes (r=0.081, p=0.166, Spearman’s correlation).

[0184] Rearrangements, resulting from copy neutral translocations, as well as inversions, duplications, or deletions, may be facilitated by crossing over of homologous sequence37. An analysis of the locations of repeat elements and tumor-specific sequence breakpoints in the 333 samples analyzed identified an enrichment of 170 elements at breakpoint locations, comprising LINES, SINES, LTRs, TEs and RNA Elements, and including 92 elements that had never been previously implicated as being altered in cancer, suggesting that these elements may play a role in facilitating these structural changes (FIG. 2B). Analysis of focal amplifications of five or more48 161564170.1copies revealed that repeat element content across all subfamilies correlated with an increase in amplicon copy number (r=0.91, p<2.2e-16, Spearman’s correlation). As an example, analysis of the 1Mb region surrounding ERBB2 in breast tumors with known gains at this region revealed significant increases in 14 repeat elements, including in eight with no previously documented changes in cancer (FIGS.2C, 13A). Similarly, gains of the ~30 Mb region on chromosome 3q containing driver genes PIK3CA and SOX2 in squamous cell lung cancer38, 39revealed increases in kmers for repeat elements overlapping these regions, including in 9 elements not previously known to be altered in cancer (FIGS.13 B and 13C).

[0185] Changes in the content of repeat landscape were not fully explained by chromosomal or focal copy number changes or genomic rearrangements. After comparing changes in repeat elements to the segmented copy number alterations observed across the cancer genomes analyzed, we determined that 88% of the repeat changes (median 706 changes per tumor, range 232-1246) were larger in magnitude than that would be expected due to copy number gains and losses alone (FIG. 14). A set of 242 elements exhibited changes not explained by copy number changes in at least 75% of tumors studied. These types of changes include reduction of kmer elements through LINE-1 mediated deletions in squamous cell lung cancers and lower than expected repeat content in regions of copy number, consistent with the concept that such repeat sequences may undergo deletion as they facilitate gains in nearby genomic content (FIGS.2D, 12, 15)37, 40. Overall, these analyses highlight the ability of kmer repeat landscapers to detect and characterize a broad variety of structural changes in human cancer, including large chromosomal changes, commonly amplified or deleted driver gene regions, and alterations that directly target repeat sequences.

[0186] To evaluate the potential clinical implications of changes in repeat elements of cancer genomes, we examined whether tumor alterations in any of these families were associated with changes in overall survival or progression free survival for patients in the PCAWG dataset. We found that alterations in SINE elements were associated with longer overall (p=0.005) and progression-free (p=0.003) survival (FIG. 2E) and remained significant for overall survival even after adjusting for tumor type (FIGS.16 A and 16B). While the overall SINE change burden appears to be a pan-cancer marker, there were differences in the distribution of individual element changes between cancer types (FIG.16C). Interestingly, this change in patient outcomes was not observed for other non-repeat genome-wide metrics, including genomic entropy, LOH, or non-modal ploidy fraction (FIGS. 17 A and 17B). Our observations are consistent with previous analyses indicating49 161564170.1that reactivation and increase of repeat elements in cancer genomes may lead to increased immune responses41–43or genomic instability44, both mechanisms that could reduce tumor cell fitness and lead to improved patient outcomes.

[0187] Despite germline variability of repeat elements among different individuals (FIG. 18), ARTEMIS scores generated from a machine learning model of kmer repeat landscapes in unmatched tumor and normal samples distinguished the 333 PCAWG tumors from normal tissues with high performance across all cancer types analyzed (range AUC = 0.93 to >0.99) (FIG.19).

[0188] Kmer repeat landscapes in cfDNA

[0189] We sought to determine whether our approach to characterize the repeat landscape could be utilized for evaluating circulating cfDNA. Detection of repeats using low-coverage whole- genome sequencing would theoretically be achievable as ARTEMIS aggregates a large number kmer-defined repeat element instances throughout the genome while maintaining sufficient granularity to identify disease specific genomic features. As a first step in this analysis, we determined that repeat landscapes in PCAWG were highly consistent even if these were subsampled to different sequencing depths ranging from >60x to 1x coverage (FIGS. 20 and 21). We further found that kmer repeat landscapes in cfDNA were consistent across different sequencing platforms and experimental batches (FIG.22).

[0190] To determine if repeat landscapes could be quantified in the plasma using low- coverage sequencing of cfDNA, we first examined satellite families with known distributions on the Y chromosome (chrY) in a collection of male and female individuals (n=158). In the plasma of males (n=87), kmer counts for human satellite types known to be found exclusively or predominantly on chrY were substantially higher than in females (n=71)(p<2.2e-16 for all types) (Fig. 3A), while satellites not found on chrY showed no significant difference between males and females (p>0.1 for all types).

[0191] We then identified repeat elements with the largest changes across the PCAWG tumors, and evaluated the occurrences of these elements in cfDNA of individuals in prospectively collected diagnostic cohorts for patients at risk for lung or liver cancer (n=287 for lung cancer cohort; n=208 for liver cancer cohort) which had been previously sequenced27, 45. Across cohorts, many of the repeat kmer increases or decreases observed in tumors were evident in the plasma of patients with lung cancers of squamous or adenocarcinoma subtypes or liver cancer as compared to plasma from individuals without cancer (FIG. 3B). These included changes in elements with previously50 161564170.1documented roles in cancer such as LINE L1 elements, but also newly identified elements that have now been revealed to have alterations in cancer, including from subfamilies such as DNA-hAT- Charlie and LTR ERV1, ERVL-MaLR, and ERVL.

[0192] We hypothesized that repeat landscapes in cfDNA could be different from the expected repeat content in genomic DNA due to genome-wide chromatin and epigenetic changes that may alter the representation of cfDNA fragments in the circulation (27, 28, 45–49). We have previously shown that cfDNA fragmentation profiles reflect open and closed chromatin states genome-wide (28, 45). Here we analyzed cfDNA from 158 individuals without cancer and showed that regions with different histone marks had differential density of repeat element types (Fig.4, A and B), and that individual cfDNA fragments derived from regions with actively transcribed chromatin or activated histone marks had shorter lengths and exhibited lower coverage in the plasma (Fig.4, C and D). Overall, repeat landscape kmer counts in cfDNA for regions with high density of activating chromatin histone marks were lower than for regions with low density of these marks, whereas the reverse was observed for repressive histone marks (Fig.4E). Genome-wide simulations suggest that repeat landscapes in cfDNA may be influenced by both tumor-specific epigenomic and genomic changes (FIG.31).

[0193] ARTEMIS kmer repeat landscape analyses for cancer detection and monitoring

[0194] Given the ability to identify repeat landscapes changes in cfDNA, we evaluated the potential of the ARTEMIS method for noninvasive detection of cancer (FIG. 23). We previously described use of a sensitive and accessible whole genome cfDNA fragmentation test (DELFI) for lung and liver cancer screening in high-risk populations27, 45. Here, we used the kmer repeat landscapes and epigenetic profiles in cfDNA regions with high density of histone marks that differentially impact repeat representation (FIG.24) to generate a cross-validated ensemble machine learning model to detect lung cancer in the LUCAS prospectively collected diagnostic cohort (n=287) or liver cancer in a high-risk population (n=208), respectively27, 45. ARTEMIS classified lung cancer patients with an AUC of 0.82 (95% CI 0.77-0.87), and when combined with the DELFI genome-wide fragmentation features28, a joint ARTEMIS-DELFI model classified lung cancer patients with an AUC of 0.91 (95% CI 0.88-0.95) (FIGS.5A and 5B, 25). Similar performance was observed in the cohort of individuals at risk for liver cancer, where ARTEMIS detected individuals with liver cancer among patients with cirrhosis or viral hepatitis with an AUC of 0.87 (95% CI 0.82- 0.92), and when combined with DELFI the AUC improved to 0.91 (95% CI 0.87-0.95) (FIGS.2551 161564170.1and 26). We validated the locked ARTEMIS and the ARTEMIS-DELFI models in an external cohort comprised of non-cancer individuals at high and average risk of lung cancer as well patients with all stages of lung cancer and observed similar performance to that in the cross-validated training cohort (FIG. 5C). We further applied these models to an independent cohort of late-stage lung cancer patients receiving tyrosine kinase inhibitor therapy50and demonstrated that the ARTEMIS and joint ARTEMIS-DELFI scores were correlated to circulating tumor DNA (ctDNA) mutant allele fractions (MAFs) observed during therapy (r=0.70, p=2.67e-12 for ARTEMIS, r=0.80, p<2.2e-16 for ARTEMIS-DELFI, Spearman’s correlation) (FIG.27).

[0195] Finally, given the observation of tumor-specific changes in repeat landscapes, we evaluated whether ARTEMIS could aid tissue of origin determination in tumor or cfDNA samples of patients with cancer. We first examined whether kmer repeat landscapes could capture a tissue specific signal. We trained a machine learning model using kmer repeat landscapes to differentiate between tissue types, and found that it classified the PCAWG tumors by tissue of origin with an average 92% accuracy among the tumor types studied despite relying only on genomic features, which are typically thought to show less tissue-specific differences than transcriptomic and epigenetic features. This is consistent with the observations that while changes in repeat landscapes are a pan-cancer feature of cancer genomes, the specific repeat elements altered vary between tumor types (FIG. 2A). We then extended this approach to cfDNA and trained a cross-validated classification model using ARTEMIS-DELFI on cfDNA from a multi-cancer cohort including 226 individuals with breast, ovarian, lung, colorectal, bile duct, gastric or pancreatic tumors28. Here, despite the small number of samples used for training such a classifier, we found that ARTEMIS- DELFI correctly categorized detected patients among the different cancer types with an average 72% or 84% accuracy, for the highest or top two predictions, respectively (Table 2). ARTEMIS kmer repeat landscape analyses for cancer detection and monitoring

[0196] Given the ability to identify repeat landscapes changes in cfDNA, we evaluated the potential of the ARTEMIS method for noninvasive detection of cancer (FIG. 23). We previously described use of a sensitive and accessible whole genome cfDNA fragmentation test (DELFI) for lung and liver cancer screening in high-risk populations27, 45. Here, we used the kmer repeat landscapes and epigenetic profiles in cfDNA regions with high density of histone marks that differentially impact repeat representation (FIG.24) as features in machine learning models to detect lung cancer in the Danish Lung Cancer Screening Study (LUCAS) prospectively collected52 161564170.1diagnostic cohort (n=287) and liver cancer in a high-risk population (n=208)27, 45. ARTEMIS classified patients with lung cancer with an AUC of 0.82 (95% CI 0.78-0.87), and when ensembled with the DELFI genome-wide fragmentation features28, a joint ARTEMIS-DELFI model classified patients with lung cancer with an AUC of 0.91 (95% CI 0.88-0.94) (Fig. 5, A and B, FIG. 25). Similar performance was observed in the cohort of individuals at risk for liver cancer, where ARTEMIS detected individuals with liver cancer among patients with cirrhosis or viral hepatitis with an AUC of 0.87 (95% CI 0.82-0.93), and when combined with DELFI the AUC improved to 0.90 (95% CI 0.86-0.94) (FIG. 30). We validated the locked ARTEMIS and ARTEMIS-DELFI models in an external cohort comprised of non-cancer individuals at high and average risk of lung cancer (n=400) as well as patients with all stages of lung cancer (n=88) and observed similar performance to that in the cross-validated training cohort (Fig.5C, FIG.32). Analysis of a separate held-out set of patients from the LUCAS cohort with a prior history of cancer (n=25) using the locked ARTEMIS and ARTEMIS-DELFI models revealed higher scores in patients who experienced cancer recurrence compared to those who did not (FIG.33). We further applied these models to an independent cohort of patients with late-stage lung cancer (n=19) receiving tyrosine kinase inhibitor therapy50and demonstrated that the ARTEMIS and joint ARTEMIS-DELFI scores were correlated to circulating tumor DNA (ctDNA) mutant allele fractions (MAFs) observed during therapy (r=0.70, p=2.67x10-12for ARTEMIS, r=0.80, p<2.2x10-16for ARTEMIS-DELFI, Spearman’s correlation). Analysis of ARTEMIS-DELFI scores in patients at the first timepoint after initiation of treatment (median = 6 days) identified that those with scores above or below the pre- treatment median had shorter or longer progression free survival, respectively (median = 1.4 months for patients in the high score group vs 8.9 months for low score group, p<0.001, log-rank test, two sided) (FIG.34).

[0197] DISCUSSION

[0198] We have shown inter alia that ARTEMIS can reconstruct genome-wide repeat landscapes that reflect underlying changes in human cancer. The alterations reflect structural changes in the cancer genome including focal amplifications, deletions, copy-number changes, and rearrangements as well as direct changes in repeat elements. Through this analysis we found that repeat elements were enriched in the genome in genes commonly altered in human cancer, including at specific tumor-derived rearrangement breakpoints. Cancer-specific changes of the repeat landscape were observed genome-wide including in elements not previously known to be altered in53 161564170.1human cancer. Such elements may provide an underlying basis for structural alterations and the genomic instability of genes, pathways, and chromosomes widely altered in human cancers. Additionally, the expansion or contraction of repeat elements that can now be comprehensively identified provides a novel way to detect and examine mechanisms affecting cancer and other diseases.

[0199] We found that changes in repeat landscapes were detectable in the circulation and that the signal in plasma was further altered by epigenetic changes to repeat elements that influence their susceptibility to fragmentation. We and others have previously shown that changes in chromatin accessibility, transcription factor binding, and methylation can alter the representation of cfDNA in the blood28, 45–48. We show in this study that epigenetic states affected by histone acetylation and methylation, leading to altered levels of gene expression, have a profound impact on the size and coverage of cfDNA at distinct regions genome-wide, including in repeat regions. These analyses suggest that kmer repeat landscapes in plasma can reveal both structural and epigenetic changes in the genome.

[0200] Repeat landscapes for cfDNA-based detection of lung, liver and other cancers in cross-validated and externally validated settings suggests that ARTEMIS alone or in combination with other genome-wide features may provide an avenue for noninvasive detection, monitoring, and tissue of origin determination of cancer. In future work, it will be important to validate the ARTEMIS method for noninvasive detection of other cancers. One of the limitations of ARTEMIS is that it relies on evaluation of changes in repeat landscapes that are inherently variable among the germline of individuals3, 14, 51–53. However, ARTEMIS improved early-stage diagnosis by identifying genome-wide changes that would perhaps not be evident in other liquid biopsy approaches when tumor features such as mutations or arm-level changes are not detected. In the future, it will be valuable to characterize kmer repeat landscapes across diverse individuals as the chm13 reference genome is from a single individual and comparisons to a representative panel of healthy genotypes of different germline backgrounds could improve performance. Moreover, the functional significance of changes in repeat families remains poorly understood and could be improved through further analyses in cancer and other disease states. Only 43% of the genome-wide occurrences of kmers used in the ARTEMIS method are within ~28k known genes, and many of the repeat types in our landscapes have not been studied in human cancers. Given the size, diversity, and potential clinical significance of these regions of the genome, our study offers unique insights54 161564170.1into the cancer genome and provides a proof-of-concept for the utility of genome-wide kmer repeat landscapes as tissue and blood-based biomarkers.55 161564170.1

[0201] References (for the above Example 1 and disclosure above as shown in numerical superscripts to correspond to the reference numbers shown in the immediately following listing) 1. M. R. Vollger, X. Guitart, P. C. Dishuck, L. Mercuri, W. T. Harvey, A. Gershman, M. Diekhans, A. Sulovari, K. M. Munson, A. P. Lewis, K. Hoekzema, D. Porubsky, R. Li, S. Nurk, S. Koren, K. H. Miga, A. M. Phillippy, W. Timp, M. Ventura, E. E. Eichler, Segmental duplications and their variation in a complete human genome. Science. 376, eabj6965 (2022). 2. S. Aganezov, S. M. Yan, D. C. Soto, M. Kirsche, S. Zarate, P. Avdeyev, D. J. Taylor, K. Shafin, A. Shumate, C. Xiao, J. Wagner, J. McDaniel, N. D. Olson, M. E. G. Sauria, M. R. Vollger, A. Rhie, M. Meredith, S. Martin, J. Lee, S. Koren, J. A. Rosenfeld, B. Paten, R. Layer, C.-S. Chin, F. J. Sedlazeck, N. F. Hansen, D. E. Miller, A. M. Phillippy, K. H. Miga, R. C. McCoy, M. Y. Dennis, J. M. Zook, M. C. Schatz, A complete reference genome improves analysis of human genetic variation. Science.376, eabl3533 (2022). 3. S. J. Hoyt, J. M. Storer, G. A. Hartley, P. G. S. Grady, A. Gershman, L. G. de Lima, C. Limouse, R. Halabian, L. Wojenski, M. Rodriguez, N. Altemose, A. Rhie, L. J. Core, J. L. Gerton, W. Makalowski, D. Olson, J. Rosen, A. F. A. Smit, A. F. Straight, M. R. Vollger, T. J. Wheeler, M. C. Schatz, E. E. Eichler, A. M. Phillippy, W. Timp, K. H. Miga, R. J. O’Neill, From telomere to telomere: The transcriptional and epigenetic state of human repeat elements. Science.376, eabk3112 (2022). 4. A. Gershman, M. E. G. Sauria, X. Guitart, M. R. Vollger, P. W. Hook, S. J. Hoyt, M. Jain, A. Shumate, R. Razaghi, S. Koren, N. Altemose, G. V. Caldas, G. A. Logsdon, A. Rhie, E. E. Eichler, M. C. Schatz, R. J. O’Neill, A. M. Phillippy, K. H. Miga, W. Timp, Epigenetic patterns in a complete human genome. Science.376, eabj5089 (2022).56 161564170.15. S. Nurk, S. Koren, A. Rhie, M. Rautiainen, A. V. Bzikadze, A. Mikheenko, M. R. Vollger, N. Altemose, L. Uralsky, A. Gershman, S. Aganezov, S. J. Hoyt, M. Diekhans, G. A. M. S. E. M. G. G. S. Y. G. V. G. K. M. P. A. C. J. S. W. S. M. de A. and Rich C.uk. 13, 1537 (2023).57 161564170.111. J. G. Martínez, J. Pérez-Escuredo, P. Castro-Santos, C. Á. Marcos, J. L. L. Pendás, M. F. Fraga, M. A. Hermsen, Hypomethylation of LINE-1, and not centromeric SAT-α, is associated with centromeric instability in head and neck squamous cell carcinoma. Cell Oncol.35, 259–267 (2012). 12. K. Karttunen, D. Patel, J. Xia, L. Fei, K. Palin, L. Aaltonen, B. Sahu, Biorxiv, in press, doi:10.1101 / 2022.12.16.520732. 13. G. S. Erwin, G. Gürsoy, R. Al-Abri, A. Suriyaprakash, E. Dolzhenko, K. Zhu, C. R. Hoerner, S. M. White, L. Ramirez, A. Vadlakonda, A. Vadlakonda, K. von Kraut, J. Park, C. M. Brannon, D. A. Sumano, R. A. Kirtikar, A. A. Erwin, T. J. Metzner, R. K. C. Yuen, A. C. Fan, J. T. Leppert, M. A. Eberle, M. Gerstein, M. P. Snyder, Recurrent repeat expansions in human cancer genomes. Nature.613, 96–102 (2023).14. L. G. de Lima, E. Howe, V. P. Singh, T. Potapova, H. Li, B. Xu, J. Castle, S. Crozier, C. J. Harrison, S. C. Clifford, K. H. Miga, S. L. Ryan, J. L. Gerton, PCR amplicons identify widespread copy number variation in human centromeric arrays and instability in cancer. Cell Genom.1, 100064 (2021). 15. A. K. Saha, M. Mourad, M. H. Kaplan, I. Chefetz, S. N. Malek, R. Buckanovich, D. M. Markovitz, R. Contreras-Galindo, The Genomic Landscape of Centromeres in Cancers. Sci Rep-uk.9, 11259 (2019). 16. S. Decombe, F. Loll, L. Caccianini, K. Affannoukoué, I. Izeddin, J. Mozziconacci, C. Escudé, J. Lopes, Epigenetic rewriting at centromeric DNA repeats leads to increased chromatin accessibility and chromosomal instability. Epigenet Chromatin.14, 35 (2021). 17. P. Ly, S. F. Brunner, O. Shoshani, D. H. Kim, W. Lan, T. Pyntikova, A. M. Flanagan, S. Behjati, D. C. Page, P. J. Campbell, D. W. Cleveland, Chromosome segregation errors generate a diverse spectrum of simple and complex genomic rearrangements. Nat Genet.51, 705–715 (2019). 18. K. Ichida, K. Suzuki, T. Fukui, Y. Takayama, N. Kakizawa, F. Watanabe, H. Ishikawa, Y. Muto, T. Kato, M. Saito, K. Futsuhara, Y. Miyakura, H. Noda, T. Ohmori, F. Konishi, T. Rikiyama, Overexpression of satellite alpha transcripts leads to chromosomal instability via segregation errors at specific chromosomes. Int J Oncol.52, 1685–1693 (2018). 19. F. Bersani, E. Lee, P. V. Kharchenko, A. W. Xu, M. Liu, K. Xega, O. C. MacKenzie, B. W. Brannigan, B. S. Wittner, H. Jung, S. Ramaswamy, P. J. Park, S. Maheswaran, D. T. Ting, D. A. Haber, Pericentromeric satellite repeat expansions through RNA-derived DNA intermediates in cancer. Proc National Acad Sci.112, 15148–15153 (2015). 20. S. Grabuschnig, J. Soh, P. Heidinger, T. Bachler, E. Hirschböck, I. R. Rodriguez, D. Schwendenwein, C. W. Sensen, Circulating cell-free DNA is predominantly composed of58 161564170.1retrotransposable elements and non-telomeric satellite DNA. J Biotechnol.313, 48–56 (2020). 21. U. Gezer, A. J. Bronkhorst, S. Holdenrieder, The Utility of Repetitive Cell-Free DNA in Cancer Liquid Biopsies. Diagnostics.12, 1363 (2022). 22. C. Douville, J. D. Cohen, J. Ptak, M. Popoli, J. Schaefer, N. Silliman, L. Dobbyn, R. E. Schoen, J. Tie, P. Gibbs, M. Goggins, C. L. Wolfgang, T.-L. Wang, I.-M. Shih, R. Karchin, A. M. Lennon, R. H. Hruban, C. Tomasetti, C. Bettegowda, K. W. Kinzler, N. Papadopoulos, B. Vogelstein, Assessing aneuploidy with repetitive element sequencing. Proc National Acad Sci.117, 4858–4863 (2020). 23. C. Douville, S. Springer, I. Kinde, J. D. Cohen, R. H. Hruban, A. M. Lennon, N. Papadopoulos, K. W. Kinzler, B. Vogelstein, R. Karchin, Detection of aneuploidy in patients with cancer through amplification of long interspersed nucleotide elements (LINEs). Proc National Acad Sci.115, 1871–1876 (2018). 24. C. Rago, D. L. Huso, F. Diehl, B. Karim, G. Liu, N. Papadopoulos, Y. Samuels, V. E. Velculescu, B. Vogelstein, K. W. Kinzler, L. A. Diaz, Serial Assessment of Human Tumor Burdens in Mice by the Analysis of Circulating DNA. Cancer Res.67, 9364–9370 (2007). 25. A. Rhie, S. Nurk, M. Cechova, S. J. Hoyt, D. J. Taylor, N. Altemose, P. W. Hook, S. Koren, M. Rautiainen, I. A. Alexandrov, J. Allen, M. Asri, A. V. Bzikadze, N.-C. Chen, C.-S. Chin, M. Diekhans, P. Flicek, G. Formenti, A. Fungtammasan, C. G. Giron, E. Garrison, A. Gershman, J. Gerton, P. G. S. Grady, A. Guarracino, L. Haggerty, R. Halabian, N. F. Hansen, R. Harris, G. A. Hartley, W. T. Harvey, M. Haukness, J. Heinz, T. Hourlier, R. M. Hubley, S. E. Hunt, S. Hwang, M. Jain, R. K. Kesharwani, A. P. Lewis, H. Li, G. A. Logsdon, J. K. Lucas, W. Makalowski, C. Markovic, F. J. Martin, A. M. M. Cartney, R. C. McCoy, J. McDaniel, B. M. McNulty, P. Medvedev, A. Mikheenko, K. M. Munson, T. D. Murphy, H. E. Olsen, N. D. Olson, L. F. Paulin, D. Porubsky, T. Potapova, F. Ryabov, S. L. Salzberg, M. E. G. Sauria, F. J. Sedlazeck, K. Shafin, V. A. Shepelev, A. Shumate, J. M. Storer, L. Surapaneni, A. M. T. Oill, F. Thibaud-Nissen, W. Timp, M. Tomaszkiewicz, M. R. Vollger, B. P. Walenz, A. C. Watwood, M. H. Weissensteiner, A. M. Wenger, M. A. Wilson, S. Zarate, Y. Zhu, J. M. Zook, E. E. Eichler, R. O’Neill, M. C. Schatz, K. H. Miga, K. D. Makova, A. M. Phillippy, Biorxiv, in press, doi:10.1101 / 2022.12.01.518724. 26. N. Altemose, K. H. Miga, M. Maggioni, H. F. Willard, Genomic Characterization of Large Heterochromatic Gaps in the Human Genome Assembly. Plos Comput Biol. 10, e1003628 (2014). 27. D. Mathios, J. S. Johansen, S. Cristiano, J. E. Medina, J. Phallen, K. R. Larsen, D. C. Bruhm, N. Niknafs, L. Ferreira, V. Adleff, J. Y. Chiao, A. Leal, M. Noe, J. R. White, A. S. Arun, C. Hruban, A. V. Annapragada, S. Ø. Jensen, M.-B. W. Ørntoft, A. H. Madsen, B. Carvalho, M. de Wit, J. Carey, N. C. Dracopoli, T. Maddala, K. C. Fang, A.-R. Hartman, P.59 161564170.1M. Forde, V. Anagnostou, J. R. Brahmer, R. J. A. Fijneman, H. J. Nielsen, G. A. Meijer, C. L. Andersen, A. Mellemgaard, S. E. Bojesen, R. B. Scharpf, V. E. Velculescu, Detection and characterization of lung cancer using cell-free DNA fragmentomes. Nat Commun. 12, 5060 (2021). 28. S. Cristiano, A. Leal, J. Phallen, J. Fiksel, V. Adleff, D. C. Bruhm, S. Ø. Jensen, J. E. Medina, C. Hruban, J. R. White, D. N. Palsgrove, N. Niknafs, V. Anagnostou, P. Forde, J. Naidoo, K. Marrone, J. Brahmer, B. D. Woodward, H. Husain, K. L. van Rooijen, M.-B. W. Ørntoft, A. H. Madsen, C. J. H. van de Velde, M. Verheij, A. Cats, C. J. A. Punt, G. R. Vink, N. C. T. van Grieken, M. Koopman, R. J. A. Fijneman, J. S. Johansen, H. J. Nielsen, G. A. Meijer, C. L. Andersen, R. B. Scharpf, V. E. Velculescu, Genome-wide cell-free DNA fragmentation in patients with cancer. Nature.570, 385–389 (2019). 29. B. A. Methé, K. E. Nelson, M. Pop, …, R. K. Wilson, O. White, A framework for human microbiome research. Nature.486, 215–21 (2011). 30. C. Huttenhower, D. Gevers, R. Knight, …, R. K. Wilson, O. White, Structure, function and diversity of the healthy human microbiome. Nature.486, 207–214 (2012). 31. J. G. Tate, S. Bamford, H. C. Jubb, Z. Sondka, D. M. Beare, N. Bindal, H. Boutselakis, C. G. Cole, C. Creatore, E. Dawson, P. Fish, B. Harsha, C. Hathaway, S. C. Jupe, C. Y. Kok, K. Noble, L. Ponting, C. C. Ramshaw, C. E. Rye, H. E. Speedy, R. Stefancsik, S. L. Thompson, S. Wang, S. Ward, P. J. Campbell, S. A. Forbes, COSMIC: the Catalogue Of Somatic Mutations In Cancer. Nucleic Acids Res.47, gky1015- (2018). 32. The ICGC / TCGA Pan-Cancer Analysis of Whole Genomes Consortium, Pan-cancer analysis of whole genomes. Nature.578, 82–93 (2020). 33. A. Babaian, D. L. Mager, Endogenous retroviral promoter exaptation in human cancer. Mob. DNA.7, 24 (2016). 34. H. S. Jang, N. M. Shah, A. Y. Du, Z. Z. Dailey, E. C. Pehrsson, P. M. Godoy, D. Zhang, D. Li, X. Xing, S. Kim, D. O’Donnell, J. I. Gordon, T. Wang, Transposable elements drive widespread expression of oncogenes in human cancers. Nat. Genet. 51, 611–617 (2019). 35. T. Sjöblom, S. Jones, L. D. Wood, D. W. Parsons, J. Lin, T. D. Barber, D. Mandelker, R. J. Leary, J. Ptak, N. Silliman, S. Szabo, P. Buckhaults, C. Farrell, P. Meeh, S. D. Markowitz, J. Willis, D. Dawson, J. K. V. Willson, A. F. Gazdar, J. Hartigan, L. Wu, C. Liu, G. Parmigiani, B. H. Park, K. E. Bachman, N. Papadopoulos, B. Vogelstein, K. W. Kinzler, V. E. Velculescu, The Consensus Coding Sequences of Human Breast and Colorectal Cancers. Science. 314, 268–274 (2006).60 161564170.136. L. D. Wood, D. W. Parsons, S. Jones, J. Lin, T. Sjöblom, R. J. Leary, D. Shen, S. M. Boca, T. Barber, J. Ptak, N. Silliman, S. Szabo, Z. Dezso, V. Ustyanksky, T. Nikolskaya, Y. Nikolsky, R. Karchin, P. A. Wilson, J. S. Kaminker, Z. Zhang, R. Croshaw, J. Willis, D. Dawson, M. Shipitsin, J. K. V. Willson, S. Sukumar, K. Polyak, B. H. Park, C. L. Pethiyagoda, P. V. K. Pant, D. G. Ballinger, A. B. Sparks, J. Hartigan, D. R. Smith, E. Suh, N. Papadopoulos, P. Buckhaults, S. D. Markowitz, G. Parmigiani, K. W. Kinzler, V. E. Velculescu, B. Vogelstein, The Genomic Landscapes of Human Breast and Colorectal Cancers. Science.318, 1108–1113 (2007). 37. B. Burssed, M. Zamariolli, F. T. Bellucco, M. I. Melaragno, Mechanisms of structural chromosomal rearrangement formation. Mol. Cytogenet.15, 23 (2022). 38. J. Qian, P. P. Massion, Role of Chromosome 3q Amplification in Lung Cancer. J Thorac Oncol.3, 212–215 (2008). 39. T. Hussenet, S. Dali, J. Exinger, B. Monga, B. Jost, D. Dembelé, N. Martinet, C. Thibault, J. Huelsken, E. Brambilla, S. du Manoir, SOX2 Is an Oncogene Activated by Recurrent 3q26.3 Amplifications in Human Lung Squamous Cell Carcinomas. Plos One. 5, e8960 (2010). 40. E. Rheinbay, M. M. Nielsen, F. Abascal, …, M. J. van de Vijver, L. van’t Veer, C. von Mering, Analyses of non-coding somatic drivers in 2,658 cancer whole genomes. Nature. 578, 102–111 (2020). 41. K. B. Chiappinelli, P. L. Strissel, A. Desrichard, H. Li, C. Henke, B. Akman, A. Hein, N. S. Rote, L. M. Cope, A. Snyder, V. Makarov, S. Budhu, S. Buhu, D. J. Slamon, J. D. Wolchok, D. M. Pardoll, M. W. Beckmann, C. A. Zahnow, T. Merghoub, T. Mergoub, T. A. Chan, S. B. Baylin, R. Strick, Inhibiting DNA Methylation Causes an Interferon Response in Cancer via dsRNA Including Endogenous Retroviruses. Cell.162, 974–86 (2014). 42. M. Russo, S. Morelli, G. Capranico, Expression of down-regulated ERV LTR elements associates with immune activation in human small-cell lung cancers. Mob. DNA.14, 2 (2023). 43. M. Onishi-Seebacher, G. Erikson, Z. Sawitzki, D. Ryan, G. Greve, M. Lübbert, T. Jenuwein, Repeat to gene expression ratios in leukemic blast cells can stratify risk prediction in acute myeloid leukemia. BMC Méd. Genom.14, 166 (2021). 44. J. R. Kemp, M. S. Longworth, Crossing the LINE Toward Genomic Instability: LINE-1 Retrotransposition in Cancer. Front. Chem.3, 68 (2015). 45. Z. H. Foda, A. V. Annapragada, K. Boyapati, D. C. Bruhm, N. A. Vulpescu, J. E. Medina, D. Mathios, S. Cristiano, N. Niknafs, H. T. Luu, M. G. Goggins, R. A. Anders, J. Sun, S. H. Mehta, D. L. Thomas, G. D. Kirk, V. Adleff, J. Phallen, R. B. Scharpf, A. K. Kim, V. E.61 161564170.1Velculescu, Detecting liver cancer using cell-free DNA fragmentomes. Cancer Discov (2022), doi:10.1158 / 2159-8290.cd-22-0659. 46. M. W. Snyder, M. Kircher, A. J. Hill, R. M. Daza, J. Shendure, Cell-free DNA Comprises an In Vivo Nucleosome Footprint that Informs Its Tissues-Of-Origin. Cell. 164, 57–68 (2016). 47. P. Ulz, G. G. Thallinger, M. Auer, R. Graf, K. Kashofer, S. W. Jahn, L. Abete, G. Pristauz, E. Petru, J. B. Geigl, E. Heitzer, M. R. Speicher, Inferring expressed genes by whole- genome sequencing of plasma DNA. Nat Genet.48, 1273–1278 (2016). 48. Q. Zhou, G. Kang, P. Jiang, R. Qiao, W. K. J. Lam, S. C. Y. Yu, M.-J. L. Ma, L. Ji, S. H. Cheng, W. Gai, W. Peng, H. Shang, R. W. Y. Chan, S. L. Chan, G. L. H. Wong, L. T. Hiraki, S. Volpi, V. W. S. Wong, J. Wong, R. W. K. Chiu, K. C. A. Chan, Y. M. D. Lo, Epigenetic analysis of cell-free DNA by fragmentomic profiling. Proc National Acad Sci. 119, e2209852119 (2022). 49. S. Y. Shen, R. Singhania, G. Fehringer, A. Chakravarthy, M. H. A. Roehrl, D. Chadwick, P. C. Zuzarte, A. Borgida, T. T. Wang, T. Li, O. Kis, Z. Zhao, A. Spreafico, T. da S. Medina, Y. Wang, D. Roulois, I. Ettayebi, Z. Chen, S. Chow, T. Murphy, A. Arruda, G. M. O’Kane, J. Liu, M. Mansour, J. D. McPherson, C. O’Brien, N. Leighl, P. L. Bedard, N. Fleshner, G. Liu, M. D. Minden, S. Gallinger, A. Goldenberg, T. J. Pugh, M. M. Hoffman, S. V. Bratman, R. J. Hung, D. D. D. Carvalho, Sensitive tumour detection and classification using plasma cell-free DNA methylomes. Nature.563, 579–583 (2018). 50. J. Phallen, A. Leal, B. D. Woodward, P. M. Forde, J. Naidoo, K. A. Marrone, J. R. Brahmer, J. Fiksel, J. E. Medina, S. Cristiano, D. N. Palsgrove, C. D. Gocke, D. C. Bruhm, P. Keshavarzian, V. Adleff, E. Weihe, V. Anagnostou, R. B. Scharpf, V. E. Velculescu, H. Husain, Early Noninvasive Detection of Response to Targeted Therapy in Non–Small Cell Lung Cancer. Cancer Res.79, 1204–1213 (2019). 51. N. Altemose, A classical revival: Human satellite DNAs enter the genomics era. Semin Cell Dev Biol.128, 2–14 (2022). 52. J. R. Iben, R. J. Maraia, tRNA gene copy number variation in humans. Gene.536, 376– 384 (2014). 53. W.-W. Liao, M. Asri, J. Ebler, …, H. Li, B. Paten, A draft human pangenome reference. Nature.617, 312–324 (2023). 54. D. C. Bruhm, D. Mathios, Z. H. Foda, A. V. Annapragada, J. E. Medina, V. Adleff, E. J. Chiao, L. Ferreira, S. Cristiano, J. R. White, S. A. Mazzilli, E. Billatos, A. Spira, A. H. Zaidi, J. Mueller, A. K. Kim, V. Anagnostou, J. Phallen, R. B. Scharpf, V. E. Velculescu, Single-62 161564170.1molecule genome-wide mutation profiles of cell-free DNA for non-invasive detection of cancer. Nat. Genet., 1–10 (2023).

[0202] Example 2

[0203] This example describes the methods utilized in Example 1 and additional results.

[0204] Study Populations

[0205] We obtained matched tumor and normal BAM files for the 343 lung, liver, ovarian, colorectal and breast cancer patients in PCAWG that were available in the Protected Data Cloud2and excluded 10 patients for which either the tumor or normal was on the PCAWG blacklist (Table S12). We then used Samtools to convert the BAM files to FASTQ for use in the ARTEMIS pipeline.

[0206] We analyzed whole-genome sequencing (1-2x coverage) data from cfDNA from 819 individuals with and without lung cancer from four cohorts, 208 individuals with and without liver cancer, and 423 individuals from a multi-cancer cohort of patients without cancer and with breast, ovarian, lung, bile duct, colorectal, gastric, duodenal and pancreatic tumors described in our previous publications3–6(FIG. 23). Briefly, the LUCAS Cohort was a prospectively collected group of 365 patients examined at Bispebjerg Hospital in Copenhagen, Denmark who presented with a positive imaging finding on a chest x-ray or CT.287 of these patients were used for early detection analyses as described in Mathios et al.3. The liver cancer cohort consisted of 208 individuals with hepatocellular carcinoma or at high risk of liver cancer (Cirrhosis, Hepatitis B or Hepatitis C).170 of these were prospectively collected as part of the HCC biomarker registry and the AIDS Linked to the IntraVenous Experience (ALIVE) study at the Johns Hopkins University School of Medicine6. 38 patients with HBV or cirrhosis were retrospectively collected by BioIVT (Westbury, NY). The external validation cohorts were comprised of 385 non-cancer individuals from a screening cohort for colorectal cancer in Denmark (Endoscopy III) and the Netherlands (COCOS, Netherlands trial register ID NTR1829)7, lung cancer patients and individuals at high risk for lung cancer from the Alleghany Health Network, Boston University and Detection of Early Lung Cancer Among Military Personnel (DECAMP) Consortium8, and patients with lung cancer from BioIVT (n=57) (Westbury, NY). The lung cancer monitoring cohort included 19 patients at the University of California San Diego or Johns Hopkins University being treated with tyrosine kinase inhibitors, at 75 time points with both targeted sequencing and whole-genome sequencing available as previously described4, 5. The multi-cancer cohort consisted of samples from ILSBio / Bioreclamation, Aarhus University,63 161564170.1Herlev Hospital of the University of Copenhagen, Hvidovre Hospital, the University Medical Center of the University of Utrecht, the Academic Medical Center of the University of Amsterdam, the Netherlands Cancer Institute and the University of California, San Diego. Collection of patient samples used in this study conformed to all relevant ethical regulations. Collection protocols were approved by the Danish Regional Ethics Committee and the Danish Data Protection Agency (LUCAS cohort, Endoscopy III samples), the Dutch Health Council (COCOS samples), the Human Research Protection Office of the Department (DECAMP samples), and Institutional Review Boards at the above named institutions (AHN samples, liver cohort, lung cancer monitoring cohort, and multi-cancer cohort). All patients provided written informed consent and the studies were performed according to the Declaration of Helsinki.

[0207] The protocols for blood collection, cfDNA extraction and sequencing have been previously described3, 4, 6, 9. In brief, whole blood was collected in EDTA or Streck tubes, and plasma was separated by centrifugation and aliquoted into EDTA tubes. cfDNA was isolated from 2-4mL of plasma, and next-generation sequencing libraries were prepared using 15ng of cfDNA when available, or the entire purified amount if less than 15ng were available. All libraries underwent four cycles of PCR amplification, except for the lung cancer monitoring samples and the multi-cancer cohort samples which underwent 12 cycles. All cfDNA data used for modeling was sequenced at 1- 2x coverage on the Illumina HiSeq2500 platform using 100bp paired-end runs. For the LUCAS cohort, we analyzed the cohort sequenced on both the HiSeq2500 and NovaSeq6000 to allow a technical comparison of sequencing replicates.

[0208] de novo kmer finding

[0209] We first extracted all repeat sequences and coordinates for known repeat element types from the RepeatMasker track in chm13 (T2T-CHM13v2.0). We excluded repeats from the families Low Complexity, Unknown and Simple Repeats, leaving 1287 types of repeats across 57 subfamilies comprising 13 families. For simplicity, we aggregated all elements in the families tRNA, srpRNA, snRNA, scRNA and rRNA as RNA Elements, and the families DNA, DNA?, Retroposon and RC as Transposable Elements leaving 6 overall families (LINE, SINE, LTR, Satellites, Transposable Elements, RNA Elements).

[0210] We then performed a de novo kmer finding procedure inspired by Altemose et al.1. We used Jellyfish10to count all unique 24-mers occurring in each of the 1287 types of repeats, as well as those occurring in the portions of the genome excluding all repeat regions. We then selected64 161564170.1all kmers that occurred only in a single repeat type and that were not present in the non-repeat regions of the genome. The kmers for a sequence and its reverse complement were counted together as the reference genome represents one strand, but we expect that half of the paired-end reads were derived from the reverse complement strand. We identified at least one unique kmer in 1266 of the 1287 repeat types. We additionally included 58,426 kmers from 14 HSATII and HSATIII subfamilies defined in Altemose et al.1supplement the RepeatMasker Satellite annotations. These kmers overlapped with broader satellite types in the RepeatMasker track but we allowed these kmers to be counted in multiple repeat types for consistency with the previous publication. In total, we identified 1,140,845,806 distinct kmers defining the 1280 repeat element types. To verify that these kmers had low co-occurrence in common human-associated microbial genomes, we counted kmers in 1545 microbial genomes from the Human Microbiome Project available for download on NCBI Entrez11,12.

[0211] Generation of kmer repeat landscapes

[0212] We obtained all sequencing reads for each sample, counting each unique kmer and its reverse complement, and aggregating the kmer counts for each repeat type. We normalized the aggregated counts to the number of reads that were aligned with MAPQ >= 30 (samtools view -c -q 30 -F 3844). Our approach considered all reads, including those from portions of the genome not provided in hg19 and / or repeat types that were not aligned.

[0213] ARTEMIS Machine Learning Models in Tissue

[0214] We centered and scaled the coverage normalized counts of the kmer repeat landscape for each tumor and normal tissue sample and trained a penalized logistic regression model to generate a cross-validated ARTEMIS score (for each sample, the ARTEMIS score was calculated as the mean across 10 repeats of 5-fold cross validation) for distinguishing tumor from normal tissue samples. We further used kmer repeat landscapes to train a multi-class gradient boosted model (GBM) to generate cross-validated (5-fold cross-validation) predictions of tumor tissues of origin (for each sample, the model generated a vector of multinomial probabilities, where each element corresponded to a possible tumor tissue of origin and the predicted class was chosen based on the element with maximum value).

[0215] ARTEMIS for early detection, tissue of origin, and monitoring of cancer in cfDNA65 161564170.1

[0216] We obtained a kmer repeat landscape for each sample using the 786 features with more than 1000 kmers per million aligned reads expected. This filtering was employed because at low coverage features with low abundance have greater technical variation (FIG. 23). In order to accommodate ensembling of diverse feature classes, we used nested cross-validation to generate the ARTEMIS score. The inner cross-validation loop trained six penalized logistic regression (PLR) models (Lasso regression, ∝ = 1, penalty chosen in the range 0.00001 - 0.1 by re-sampling within each cross-validation fold) with repeat landscapes as features (a PLR for each of five repeat families and a PLR for the epigenetic profile, Supplementary Materials and Methods). The outer cross- validation loop was trained with a leave-one-individual-out architecture; we ensembled the six scores available for each of the N-1 individuals using a PLR model. The score obtained by applying this PLR model to the 6 scores for the held-out patient’s features is the ARTEMIS Score for cancer detection in cfDNA.

[0217] To incorporate DELFI fragmentation profiles we ensembled the ARTEMIS score with three additional models: a PLR model using principal component analysis of the ratio of short to long fragments in 5 Mb bins genome wide (D. Mathios et al., Nat Commun 12, 5060 (2021)), a PLR model on 39 chromosomal arm z-scores for aneuploidy (D. Mathios et al., Nat Commun 12, 5060 (2021)) and a gradient boosted model on coverage in 5 Mb bins genome wide (28). This combined ensemble produced a joint ARTEMIS-DELFI score. We retrained the lung cancer model on the full LUCAS cohort and then applied the locked models to four external validation cohorts: The JHU Validation set (n=431) from Mathios et al., 2021 (D. Mathios et al., Nat Commun 12, 5060 (2021)), a subset of patients with prior cancers, with and without cancer recurrence in the LUCAS cohort (n=25), the validation set from the AHN / DECAMP cohort in Bruhm et al., 2023 (56), and the lung cancer monitoring cohort from Phallen et al., 2019 (n=19)50.

[0218] Finally, we trained multi-class GBMs using the features described above to generate an ARTEMIS and ARTEMIS-DELFI score for tissue of origin classification. The final ARTEMIS model for tissue of origin used the ensemble components described above as features, and the ARTEMIS-DELFI model used these components and additional fragmentation features (the 39 chromosomal arm z-scores for aneuploidy, and the ratio of short to long fragments in 5 Mb bins genome wide). Each ensemble component produced a vector of multinomial probabilities, with one for each possible tumor site. We determined classification based on the element of the vector with the maximum value. When ensembling multiple GBM classifiers, all elements of the vector were66 161564170.1used as feature inputs to the ensemble model. The models were trained using the nested cross- validation procedure described above on all cancer samples from Cristiano et al., 2019 (n=423) (28) and performance was reported for all cancers detected at the 90% specificity threshold by the ARTEMIS and ARTEMIS-DELFI detection models when trained on the full cohort including patients without cancer. Consistent with that previous publication, for the tissue of origin analyses, we included the baseline timepoints from the lung cancer monitoring cohort above to increase the number of lung cancers available for classification analyses.

[0219] Gene-Set Enrichment Analysis

[0220] We downloaded the coordinates of all genes in the chm13 version of the curated ncbiRefSeq from the UCSC genome browser. For genes with multiple transcripts we defined the region encompassing all transcripts, and then counted which of the 1.1 billion possible kmers occurred in each gene. We then ranked the genes by two metrics: (1) the total number of kmers, and (2) a corrected kmer density metric – the number of types occurring in the gene divided by the kmer density (total kmers / Mb). We employed this correction because our requirement that a kmer occur only in a single repeat element type means that genes with a large number of repeat types occurring within them have less opportunities for kmers to be found despite having a higher overall repeat density. We used these rank lists in a Gene Set Enrichment Analysis using both the KEGG gene sets and the COSMIC cancer gene census13–15. For KEGG, the default minimum gene set size of 25 and maximum gene set size of 500 was used; for the COSMIC analysis, we used minimum of 15 and maximum of 1000 to ensure that all gene sets were included.

[0221] Simulations of kmer repeat landscapes in short-read sequencing

[0222] We simulated 50 million paired end reads with a 0.1% error rate (corresponds to Q30) from the chm13 reference genome, partitioned the reads by their repeat element type of origin (or reads occurring in non-repeat regions) and counted the 1.1 billion possible kmers for each set. We then calculated the percentage of all counted kmer occurrences for each repeat element type that occurred in reads originating from that repeat element type.

[0223] Analysis of kmer repeat landscapes in PCAWG tumors

[0224] We generated kmer repeat landscapes for all 686 PCAWG samples (343 matched Tumor / Normal pairs) and excluded 10 pairs on the PCAWG blacklist, leaving 666 samples. To define a significant change in a repeat element, we selected 100 Normal samples and subsampled two in silico Normal samples, each having approximately half the coverage of the original sample.67 161564170.1We then calculated the ratio of the kmer repeat landscapes between the two samples and set the thresholds to call a significant change in a tumor as a Tumor / Normal ratio below the 1stpercentile or above the 99thpercentile of the Normal / Normal ratios. This ensures that changes seen in tumor are larger than that expected merely due to chance variation. We correlated the number of such changes with several metrics of genomic instability (entropy, non-modal ploidy fraction, non-diploid fraction, loss of heterozygosity fraction, breakpoint count, tumor mutation burden, ploidy, and modal ploidy)16. The visualized sample set (n=293, FIG. 2a) was defined by tumors where the genomic stability metrics could be retrieved from Anagnostou et al.16. Briefly, this required that a tumor had TCGA copy number data (for ploidy-related metrics) and was within the mc3 (Multi-Center Mutation Calling in Multiple Cancers) mutation call set used for TMB calculation. Correlation analyses were performed on this subset from Anagnostou et al., 202016(293 of the original 333), while the remaining analyses used all 333 non-blacklisted samples.

[0225] To determine whether these observed changes were greater than that expected due to chromosomal arm gains and losses alone we used TCGA array-based copy number data to calculate an average arm-level ploidy for 329 of 333 tumors that had the data available. We counted all kmers on each chromosomal arm and determined theoretical diploid kmer counts. We then adjusted these counts at the chromosomal arm level for each individual samples’ copy number profile. This adjustment was not possible for the acrocentric chromosomal arms or chromosome Y which were not provided in the TCGA data, so we assumed these to be diploid. We then calculated predicted Tumor:Normal ratios if all changes to the kmer repeat landscape were due to chromosomal arm gains and losses alone and defined a confidence interval around these for each element in each sample using the range of observed values in the random downsampling experiments above. We then further filtered the changes to those that were also outside the range predicted to be due to chromosomal arm gains and losses.

[0226] To extend these copy number analysis, we identified repeat kmers found on each chromosomal arm, and calculated the ratio of arm-level Tumor:Normal kmer counts for each sample and each of the chromosomal arms for which copy number data was available in TCGA. We compared the expected ratio of kmer counts to the observed ratio across the range of arm-level ploidies. We repeated this analysis for amplicon segments with copy number greater than 4, for up to the 10 most amplified segments in each tumor. We also randomly sampled 10 diploid segments to serve as controls from each tumor with one of the amplified segments above. After filtering for68 161564170.1amplicons between 10 kb and 5 Mb in size, we had 1976 regions, 1972 of which contained repeat kmers. We compared the expected ratio of kmer counts to the observed ratio across the range of amplicon copy numbers.

[0227] Next, we identified two common copy number gains in tumors – a 1 Mb region containing ERBB2 in breast tumors and a 31 Mb region containing SOX2 and PIK3CA in lung squamous cell tumors. We extracted repeat element types with kmers overlapping these regions and evaluated differences in genome-wide kmer counts for these repeat elements between tumors with and without focal amplification.

[0228] To assess the impact of small kb-scale repeat-related structural changes on kmer repeat landscapes, we identified 19 LINE-1 mediated deletions in PCAWG that occurred in lung squamous cell tumors17. Of the 19 LINE-1 mediated deletions ranging in size from <1kb to >50 Mb, 11 contained at least one ARTEMIS kmer. We further restricted the set of these kmers to those that occurred only within the genomic region delimited by the deletion and compared the count of these kmers between Tumor and Normal samples with and without the deletion.

[0229] We stratified patients into two groups with numbers of repeat element changes above or below the median for the 6 main families of repeat elements (LINEs, SINEs, LTRs, Satelites, DNA Elements, and RNA Elements) and compared overall survival and progression free survival between the two groups using the R packages survival and survminer. Only the number of SINE changes had a significant association with survival. To correct for the effect of tumor type in this analysis, we also performed the analysis stratifying patients based on the tumor-type specific median number of changes, in effect ensuring that half of the patients from every tumor type were in each group. We also performed survival analyses for the genomic stability metrics defined in Anagnostou et al., 202016.

[0230] Finally, we centered and scaled the coverage normalized counts and used these kmer repeat landscapes to train a penalized logistic regression model to generate a cross-validated ARTEMIS score (mean scores across 10 repeats of 5-fold cross validation) for distinguishing tumor from normal tissue samples. We further used kmer repeat landscapes to train a multi-class gradient boosted model (GBM) to generate cross-validated ARTEMIS scores (5-fold cross-validation) to distinguish between tumor tissues of origin. We conservatively excluded breast tum33 ors from this analysis as they were sequenced at a different site than the other PCAWG samples and could inflate performance due to technical differences. To assess the extent to which repeat landscapes69 161564170.1were affected by sequencing coverage, we performed subsampling of sequencing data from PCAWG lung cancers with a matched blood derived normal sample (n=42) to 30x and 1-2x coverage and assessed concordance of the subsampled normalized kmer counts to the original normalized kmer counts.

[0231] Observation of novel changes in repeat elements

[0232] We quantified changes to each repeat element using the analyses described above for individual tumor normal pairs. Additionally we compared kmer counts for each feature between all tumors and normal using the Wilcoxon rank sum test with Benjamini-Hochberg correction for multiple tests, and Cohen’s D for effect size. To assess the novelty of these changes to 57 subfamilies of repeats comprising 1280 repeat element types we first performed a systematic search for the description of each of 57 subfamilies as follows:

[0233] We first searched for prior evidence in PCAWG / TCGA using manual curation of pan-cancer papers on repeat elements that used paired Tumor / Normal studies of PCAWG and TCGA17, 18. We categorized subfamilies with elements described in these studies as having “Prior evidence in TCGA / PCAWG”. For the elements within these subfamilies, if they were specifically identified as involved in a cancer-related change, we considered them to have prior evidence for the change, and if they were not we categorized the changes we saw as “Novel”. For the LINE-L1 subfamily, we conservatively considered all 132 elements to have prior evidence as there were nearly 20,000 LINE-L1 related changes identified in PCAWG samples in Rodriguez-Martin et. al, 202017but the element naming convention was drastically different pre-chm13 which made disaggregation difficult. This approach yielded description of 18 subfamilies comprising 1008 elements. Of these, 275 elements were described as changed in cancer and 733 were novel changes we identified.

[0234] For subfamilies not represented in the above studies, we conducted a Pubmed search for “Subfamily name AND cancer", or "Subfamily name" or "Individual element name" for subfamilies with very small numbers of elements and search results. For subfamilies with evidence found through this method, we categorized elements within them to have “Prior Evidence for Subfamily”. If evidence for a subfamily and its elements was found via Pubmed search for another subfamily, we placed these within the same category. This approach yielded descriptions of 12 subfamilies comprising 185 elements19–31.70 161564170.1

[0235] If neither of the above two search attempts yielded evidence for a subfamily or the elements within it, we categorized them as “Novel”. This included the final 27 subfamilies comprising 87 elements.

[0236] Analysis of kmer repeat landscapes in cfDNA

[0237] We generated kmer repeat landscapes for the LUCAS Cohort and its corresponding external validation set3, and for the AHN / DECAMP9, lung cancer monitoring4and liver cancer6cohorts. In the LUCAS cohort, we compared kmer counts in lung cancer and non-cancer samples for repeat element types frequently altered in PCAWG lung tumors and for satellite repeat element types with varying percentages of kmers occurring on chrY.

[0238] We looked at the impact of epigenetic features on the representation of cfDNA fragments in the circulation. In the non-cancer patients in the LUCAS cohort, we looked at differences in fragment length and coverage for fragments occurring in regions of different histone marks. We defined these regions from ENCODE Histone CHIP-Seq experiments32in lymphoblastoid cell line GM12878 (ENCFF001SUG, ENCFF001SUI, ENCFF001SUJ, ENCFF001SUE, ENCFF001SUL, ENCFF001SUF, ENCFF001SUN, ENCFF001SUO, ENCFF001SUP, ENCFF001SUQ). We further used ENCODE chromatin state definitions33, grouping states 1-5 (Promoter, Enhancer 1, Enhancer 2, Transcription 5` 1, Transcription 5` 2) as Activating, States 7-9 (Transcription 3` 1, Transcription 3` 2, Transcription 3` 3) as 3` Transcription, and States 10-13 (PC Repressed 1, PC Repressed 2, Heterochromatin 1, Heterochromatin 2) as Repressed. We then looked at the localization of these histone marks within repeat element types in chm13 and identified repeat element types with higher and lower densities of each histone mark. Finally, we analyzed how the ratio of observed kmer counts to expected kmer counts (based on the chm13 reference genome) varied between repeat element types based on histone mark density.

[0239] Early detection, tissue of origin and monitoring of cancer with ARTEMIS

[0240] We trained ensembled penalized logistic regression models using the kmer repeat landscapes and defined the predictions as the ARTEMIS score. First, we obtained a kmer repeat landscape for each sample using the 786 features with more than 1000 kmers per million aligned reads expected. This filtering was employed because at low coverage features with low abundance have greater technical variation (FIG. 22). We partitioned the remaining features into 6 families (LINEs, SINEs, Satellites, LTRs, RNA Elements, DNA Elements) and centered and scaled features within their family in each sample. RNA Elements and DNA Elements were combined because after71 161564170.1filtering for the 1000 kmers / million aligned reads threshold, only two RNA Element features remained. We further defined 5611 Mb bins using the epigenetic analyses above that had either >90% of their bases covered by peaks from one of the Histone CHIP-Seq experiments above, or >30% of their bases covered by one of the three groups of chromatin states defined above. We calculated aligned fragment coverage in these bins. Bins with mappability < 0.9 or GC content < 0.3 were excluded from downstream analysis, as in our prior fragmentation-based analyses3, 6. We then trained six penalized logistic regression (PLR) models using repeat landscapes for the five repeat families above and the epigenetic profiles described above and ensembled their scores together with a PLR model using leave-one-out cross validation with nested 5-fold cross validation to train each learner. The combined score was defined as the ARTEMIS Score.

[0241] To incorporate DELFI fragmentation profiles we ensembled the ARTEMIS score with three additional models: a PLR model using principal component analysis of the ratio of short to long fragments in 5 Mb bins genome wide3, a PLR model on 39 arm-level z-scores for aneuploidy3and a gradient boosted model on coverage in 5 Mb bins genome wide5. This combined ensemble produced a joint ARTEMIS-DELFI score. We compared the ARTEMIS score and the joint ARTEMIS-DELFI score to the scores from the DELFI models described in Mathios et al., 20213. We further retrained the lung cancer model on the full LUCAS cohort and then applied the locked models to three external validation cohorts: The JHU Validation set from Mathios et al., 20213, the AHN / DECAMP Validation set from Bruhm et al., 20239and the lung cancer monitoring cohort from Phallen et al., 20194.

[0242] Finally, we trained multi-class GBMs using the ensembled architectures described above to generate an ARTEMIS and ARTEMIS-DELFI score for tissue of origin classification. This model was trained in a cross-validated fashion on all cancer samples from Cristiano et al., 20195and performance was reported for all cancers detected at the 90% specificity threshold by ARTEMIS and ARTEMIS-DELFI when trained on the full cohort including non-cancer patients. Consistent with that previous publication, for the tissue of origin analyses, we included the baseline timepoints from the lung cancer monitoring cohort above to increase the number of lung cancers available for classification analyses.

[0243] Bioinformatic and statistical software

[0244] All statistical analysis were performed using R version >4.0.5. GSEA analysis were performed using the Broad Institute software GSEA v4.3.213, 15. We used Jellyfish10for de novo72 161564170.1kmer finding in the chm13 reference genome, and then to count these kmers in patient samples. For counting of aligned reads and for fragmentation analyses, fastp (0.20.0) was used for trimming of adapter sequences, Bowtie2 (2.3.0) was used to align paired-end reads to the hg19 reference genome and samtools (1.17) was used for counting reads. The R package Caret was used to implement classification. The R packages regioneR34, fgsea (github.com / ctlab / fgsea / ), ComplexHeatmap35, and ggplot236were among those used for visualizations.

[0245] Statistics and reproducibility

[0246] Computer code, software versions and the computing environment for reproducing results in this study will be made publicly available in a GitHub repository upon publication. P- values for two group comparisons were performed using the Wilcoxon rank sum test, as indicated in the text. Correlation of continuous variables was performed using either the Pearson product- moment correlation coefficient or Spearman’s rank correlation coefficient as indicated in the text. ROC curves were compared using DeLong’s test. All confidence intervals for area under the ROC curve indicate a confidence level of 95% and were based on DeLong’s method.

[0247] References (for the above Example 2 as shown in numerical superscripts in Example 2 to correspond to the reference numbers shown in the immediately following listing) 1. N. Altemose, K. H. Miga, M. Maggioni, H. F. Willard, Genomic Characterization of Large Heterochromatic Gaps in the Human Genome Assembly. Plos Comput Biol.10, e1003628 (2014). 2. The ICGC / TCGA Pan-Cancer Analysis of Whole Genomes Consortium, Pan-cancer analysis of whole genomes. Nature.578, 82–93 (2020). 3. D. Mathios, J. S. Johansen, S. Cristiano, J. E. Medina, J. Phallen, K. R. Larsen, D. C. Bruhm, N. Niknafs, L. Ferreira, V. Adleff, J. Y. Chiao, A. Leal, M. Noe, J. R. White, A. S. Arun, C. Hruban, A. V. Annapragada, S. Ø. Jensen, M.-B. W. Ørntoft, A. H. Madsen, B. Carvalho, M. de Wit, J. Carey, N. C. Dracopoli, T. Maddala, K. C. Fang, A.-R. Hartman, P. M. Forde, V. Anagnostou, J. R. Brahmer, R. J. A. Fijneman, H. J. Nielsen, G. A. Meijer, C. L. Andersen, A. Mellemgaard, S. E. Bojesen, R. B. Scharpf, V. E. Velculescu, Detection and characterization of lung cancer using cell-free DNA fragmentomes. Nat Commun.12, 5060 (2021). 4. J. Phallen, A. Leal, B. D. Woodward, P. M. Forde, J. Naidoo, K. A. Marrone, J. R. Brahmer, J. Fiksel, J. E. Medina, S. Cristiano, D. N. Palsgrove, C. D. Gocke, D. C. Bruhm, P. Keshavarzian, V. Adleff, E. Weihe, V. Anagnostou, R. B. Scharpf, V. E. Velculescu, H. Husain, Early Noninvasive Detection of Response to Targeted Therapy in Non–Small Cell Lung Cancer. Cancer Res.79, 1204– 1213 (2019).73 161564170.15. S. Cristiano, A. Leal, J. Phallen, J. Fiksel, V. Adleff, D. C. Bruhm, S. Ø. Jensen, J. E. Medina, C. Hruban, J. R. White, D. N. Palsgrove, N. Niknafs, V. Anagnostou, P. Forde, J. Naidoo, K. Marrone, J. Brahmer, B. D. Woodward, H. Husain, K. L. van Rooijen, M.-B. W. Ørntoft, A. H. Madsen, C. J. H. van de Velde, M. Verheij, A. Cats, C. J. A. Punt, G. R. Vink, N. C. T. van Grieken, M. Koopman, R. J. A. Fijneman, J. S. Johansen, H. J. Nielsen, G. A. Meijer, C. L. Andersen, R. B. Scharpf, V. E. Velculescu, Genome-wide cell-free DNA fragmentation in patients with cancer. Nature.570, 385– 389 (2019). 6. Z. H. Foda, A. V. Annapragada, K. Boyapati, D. C. Bruhm, N. A. Vulpescu, J. E. Medina, D. Mathios, S. Cristiano, N. Niknafs, H. T. Luu, M. G. Goggins, R. A. Anders, J. Sun, S. H. Mehta, D. L. Thomas, G. D. Kirk, V. Adleff, J. Phallen, R. B. Scharpf, A. K. Kim, V. E. Velculescu, Detecting liver cancer using cell-free DNA fragmentomes. Cancer Discov (2022), doi:10.1158 / 2159-8290.cd-22- 0659. 7. E. M. Stoop, M. C. de Haan, T. R. de Wijkerslooth, P. M. Bossuyt, M. van Ballegooijen, C. Y. Nio, M. J. van de Vijver, K. Biermann, M. Thomeer, M. E. van Leerdam, P. Fockens, J. Stoker, E. J. Kuipers, E. Dekker, Participation and yield of colonoscopy versus non-cathartic CT colonography in population-based screening for colorectal cancer: a randomised controlled trial. Lancet Oncol. 13, 55–64 (2012). 8. E. Billatos, F. Duan, E. Moses, H. Marques, I. Mahon, L. Dymond, C. Apgar, D. Aberle, G. Washko, A. Spira, D. investigators, Detection of early lung cancer among military personnel (DECAMP) consortium: study protocols. Bmc Pulm Med.19, 59 (2019). 9. D. Bruhm, D. Mathios, Z. H. FOda, A. V. Annapragada, J. E. Medina, Vi. Adleff, E. J. Chiao, L. Ferreira, S. Cristiano, J. R. White, S. A. Mazzilli, E. Billatos, A. Spira, A. H. Zaidi, J. Mueller, A. K. Kim, V. Anagnostou, J. Phallen, R. B. Scharpf, V. E. Velculescu, Single molecule genome-wide mutation profiles of cell-free DNA for non-invasive detection of cancer. In Press. Nature Genetics. 10. G. Marçais, C. Kingsford, A fast, lock-free approach for efficient parallel counting of occurrences of k-mers. Bioinformatics.27, 764–770 (2011). 11. B. A. Methé, K. E. Nelson, M. Pop, …, R. K. Wilson, O. White, A framework for human microbiome research. Nature.486, 215–21 (2011). 12. C. Huttenhower, D. Gevers, R. Knight, …, R. K. Wilson, O. White, Structure, function and diversity of the healthy human microbiome. Nature.486, 207–214 (2012). 13. A. Subramanian, P. Tamayo, V. K. Mootha, S. Mukherjee, B. L. Ebert, M. A. Gillette, A. Paulovich, S. L. Pomeroy, T. R. Golub, E. S. Lander, J. P. Mesirov, Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proc National Acad Sci.102, 15545–15550 (2005).74 161564170.114. J. G. Tate, S. Bamford, H. C. Jubb, Z. Sondka, D. M. Beare, N. Bindal, H. Boutselakis, C. G. Cole, C. Creatore, E. Dawson, P. Fish, B. Harsha, C. Hathaway, S. C. Jupe, C. Y. Kok, K. Noble, L. Ponting, C. C. Ramshaw, C. E. Rye, H. E. Speedy, R. Stefancsik, S. L. Thompson, S. Wang, S. Ward, P. J. Campbell, S. A. Forbes, COSMIC: the Catalogue Of Somatic Mutations In Cancer. Nucleic Acids Res. 47, gky1015- (2018). 15. V. K. Mootha, C. M. Lindgren, K.-F. Eriksson, A. Subramanian, S. Sihag, J. Lehar, P. Puigserver, E. Carlsson, M. Ridderstråle, E. Laurila, N. Houstis, M. J. Daly, N. Patterson, J. P. Mesirov, T. R. Golub, P. Tamayo, B. Spiegelman, E. S. Lander, J. N. Hirschhorn, D. Altshuler, L. C. Groop, PGC-1α- responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet.34, 267–273 (2003). 16. V. Anagnostou, N. Niknafs, K. Marrone, D. C. Bruhm, J. R. White, J. Naidoo, K. Hummelink, K. Monkhorst, F. Lalezari, M. Lanis, S. Rosner, J. E. Reuss, K. N. Smith, V. Adleff, K. Rodgers, Z. Belcaid, L. Rhymee, B. Levy, J. Feliciano, C. L. Hann, D. S. Ettinger, C. Georgiades, F. Verde, P. Illei, Q. K. Li, A. S. Baras, E. Gabrielson, M. V. Brock, R. Karchin, D. M. Pardoll, S. B. Baylin, J. R. Brahmer, R. B. Scharpf, P. M. Forde, V. E. Velculescu, Multimodal genomic features predict outcome of immune checkpoint blockade in non-small-cell lung cancer. Nat Cancer.1, 99–111 (2020). 17. B. Rodriguez-Martin, E. G. Alvarez, A. Baez-Ortega, …, L. van’t Veer, C. von Mering, Pan-cancer analysis of whole genomes identifies driver rearrangements promoted by LINE-1 retrotransposition. Nat Genet.52, 306–319 (2020). 18. H. S. Jang, N. M. Shah, A. Y. Du, Z. Z. Dailey, E. C. Pehrsson, P. M. Godoy, D. Zhang, D. Li, X. Xing, S. Kim, D. O’Donnell, J. I. Gordon, T. Wang, Transposable elements drive widespread expression of oncogenes in human cancers. Nat. Genet.51, 611–617 (2019). 19. C. Yandım, G. Karakülah, Dysregulated expression of repetitive DNA in ER+ / HER2- breast cancer. Cancer Genet.239, 36–45 (2019). 20. K. Zillner, J. Komatsu, K. Filarsky, R. Kalepu, A. Bensimon, A. Nmeth, Active human nucleolar organizer regions are interspersed with inactive rDNA repeats in normal and tumor cells. Epigenomics.7, 363–378 (2015). 21. M. Uemura, Q. Zheng, C. M. Koh, W. G. Nelson, S. Yegnasubramanian, A. M. D. Marzo, Overexpression of ribosomal RNA in prostate cancer is common but not linked to rDNA promoter hypomethylation. Oncogene.31, 1254–1263 (2012). 22. V. Valori, K. Tus, C. Laukaitis, D. T. Harris, L. LeBeau, K. A. Maggert, Human rDNA copy number is unstable in metastatic breast cancers. Epigenetics.15, 85–106 (2020).75 161564170.123. S. Shuai, H. Suzuki, A. Diaz-Navarro, F. Nadeu, S. A. Kumar, A. Gutierrez-Fernandez, J. Delgado, M. Pinyol, C. López-Otín, X. S. Puente, M. D. Taylor, E. Campo, L. D. Stein, The U1 spliceosomal RNA is recurrently mutated in multiple cancers. Nature.574, 712–716 (2019). 24. H. Suzuki, S. A. Kumar, S. Shuai, A. Diaz-Navarro, A. Gutierrez-Fernandez, P. D. Antonellis, F. M. G. Cavalli, K. Juraschka, H. Farooq, I. Shibahara, M. C. Vladoiu, J. Zhang, N. Abeysundara, D. Przelicki, P. Skowron, N. Gauer, B. Luu, C. Daniels, X. Wu, A. Forget, A. Momin, J. Wang, W. Dong, S.-K. Kim, W. A. Grajkowska, A. Jouvet, M. Fèvre-Montange, M. L. Garrè, A. A. N. Rao, C. Giannini, J. M. Kros, P. J. French, N. Jabado, H.-K. Ng, W. S. Poon, C. G. Eberhart, I. F. Pollack, J. M. Olson, W. A. Weiss, T. Kumabe, E. López-Aguilar, B. Lach, M. Massimino, E. G. V. Meir, J. B. Rubin, R. Vibhakar, L. B. Chambless, N. Kijima, A. Klekner, L. Bognár, J. A. Chan, C. C. Faria, J. Ragoussis, S. M. Pfister, A. Goldenberg, R. J. Wechsler-Reya, S. D. Bailey, L. Garzia, A. S. Morrissy, M. A. Marra, X. Huang, D. Malkin, O. Ayrault, V. Ramaswamy, X. S. Puente, J. A. Calarco, L. Stein, M. D. Taylor, Recurrent noncoding U1 snRNA mutations drive cryptic splicing in SHH medulloblastoma. Nature. 574, 707–711 (2019). 25. C. Tan, J. Cao, L. Chen, X. Xi, S. Wang, Y. Zhu, L. Yang, L. Ma, D. Wang, J. Yin, T. Zhang, Z. J. Lu, Noncoding RNAs Serve as Diagnosis and Prognosis Biomarkers for Hepatocellular Carcinoma. Clin. Chem.65, 905–915 (2019). 26. S. Ganesan, Breaking satellite silence: human satellite II RNA expression in ovarian cancer. J. Clin. Investig.132, e161981 (2022). 27. D. T. Ting, D. Lipson, S. Paul, B. W. Brannigan, S. Akhavanfard, E. J. Coffman, G. Contino, V. Deshpande, A. J. Iafrate, S. Letovsky, M. N. Rivera, N. Bardeesy, S. Maheswaran, D. A. Haber, Aberrant Overexpression of Satellite Repeats in Pancreatic and Other Epithelial Cancers. Science. 331, 593–596 (2011). 28. W. Arancio, C. Coronnello, Repetitive Sequence Transcription in Breast Cancer. Cells.11, 2522 (2022). 29. A. K. Saha, M. Mourad, M. H. Kaplan, I. Chefetz, S. N. Malek, R. Buckanovich, D. M. Markovitz, R. Contreras-Galindo, The Genomic Landscape of Centromeres in Cancers. Sci Rep-uk.9, 11259 (2019). 30. K. Tsumagari, L. Qi, K. Jackson, C. Shao, M. Lacey, J. Sowden, R. Tawil, V. Vedanarayanan, M. Ehrlich, Epigenetics of a tandem DNA repeat: chromatin DNaseI sensitivity and opposite methylation changes in cancers. Nucleic Acids Res.36, 2196–2207 (2008). 31. C. S. Herrington, M. Worsham, S. A. Southern, P. Mackowiak, S. R. Wolman, Loss of sequences on the short arm of chromosome 17 is a late event in squamous carcinoma of the cervix. Mol. Pathol.54, 160 (2001).76 161564170.132. The ENCODE Project Consortium, An Integrated Encyclopedia of DNA Elements in the Human Genome. Nature.489, 57–74 (2012). 33. J. W. K. Ho, Y. L. Jung, T. Liu, B. H. Alver, S. Lee, K. Ikegami, K.-A. Sohn, A. Minoda, M. Y. Tolstorukov, A. Appert, S. C. J. Parker, T. Gu, A. Kundaje, N. C. Riddle, E. Bishop, T. A. Egelhofer, S. S. Hu, A. A. Alekseyenko, A. Rechtsteiner, D. Asker, J. A. Belsky, S. K. Bowman, Q. B. Chen, R. A.-J. Chen, D. S. Day, Y. Dong, A. C. Dose, X. Duan, C. B. Epstein, S. Ercan, E. A. Feingold, F. Ferrari, J. M. Garrigues, N. Gehlenborg, P. J. Good, P. Haseley, D. He, M. Herrmann, M. M. Hoffman, T. E. Jeffers, P. V. Kharchenko, P. Kolasinska-Zwierz, C. V. Kotwaliwale, N. Kumar, S. A. Langley, E. N. Larschan, I. Latorre, M. W. Libbrecht, X. Lin, R. Park, M. J. Pazin, H. N. Pham, A. Plachetka, B. Qin, Y. B. Schwartz, N. Shoresh, P. Stempor, A. Vielle, C. Wang, C. M. Whittle, H. Xue, R. E. Kingston, J. H. Kim, B. E. Bernstein, A. F. Dernburg, V. Pirrotta, M. I. Kuroda, W. S. Noble, T. D. Tullius, M. Kellis, D. M. MacAlpine, S. Strome, S. C. R. Elgin, X. S. Liu, J. D. Lieb, J. Ahringer, G. H. Karpen, P. J. Park, Comparative analysis of metazoan chromatin organization. Nature.512, 449–452 (2014). 34. B. Gel, A. Díez-Villanueva, E. Serra, M. Buschbeck, M. A. Peinado, R. Malinverni, regioneR: an R / Bioconductor package for the association analysis of genomic regions based on permutation tests. Bioinformatics.32, 289–291 (2016). 35. Z. Gu, R. Eils, M. Schlesner, Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics.32, 2847–2849 (2016). 36. H. Wickham, ggplot2, Elegant Graphics for Data Analysis (2009), doi:10.1007 / 978-0-387- 98141-3 OTHER EMBODIMENTS

[0248] From the foregoing description, it will be apparent that variations and modifications may be made to the disclosure described herein to adopt it to various usages and conditions. Such embodiments are also within the scope of the following claims.

[0249] All citations to sequences, patents and publications in this specification are herein incorporated by reference to the same extent as if each independent patent and publication was specifically and individually indicated to be incorporated by reference. By their citation of various references in this document, Applicants do not admit any particular reference is “prior art” to their disclosure.77 161564170.1outcome of immune checkpoint blockade in non-small-cell lung cancer. Nat Cancer. 1, 99-111 (2020).17. B. Rodriguez-Martin, E. G. Alvarez, A. Baez-Ortega, ..., L. van't Veer, C. von Mering, Pan-cancer analysis of whole genomes identifies driver rearrangements promoted by LINE-1 retrotransposition. Nat Genet. 52, 306-319 (2020).18. H. S. Jang, N. M. Shah, A. Y. Du, Z. Z. Dailey, E. C. Pehrsson, P. M. Godoy, D. Zhang, D. Li, X. Xing, S. Kim, D. O'Donnell, J. I. Gordon, T. Wang, Transposable elements drive widespread expression of oncogenes in human cancers. Nat. Genet. 51, 611-617 (2019).19. C. Yandim, G. Karakulah, Dysregulated expression of repetitive DNA in ER+ / HER2- breast cancer. Cancer Genet. 239, 36-45 (2019).20. K. Zillner, J. Komatsu, K. Filarsky, R. Kalepu, A. Bensimon, A. Nmeth, Active human nucleolar organizer regions are interspersed with inactive rDNA repeats in normal and tumor cells. Epigenomics. 7, 363-378 (2015).21. M. Uemura, Q. Zheng, C. M. Koh, W. G. Nelson, S. Yegnasubramanian, A. M. D. Marzo, Overexpression of ribosomal RNA in prostate cancer is common but not linked to rDNA promoter hypomethylation. Oncogene. 31, 1254-1263 (2012).22. V. Valori, K. Tus, C. Laukaitis, D. T. Harris, L. LeBeau, K. A. Maggert, Human rDNA copy number is unstable in metastatic breast cancers. Epigenetics. 15, 85-106 (2020).23. S. Shuai, H. Suzuki, A. Diaz-Navarro, F. Nadeu, S. A. Kumar, A. Gutierrez-Fernandez, J. Delgado, M. Pinyol, C. Lopez-Otin, X. S. Puente, M. D. Taylor, E. Campo, L. D. Stein, The U1 spliceosomal RNA is recurrently mutated in multiple cancers. Nature. 574, 712-716 (2019).24. H. Suzuki, S. A. Kumar, S. Shuai, A. Diaz-Navarro, A. Gutierrez-Fernandez, P. D. Antonellis, F. M. G. Cava I li, K. Juraschka, H. Farooq, I. Shibahara, M. C. Vladoiu, J. Zhang, N. Abeysundara, D. Przelicki, P. Skowron, N. Gauer, B. Luu, C. Daniels, X. Wu, A. Forget, A. Momin, J. Wang, W. Dong, S.-K. Kim, W. A. Grajkowska, A. Jouvet, M. Fevre-Montange, M. L. Garre, A. A. N. Rao, C. Giannini, J. M. Kros, P. J. French, N. Jabado, H.-K. Ng, W. S. Poon, C. G. Eberhart, I. F. Pollack, J. M. Olson,W. A. Weiss, T. Kumabe, E. Lopez-Aguilar, B. Lach, M. Massimino, E. G. V. Meir, J. B. Rubin, R. Vibhakar, L. B. Chambless, N. Kijima, A. Klekner, L. Bognar, J. A. Chan, C. C. Faria, J. Ragoussis, S. M. Pfister, A. Goldenberg, R. J. Wechsler-Reya, S. D. Bailey, L. Garzia, A. S. Morrissy, M. A. Marra,X. Huang, D. Malkin, O. Ayrault, V. Ramaswamy, X. S. Puente, J. A. Calarco, L. Stein, M. D. Taylor, Recurrent noncoding U1 snRNA mutations drive cryptic splicing in SHH medulloblastoma. Nature. 574, 707-711 (2019).78SUBSTITUTE SHEET (RULE 26)25. C. Tan, J. Cao, L. Chen, X. Xi, S. Wang, Y. Zhu, L. Yang, L. Ma, D. Wang, J. Yin, T. Zhang, Z. J. Lu, Noncoding RNAs Serve as Diagnosis and Prognosis Biomarkers for Hepatocellular Carcinoma. Clin. Chem. 65, 905-915 (2019).26. S. Ganesan, Breaking satellite silence: human satellite II RNA expression in ovarian cancer. J. Clin. Investig. 132, el61981 (2022).27. D. T. Ting, D. Lipson, S. Paul, B. W. Brannigan, S. Akhavanfard, E. J. Coffman, G. Contino, V. Deshpande, A. J. lafrate, S. Letovsky, M. N. Rivera, N. Bardeesy, S. Maheswaran, D. A. Haber, Aberrant Overexpression of Satellite Repeats in Pancreatic and Other Epithelial Cancers. Science. 331, 593-596 (2011).28. W. Arancio, C. Coronnello, Repetitive Sequence Transcription in Breast Cancer. Cells. 11, 2522 (2022).29. A. K. Saha, M. Mourad, M. H. Kaplan, I. Chefetz, S. N. Malek, R. Buckanovich, D. M. Markovitz,R. Contreras-Galindo, The Genomic Landscape of Centromeres in Cancers. Sci Rep-uk. 9, 11259 (2019).30. K. Tsumagari, L. Qi, K. Jackson, C. Shao, M. Lacey, J. Sowden, R. Tawil, V. Vedanarayanan, M. Ehrlich, Epigenetics of a tandem DNA repeat: chromatin DNasel sensitivity and opposite methylation changes in cancers. Nucleic Acids Res. 36, 2196-2207 (2008).31. C. S. Herrington, M. Worsham, S. A. Southern, P. Mackowiak, S. R. Wolman, Loss of sequences on the short arm of chromosome 17 is a late event in squamous carcinoma of the cervix. Mol. Pathol. 54, 160 (2001).32. The ENCODE Project Consortium, An Integrated Encyclopedia of DNA Elements in the Human Genome. Nature. 489, 57-74 (2012).33. J. W. K. Ho, Y. L. Jung, T. Liu, B. H. Alver, S. Lee, K. Ikegami, K.-A. Sohn, A. Minoda, M. Y. Tolstorukov, A. Appert, S. C. J. Parker, T. Gu, A. Kundaje, N. C. Riddle, E. Bishop, T. A. Egelhofer, S.S. Hu, A. A. Alekseyenko, A. Rechtsteiner, D. Asker, J. A. Belsky, S. K. Bowman, Q. B. Chen, R. A. -J. Chen, D. S. Day, Y. Dong, A. C. Dose, X. Duan, C. B. Epstein, S. Ercan, E. A. Feingold, F. Ferrari, J. M. Garrigues, N. Gehlenborg, P. J. Good, P. Haseley, D. He, M. Herrmann, M. M. Hoffman, T. E. Jeffers, P. V. Kharchenko, P. Kolasinska-Zwierz, C. V. Kotwaliwale, N. Kumar, S. A. Langley, E. N. Larschan, I. Latorre, M. W. Libbrecht, X. Lin, R. Park, M. J. Pazin, H. N. Pham, A. Plachetka, B. Qin, Y. B. Schwartz, N. Shoresh, P. Stempor, A. Vielle, C. Wang, C. M. Whittle, H. Xue, R. E. Kingston, J. H. Kim, B. E. Bernstein, A. F. Dernburg, V. Pirrotta, M. I. Kuroda, W. S. Noble, T. D. Tullius, M. Kellis, D. M. MacAlpine, S. Strome, S. C. R. Elgin, X. S. Liu, J. D. Lieb, J. Ahringer, G. H. Karpen, P. J. Park, Comparative analysis of metazoan chromatin organization. Nature. 512, 449-452 (2014).79SUBSTITUTE SHEET (RULE 26)34. B. Gel, A. Diez-Villanueva, E. Serra, M. Buschbeck, M. A. Peinado, R. Malinverni, regioneR: an R / Bioconductor package for the association analysis of genomic regions based on permutation tests. Bioinformatics. 32, 289-291 (2016).35. Z. Gu, R. Eils, M. Schlesner, Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics. 32, 2847-2849 (2016).36. H. Wickham, ggplot2, Elegant Graphics for Data Analysis (2009), doi:10.1007 / 978-0-387- 98141-3OTHER EMBODIMENTS

[0248] From the foregoing description, it will be apparent that variations and modifications may be made to the disclosure described herein to adopt it to various usages and conditions. Such embodiments are also within the scope of the following claims.

[0249] All citations to sequences, patents and publications in this specification are herein incorporated by reference to the same extent as if each independent patent and publication was specifically and individually indicated to be incorporated by reference. By their citation of various references in this document, Applicants do not admit any particular reference is “prior art” to their disclosure.80SUBSTITUTE SHEET (RULE 26)

Claims

What is claimed:

1. A method of identifying repeat element types, comprising extracting repeat sequences and genomic coordinates from known repeat element types; selecting nucleic acid sequences (kmers) occurring in a single repeat type and identifying unique kmers of repeat element types; wherein the unique kmers identify one or a plurality of repeat element types, and identifying the type and / or frequency of kmers in genomic or cell free nucleic acid sequences.

2. The method of claim 1 wherein the type and / or frequency of kmers in genomic or cell free DNA sequences are identified.

3. The method of claim 1 or 2 wherein the type and / or frequency of kmers in genomic or cell free RNA are identified.

4. The method of any one of claims 1 to 3 wherein the type and frequency of kmers in genomic or cell free nucleic acid sequences are identified.

5. The method of any one of claims 1 to 4 wherein repeat element types are excluded from families comprising low complexity, unknown, simple repeats or combinations thereof.

6. The method of any one of claims 1 to 5 wherein elements from each family comprising tRNAs, srpRNAs, snRNAs, scRNAs, rRNAs, RNA elements, DNA, retroposons, retrotransposons, or combinations thereof, are aggregated.

7. The method of any one of claims 1 to 6 wherein the families comprise long interspersed nuclear elements (LINEs), short interspersed nuclear elements (SINEs), long terminal repeats (LTRs), satellites, transposable elements, RNA elements or combinations thereof.

8. The method of any one of claims 1 to 7 wherein the kmer comprises up to or less than 200 nucleotides.78 161564170.

19. The method of any one of claims 1 to 8 wherein the kmer comprises up to or less than 100 nucleotides.

10. The method of c any one of claims 1 to 8 wherein the kmer comprises 10 to 40 nucleotides.

11. The method of any one of claims 1 to 8 wherein the kmer comprises 15 to 35 nucleotides.

12. The method of any one of claims 1 to 8 wherein the kmer comprises 20 to 30 nucleotides.

13. The method of any one of claims 1 to 8 wherein the kmer comprises 22 to 28 nucleotides.

14. The method of any one of claims 1 to 8 wherein the kmer comprises 23 to 26 nucleotides.

15. The method of any one of claims 1 to 5 wherein the kmer comprises about 24 nucleotides.

16. The method of any one of claims 1 through 15 wherein kmer repeat landscapes are generated.

17. The method of claim 16 wherein each unique kmer and reverse complement thereof, is counted and kmer counts for each repeat type element are aggregated.

18. The method of any one of claims 1 through 17 wherein regions of genes comprising a plurality of transcripts are counted to identify kmers occurring in each gene.

19. The method of claim 18 wherein genes are ranked by total number of kmers and a corrected kmer density.

20. The method of claim 19 wherein the corrected kmer density comprises the number of types occurring in a gene divided by kmer density.

21. The method of claim 20 wherein the kmer density is total kmers per Mb.79 161564170.

122. The method of any one of claims 1 through 21 wherein repeat element types are increased in genes associated with a cancer diagnosis as compared to a non-cancer control.

23. The method of any one of claims 1 through 22, wherein kmer-defined repeat element type families are altered genome-wide as compared to a normal control.

24. The method of claim 23, wherein the alterations in the genome as compared to a non- cancer genome comprises focal amplifications, deletions, copy-number changes, rearrangements, changes in repeat elements or combinations thereof.

25. The method of any one of claims 1 to 24 wherein repeat element sequences are identified from RNA transcripts.

26. A method of diagnosing and treating cancer in a subject, comprising: generating kmer repeat landscapes from a subject’s biological sample, generating kmer repeat landscapes from a population of normal samples, comparing the kmer repeat landscapes between the subject sample and the normal samples to diagnose the subject as having cancer and, treating the subject diagnosed with cancer with a cancer treatment.

27. The method of claim 26 wherein changes in the kmer repeat landscape are correlated with one or more metrics of genomic instability.

28. The method of claim 27 wherein the one or more metrics of genomic instability comprise entropy, non-modal ploidy fraction, non-diploid fraction, loss of heterozygosity fraction, breakpoint count, tumor mutation burden, ploidy, modal ploidy or combinations thereof.

29. The method of any one of claims 26 through 28 further comprising repeat element types with overlapping kmers and assaying differences in aggregate kmer counts for the repeat element types between tumors with and without focal amplification.80 161564170.

130. The method of any one of claims 26 through 29 wherein the kmer repat landscapes are inputted into a machine-learning or artificial intelligence program to generate a cross-validated score for distinguishing samples containing tumor material from normal samples.

31. The method of any one of claims 26 through 30 wherein the biological sample comprises genomic DNA, cell free DNA (cfDNA), RNA or combinations thereof.

32. The method of any one of claims 26 through 31 wherein the biological sample comprises cell free DNA (cfDNA) and features in cfDNA comprising fragment length and coverage for fragments in regions identified by frequency of histone marks are evaluated.

33. The method of claim 32 wherein localization of histone marks within repeat element types and densities of each histone mark are evaluated.

34. The method of claim 33 wherein ratios of observed kmer counts to expected kmer count variations between repeat element types are evaluated based on histone mark density.

35. The method of any one of claims 26 through 34 wherein the treatment is selected from surgery, adjuvant chemotherapy, neoadjuvant chemotherapy, radiation therapy, hormone therapy, cytotoxic therapy, immunotherapy, adoptive T cell therapy, targeted therapy, and combinations thereof.

36. A method of detecting and monitoring of cancer progression in a subject, comprising: assaying a plurality of features of a subject’s sample to produce a kmer repeat landscape; partitioning one or more features into families, wherein the families comprise long interspersed nuclear elements (LINEs), short interspersed nuclear elements (SINEs), long terminal repeats (LTRs), satellites, transposable elements, RNA elements or combinations thereof, defining a plurality of features comprising fragment length and coverage for fragments in regions identified by frequency of histone marks training of regression models using repeat landscapes and epigenetic profiles, and ensembling the scores together with the regression model to produce a combined score.81 161564170.

137. The method of claim 36 further comprising incorporating genome-wide fragmentation and a score obtained by one or more software models, machine learning models, artificial intelligence or combinations thereof.82 161564170.1

Images