Disease classification using fragment images

Abstract

Techniques for classifying one or more cancers of a subject are described. An example method includes identifying data indicative of cell-free DNA (cfDNA) from a sample derived from a subject. A fragment image is generated based on the data. The fragment image indicates a plurality of fragments among the cfDNA corresponding to a region of a genome. Input data, including the fragment image, is input into a model configured to generate a prediction of whether the subject has at least one cancer. A report is generated based on the prediction.

Description

FMI Docket No.0032-P L&H Docket No.: F171-0008PCT DISEASE CLASSIFICATION USING FRAGMENT IMAGES CROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims priority to U.S. Provisional App. No.63 / 544,134, which was filed on October 13, 2023, is titled “DISEASE CLASSIFICATION USING FRAGMENT IMAGES,” and is incorporated by reference herein in its entirety. BACKGROUND

[0002] Cancer is characterized by uncontrolled division of abnormal cells in the body. However, different types of cancers have different types of clinically relevant attributes. Different types of cancers may have different types of prognoses. For instance, the five-year survival rate of patients with localized Hodgkin lymphoma is significantly higher than the five-year survival rate of patients with small cell lung cancer. Further, different types of cancers or cancer subtypes respond to different types of therapies. For instance, HER2 inhibitors may be effective at treating HER2- positive breast cancers, but may be ineffective at treating certain HER2-negative breast cancers. Accordingly, appropriate clinical diagnosis and treatments can be dependent on determining cancer types.

[0003] Cancer type can be identified based on a histological study of a tissue biopsy sample from a tumor. In some cases, the tissue biopsy is obtained using an invasive surgical procedure during which a patient undergoes general anesthesia. Once the sample is obtained, it can be processed, stained, and reviewed by a pathologist. Based on visual characteristics of the stained sample, the pathologist may be able to identify a type of cancer cell present in the tumor. However, some cancer types are difficult to ascertain using traditional histological techniques. Moreover, other types of disease classifications can be difficult using previously available technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0004] Various aspects of the disclosed methods, devices, and systems are set forth with particularity in the appended claims. A better understanding of the features and advantages of the disclosed methods, devices, and systems will be obtained by reference to the following detailed description of illustrative embodiments and the accompanying drawings, of which:

[0005] FIG.1 illustrates an example environment for predicting cancer types by converting fragmentomic information into images.

[0006] FIG.2 illustrates an example environment illustrating DNA fragments, which can be utilized to analyze cancer cells of a subject.

[0007] FIG.3 illustrates an example fragment image that is generated based on fragments obtained from a sample of a subject.

[0008] FIG. 4 illustrates an example environment for training and utilizing a predictive model to using fragment images. In some cases, the predictive model is used to determine the methylation status of various cells of interest. The methylation status may be used to detect diseases (e.g., cancer types and subtypes), select therapies, and to monitor disease progression.

[0009] FIG.5 illustrates an example of training data utilized to train one or more machine learning models.

[0010] FIG.6 illustrates an example report summarizing predicted categories of a cancer of a subject.

[0011] FIG.7 illustrates an example process for predicting a type of cancer cell by converting fragmentomic data into an image.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0012] FIG.8 illustrates an example process for predicting a type of a cancer cell present in the body of a subject using a CNN.

[0013] FIG.9 illustrates an example environment for sequencing various nucleic acid molecules.

[0014] FIG.10 illustrates one or more devices configured to perform various operations described herein.

[0015] FIGS.11A to 11C illustrate examples of fragment images generated based on a sample obtained from a subject without cancer.

[0016] FIGS.12A to 12D illustrate examples of fragment images generated based on a sample obtained from a subject with cancer.

[0017] FIG. 13 illustrates results from an example convolutional neural network trained to differentiate between fragment images generated based on samples from subjects without cancer and fragment images generated based on samples from subjects with cancer. DETAILED DESCRIPTION

[0018] Various implementations of the present disclosure relate to techniques for detecting and classifying pathologies using images that represent nucleic acid fragments in patient samples. Sequences, lengths, and placements of the fragments represent complex fragmentomic features that can be summarized in images described herein. In particular cases, techniques are used for classifying cancers based on the images.

[0019] According to some examples, DNA fragments present in a patient sample (e.g., a liquid biopsy sample) are sequenced. At least a portion of the DNA fragments are determined to originate from a predetermined genomic region (also referred to as a “genomic interval”). Characteristics of the portion of the DNA fragments are mapped onto a digital image referred to as a “fragment image.” For instance, rows of the fragment image represent individual DNA fragments that overlap the genomic region and columns of the fragment image represent different genomic positions within the genomic region. In particular cases, different channels of the fragment image represent different characteristics of the relevant DNA fragments. For instance, a first channel may represent the presence of a base or base pair in a given DNA fragment at a given genomic position, a second channel may represent the presence of a CG dinucleotide in the given DNA fragment at the given genomic position, a third channel may represent the presence of a methylated cytosine in the given DNA fragment at the given genomic position, and so on. In various implementations described herein, complex sequencing data representing DNA fragments can be summarized and represented into a single digital image.

[0020] By representing complex fragmentomic data in an image format, sophisticated image processing techniques can be used to identify the presence of circulating tumor DNA (ctDNA) in the patient sample and / or to classify a type of cell (e.g., cancer cell) that released the ctDNA in the patient sample. According to some implementations, one or more machine learning (ML) models can be used to perform the classification. For instance, convolutional neural networks (CNNs) can be used to classify one or more cell types present in the patient from which the patient sample originated. The CNNs, for instance, can be trained based on fragment images generated from a population of individuals with known diseases, such as cancer types. Accordingly, the fragment images can be used to classify a disease of the patient.

[0021] Implementations of the present disclosure provide significant improvements to the technical field of medical diagnostics. Previously, clinicians relied on invasive tissue biopsies and complex immunohistochemistry techniquesFMI Docket No.0032-P L&H Docket No.: F171-0008PCT in order to infer a type of cancer cell present in a patient. However, these previous techniques were often highly invasive, and could subject the patient to immense discomfort and harm. In contrast, implementations of the present disclosure enable cancer cell classification using minimally invasive samples, such as liquid biopsy samples. Pathological classifications that were previously achieved using invasive surgical procedures can now be achieved using a simple blood draw procedure.

[0022] Further, accurate pathology-based classifications can be performed by highly trained pathologists. However, patients in rural, underserved, or low-resource settings may not have access to such highly trained pathologists. Using techniques described herein, a sample can be obtained using a care provider that may not have advanced medical training and may be more likely to be available in low-resource settings. The sample can be easily transported to a facility where DNA fragments in the sample can be sequenced and analyzed. Thus, various implementations of the present disclosure enable broader access for accurate disease classification than previous techniques.

[0023] Various analyses described herein cannot be performed in the human mind, or by pen and paper. For example, a fragment image may represent hundreds or thousands of unique fragment sequences and hundreds or thousands of unique genomic positions. Subtle characteristics of the fragment image that are indicative of a cancer classification would be impossible for a clinician to visually identify. However, various predictive models described herein (e.g., CNNs) are capable of identifying these characteristics of the fragment image. Example Definitions

[0024] As used herein, the terms “deoxyribonucleic acid,” “DNA,” “DNA molecule,” and their equivalents, may refer to a polymer of nucleotides (also referred to as “nucleobases”) containing deoxyribose. The nucleotides in DNA include cytosine (C), guanine (G), adenine (A), and thymine (T). Each DNA nucleotide includes a deoxyribose and a phosphate group. An example single-stranded DNA (ssDNA) molecule includes a chain of covalently bonded DNA nucleotides. In the example ssDNA molecule, the phosphate group of the mth nucleotide is covalently bonded to the deoxyribose of the (m-1)th nucleotide, wherein m is a positive integer greater than 2 and less than or equal to the number of DNA nucleotides in the chain. In various examples, DNA is double-stranded and includes two ssDNA molecules that are complementary to one another and coiled around each other in a double helix form. The nucleotides of one ssDNA molecule are hydrogen bonded to the nucleotides of the other ssDNA molecule. In particular, the pyrimidines (A and T) hydrogen bond to each other, and the purines (C and G) hydrogen bond to each other.

[0025] As used herein, the terms “ribonucleic acid,” “RNA,” “RNA molecule,” and their equivalents, may refer to a polymer of nucleotides containing ribose. The nucleotides in RNA include cytosine (C), guanine (G), adenine (A), and uracil (U). Each RNA nucleotide includes a ribose and a phosphate group. In an example RNA molecule, the phosphate group of the nth nucleotide is covalently bonded to the ribose of the (n-1)th nucleotide, wherein n is a positive integer greater than 2 and less than or equal to the number of RNA nucleotides in the chain. Messenger RNA (mRNA) is a type of RNA molecule that is synthesized (or “transcribed”) by RNA polymerase (an enzyme) to be complementary to a gene encoded in a DNA sequence, and is also used by a ribosome to synthesize a polypeptide or protein. An mRNA is therefore an example of a “coding RNA.” In various cases, intron sequences are removed from an mRNA via a process known as “RNA splicing.” MicroRNA (“miRNA”) are single-stranded RNA molecules that perform post-transcriptional gene expression regulation. For instance, a miRNA may bind to a complementary mRNA molecule, thereby cleaving, destabilizing, or otherwise preventing the mRNA molecule from being translated into aFMI Docket No.0032-P L&H Docket No.: F171-0008PCT polypeptide or protein by a ribosome. In various examples, a miRNA has a length in a range of 21 to 23 RNA nucleotides. As used herein, the terms “non-coding RNA” may refer to a type of RNA that is not translated into a protein. Examples of non-coding RNA include miRNA, transfer RNA (tRNA), and ribosomal RNA (rRNA). The term “functional RNA,” and its equivalents, may refer to any RNA molecule that impacts a biological process. For instance, functional RNA may include mRNA, miRNA, tRNA, rRNA, and the like.

[0026] As used herein, the term “base,” and its equivalents, may refer to a monomer of a polymer. For example, a base of DNA or RNA is a nucleotide.

[0027] As used herein, the term “base pair,” and its equivalents, may refer to a pair of complementary DNA nucleotides, which are hydrogen-bonded to one another in a double-stranded DNA molecule. For example, a base pair includes a first base in a first ssDNA and a second base in a second ssDNA, wherein the first and second bases are complementary and hydrogen-bonded to one another.

[0028] As used herein, the terms “nucleotide,” “nucleobase,” “nucleic acid,” “nucleic acid molecule,” and their equivalents, may refer to an organic molecule that includes a nitrogenous base, a sugar, and a phosphate group. In various cases, a nucleotide is a monomer of DNA or RNA. A nucleotide, for instance, is a chemical structure.

[0029] As used herein, the terms “3’ end,” “3-prime end,” and their equivalents, may refer to a terminus of a single- stranded nucleotide polymer that includes a base whose third carbon in its deoxyribose or ribose is bound to a hydroxyl group while being unbound to another base.

[0030] As used herein, the terms “5’ end,” “5-prime end,” and their equivalents, may refer to a terminus of a single- stranded nucleotide polymer that includes a base whose fifth carbon in its deoxyribose or ribose ring is unbound to another base. In some cases, the fifth carbon is bound to a phosphate group.

[0031] As used herein, the “length” of a polymer refers to a number of covalently bonded monomers that are included in the polymer. For instance, the length of a DNA molecule may be the number of covalently bonded nucleotides in at least one strand of the DNA molecule and / or the number of base pairs in the DNA molecule. In various examples, the length of an RNA molecule may be the number of covalently bonded nucleotides in the RNA molecule.

[0032] As used herein, the term “gene,” and its equivalents, refers to a sequence of DNA nucleotides that is transcribed into a functional RNA. The functional RNA, for instance, is RNA that is translated into a polypeptide or protein (e.g., mRNA) or that has some other biological function (e.g., miRNA, tRNA, etc.). A gene is “expressed” when it is used as a template to generate a functional RNA. A subject, for instance, has numerous genes contained in the subject’s genome. A gene may include both introns and exons. As used herein, the term “intron,” and its equivalents, may refer to a subset of DNA nucleotides in a gene that is not used to code for any functional RNA that is expressed by the organism. As used herein, the term “exon,” and its equivalents, may refer to a subset of DNA nucleotides in a gene that is used to code for a functional RNA. For instance, an exon may encode a polypeptide or protein that is expressed by the organism. In various examples, a gene can be represented in data (e.g., as data representative of the sequence of DNA nucleotides in the gene) or as a chemical structure (e.g., as the sequence of DNA nucleotides itself).

[0033] As used herein, the term “genome,” and its equivalents, refers to the aggregate of genes of a subject. In various cases, a genome represents the sequences of several linear DNA molecules that are present in a subject’s chromosomes. A “reference genome” refers to an aggregation of genes of one or more reference subjects. In various cases, a genome is represented in data.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0034] As used herein, the terms “pangenome,” “pan-genome,” “supragenome,” and their equivalents, refers to an aggregate set of genes from multiple subgroups (e.g., strains) within a population (e.g., a clade) of subjects. A pangenome, for example, indicates genes that are present in all subjects within the population, as well as genes that are present in some of the subjects of the population. A pangenome is represented in data, for instance.

[0035] As used herein, the term “transcriptome,” and its equivalents, refers to the aggregate of RNA sequences of a subject. In some cases, a transcriptome is limited to mRNA sequences. In various examples, a transcriptome is represented in data.

[0036] As used herein, the term “genomic DNA,” “gDNA,” “chromosomal DNA,” and their equivalents, may refer to DNA molecules that are obtained from a chromosome and / or nucleus of a cell.

[0037] As used herein, the terms “DNA fragment,” “fragment,” and their equivalents, may refer to DNA molecules that are excised and / or broken off from a larger DNA molecule.

[0038] As used herein, the terms “cell-free DNA,” “cfDNA,” and their equivalents, may refer to DNA fragments that are non-encapsulated and obtained outside of cells within a sample (e.g., a liquid biopsy sample).

[0039] As used herein, the terms “circulating tumor DNA,” “ctDNA,” and their equivalents, may refer to a cfDNA molecule that originates from a cancer cell.

[0040] As used herein, the terms “end motif,” “terminal sequences,” and their equivalents, may refer to a sequence of nucleotides extending from a 3’ or 5’ end of a DNA or RNA molecule. In various cases, the end motif is shorter than a length of the DNA or RNA molecule. For example, the end motif may have a length in a range of 5 to 30 bases or base pairs, a range of 3 to 30 bases or base pairs, or a range of 1 to 30 base pairs.

[0041] As used herein, the term “promoter,” and its equivalents, may refer to a portion of a DNA molecule that binds one or more proteins in order to initiate transcription of a gene. For example, the promotor is located “upstream” of the gene. For example, the promotor is located between the 5’ end of the DNA molecule and the gene. A promotor may include one or more binding sites for RNA polymerase, and / or one or more transcription factor binding sites. In some examples, a promotor includes one or more CpG islands. A promoter, for instance, includes a transcription start site.

[0042] As used herein, the terms “CpG island,” “CGI,” “CpG site,” and their equivalents, may refer to a continuous portion of a DNA molecule whose sequence includes greater than a threshold amount (e.g., greater than 50%) of G-C base pairs.

[0043] As used herein, the term “enhancer,” and its equivalents, may refer to a portion of a DNA molecule that binds one or more proteins in order to increase the chance that a gene will be transcribed. For instance, an enhancer includes one or more transcription factor binding sites. In various cases, an enhancer includes one or more CpG islands.

[0044] As used herein, the term “cancer,” and its equivalents, may refer to a condition of a subject in which particular cells (referred to as “cancer cells”) divide uncontrollably in the subject’s body. In some cases, a cancer is characterized by a location or tissue type from which the cancer cells originated. In some examples, a cancer is characterized by a location or tissue type in which the cancer cells are located.

[0045] As used herein, the terms “tumor,” “neoplasm,” and their equivalents, may refer to a mass of tissue including cancer cells.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0046] As used herein, the terms “tissue of origin,” “tissue origin,” and their equivalents, refers to a differentiated type of tissue from which cancer cells in the body of a subject began dividing uncontrollably in the subject’s body.

[0047] As used herein, the terms “liquid biopsy,” “fluid biopsy,” and their equivalents, may refer to a process of obtaining a fluid sample from a subject’s body. The sample, for instance, can be referred to as a “liquid biopsy sample.” Examples of fluids that are sampled from the body include blood, plasma, cerebrospinal fluid, sputum, stool, urine, lymphatic fluid, and saliva.

[0048] As used herein, the term “tissue biopsy,” and its equivalents, may refer to a process of obtaining a sample of cells from a subject’s body. A tissue biopsy, in various cases, is performed by cutting a mass of cells from the subject’s body. For instance, a tissue biopsy is a procedure performed by a surgeon, interventional radiologist, interventional cardiologist, or other specialized clinician. The term “tissue” or “tissue biopsy sample” can be used to refer to the sample of cells obtained using a tissue biopsy.

[0049] As used herein, the term “subject,” and its equivalents, may refer to a human or non-human animal. A subject that is receiving care from at least one care provider may be referred to as a “patient.”

[0050] As used herein, the terms “machine learning,” “ML,” “computer learning,” “artificial intelligence,” and their equivalents, may refer to the use of a computing devices to learn patterns in training data. The process of learning these patterns may be referred to as “training.” In particular cases, one or more computing devices may perform machine learning by executing a machine learning model. As used herein, the terms “machine learning model,” “ML model,” and their equivalents, may refer to data encoding instructions that, when executed by at least one computing device, causes the at least one computing device to learn patterns in training data by optimizing one or more metrics, values, or other types of parameters. After training, an ML model, when executed by at least one computing device, causes the at least one computing device to utilize the optimized parameters in order to perform one or more tasks.

[0051] As used herein, the terms “deep learning,” “DL,” and their equivalents, may refer to a type of ML in which multi-layered, artificial neural networks are used to learn patterns in training data. An example layer, for instance, is configured to perform one or more transformations on data output by a previous layer, and outputs the transformed data to a subsequent layer in the neural network. Examples of DL models include, for instance, autoencoders, deep stacking networks, convolutional neural networks (CNNs), probabilistic neural networks (PNNs), time delay neural networks (TDNNs), and the like.

[0052] As used herein, the term “variant,” and its equivalents, may refer to a difference between a subject genetic sequence and a reference sequence. For instance, a variant may correspond to a difference between one or more nucleotides in a genome of a subject and one or more corresponding nucleotides in at least one reference genome or pangenome. A variant may be characterized by its identity (e.g., what nucleotides are different), its position (e.g., where are the nucleotides located in the genome, what chromosome contains the nucleotides, what gene contains the nucleotides, etc.), its length (e.g., how many nucleotides are different from the reference sequence), its type (e.g., substitution, insertion, deletion, copy number alternation, rearrangement of fusion, etc.), and other features that indicates its significance and / or relevance. In some cases, a variant represents any apparent alteration in a sequence that has been read from a nucleic acid molecule with respect to the reference sequence, such as reads cleaved by restriction enzymes (RE). In various examples, a variant can be represented in data (e.g., by data characterizing the variant) or as a chemical structure (e.g., the nucleotides themselves). As used herein, the term “mutation,” and its equivalents, may refer to a change in a gene.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0053] As used herein, the term “substitution,” and its equivalents, can refer to a nucleotide in a subject sequence that is different than an equivalent nucleotide (e.g., a nucleotide at the same position) in a reference sequence.

[0054] As used herein, the term “insertion,” and its equivalents, can refer to a nucleotide in a subject sequence that is added with respect to a reference sequence.

[0055] As used herein, the term “deletion,” and its equivalents, can refer to the removal of a nucleotide from a nucleotide sequence.

[0056] As used herein, the terms “copy number alternation,” “CNA,” “copy number variation,” “CNV,” and their equivalents, can refer to a portion of a reference sequence that is repeated.

[0057] As used herein, the terms “rearrangement of fusion,” “fusion rearrangement,” “translocation,” and their equivalents, can refer to a change in the relative position of one or more portions of a reference sequence, thereby generating a gene that was not present in the reference sequence.

[0058] As used herein, the term “sequencing,” and its equivalents, may refer to a process of identifying the order and identity of monomers in a polymer chain, such as the order and identity of nucleotides in a DNA or RNA molecule. The terms “whole genome sequencing,” “WGS,” and their equivalents, may refer to the process of sequencing an entire genome of a subject, including the introns and exons of the genes of the subject. The term “whole exome sequencing,” and its equivalents, may refer to the process of sequencing all exomes of a subject. The term “targeted sequencing,” and its equivalents, may refer to the process of sequencing a portion of the genome of a subject, such as sequencing a single gene of the subject. Various techniques can be utilized to sequence a DNA or RNA molecule, such as massively parallel sequencing (MPS), nanopore sequencing, direct sequencing, Sanger sequencing, or next- generation sequencing. In various cases, sequencing is performed on physical molecules (e.g., RNA or DNA) and is used to generate data.

[0059] As used herein, the terms “massive parallel sequencing,” “massively parallel sequencing,” “MPS,” and their equivalents, may refer to a technique for simultaneously performing multiple reactions that can be used to identify the order and identity of monomers in multiple polymer chains. In particular cases, massive parallel sequencing can be performed using sequencing-by-synthesis on clonally amplified DNA molecules that are located in spatially separated regions, which are individually monitored by sensors.

[0060] As used herein, the term “nanopore sequencing,” and its equivalents, may refer to a technique for identifying the order and identity of monomers in a polymer chain by transporting the polymer chain from a first space to a second space, wherein the first space and the second space are separated by a substrate, by directing the polymer chain through a small hole (known as a “nanopore”) embedded in the substrate, and monitoring a relative electrical signal (e.g., a voltage or current) between the first space and the second space.

[0061] As used herein, the term “sensor,” and its equivalents, may refer to a physical device or other apparatus that is configured to detect one or more detection signals.

[0062] As used herein, the term “detection signal,” and its equivalents, may refer to a physical signal that can be identified, characterized, or otherwise perceived by a sensor.

[0063] As used herein, the term “sequence read data,” and its equivalents, may refer to data that is indicative of an order and identity of monomers in a polymer, such as the order and identity of nucleotides in a DNA or RNA sequence. In various implementations, sequence read data is generated via a sequencing operation.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0064] As used herein, the term “image,” and its equivalents, may refer to 2D or 3D array of data indicative of an array of pixels or voxels.

[0065] As used herein, the term “ligating,” and its equivalents, may refer to a process of joining two molecules together, for example, with a chemical bond.

[0066] As used herein, the term “adapter,” and its equivalents, may refer to an oligonucleotide that can be ligated to a target nucleic acid molecule. In various cases, an adapter prepares the target nucleic acid molecule for sequencing.

[0067] As used herein, the term “bait molecule,” and its equivalents, may refer to a nucleic acid molecule having a portion that is complementary to a portion of a target molecule (e.g., cfDNA). A bait molecule includes, for instance, a nucleic acid molecule that can hybridize to (i.e., is complementary to) a target molecule can be used to capture the target molecule. In some instances, the bait molecule is a capture oligonucleotide (or capture probe). In some instances, the bait molecule is suitable for solution phase hybridization to the target molecule. In some instances, the bait molecule is suitable for solid phase hybridization to the target molecule. In some instances, the bait molecule is suitable for both solution-phase and solid-phase hybridization to the target molecule. The design and construction of bait molecules is described in more detail in, e.g., International Patent Application Publication No. WO 2020 / 236941.

[0068] As used herein, the term “amplifying,” and its equivalents, may refer to a process of generating copies of a target molecule, such as a nucleic acid molecule.

[0069] As used herein, the term “hybridization,” and its equivalents, may refer to a process by which to complementary single-stranded nucleic acid molecules bind to one another, thereby forming a double-stranded nucleic acid molecule. In certain examples, the double-stranded nature of the nucleic acid molecule is maintained under stringent hybridization conditions. Exemplary stringent hybridization conditions include an overnight incubation at 42 °C in a solution including 50% formamide, 5XSSC (750 mM NaCl, 75 mM trisodium citrate), 50 mM sodium phosphate (pH 7.6), 5XDenhardt's solution, 10% dextran sulfate, and 20 µg / ml denatured, sheared salmon sperm DNA, followed by washing the filters in 0.1XSSC at 50 °C.

[0070] As used herein, the term “complementary,” and its equivalents, may refer to a state of two single-stranded nucleic acid molecules with respective sequences that cause the nucleic acid molecules to spontaneously hybridize to one another. One nucleic acid molecule, for instance, may have a sequence that causes each nucleic acid to hydrogen bond to a respective nucleic acid in the other nucleic acid molecule.

[0071] As used herein, the terms “therapy,” “treatment,” and their equivalents, may refer to a composition or process that can be used to remediate a health problem. Cancer therapies, for instance, include surgery, radiotherapy, chemotherapy, immunotherapy, cell-based therapies, and the like. Examples of cancer therapies include abemaciclib (Verzenio), abiraterone acetate (Zytiga), acalabrutinib (Calquence), ado-trastuzumab emtansine (Kadcyla), afatinib dimaleate (Gilotrif), aldesleukin (Proleukin), alectinib (Alecensa), alemtuzumab (Campath), alitretinoin (Panretin), alpelisib (Piqray), amivantamab-vmjw (Rybrevant), anastrozole (Arimidex), apalutamide (Erleada), asciminib hydrochloride (Scemblix), atezolizumab (Tecentriq), avapritinib (Ayvakit), avelumab (Bavencio), axicabtagene ciloleucel (Yescarta), axitinib (Inlyta), belantamab mafodotin-blmf (Blenrep), belimumab (Benlysta), belinostat (Beleodaq), belzutifan (Welireg), bevacizumab (Avastin), bexarotene (Targretin), binimetinib (Mektovi), blinatumomab (Blincyto), bortezomib (Velcade), bosutinib (Bosulif), brentuximab vedotin (Adcetris), brexucabtagene autoleucel (Tecartus), brigatinib (Alunbrig), cabazitaxel (Jevtana), cabozantinib (Cabometyx), cabozantinib (Cabometyx,FMI Docket No.0032-P L&H Docket No.: F171-0008PCT Cometriq), canakinumab (Ilaris), capmatinib hydrochloride (Tabrecta), carfilzomib (Kyprolis), cemiplimab-rwlc (Libtayo), ceritinib (LDK378 / Zykadia), cetuximab (Erbitux), cobimetinib (Cotellic), copanlisib hydrochloride (Aliqopa), crizotinib (Xalkori), dabrafenib (Tafinlar), dacomitinib (Vizimpro), daratumumab (Darzalex), daratumumab and hyaluronidase-fihj (Darzalex Faspro), darolutamide (Nubeqa), dasatinib (Sprycel), denileukin diftitox (Ontak), denosumab (Xgeva), dinutuximab (Unituxin), dostarlimab-gxly (Jemperli), durvalumab (Imfinzi), duvelisib (Copiktra), elotuzumab (Empliciti), enasidenib mesylate (Idhifa), encorafenib (Braftovi), enfortumab vedotin-ejfv (Padcev), entrectinib (Rozlytrek), enzalutamide (Xtandi), erdafitinib (Balversa), erlotinib (Tarceva), everolimus (Afinitor), exemestane (Aromasin), fam-trastuzumab deruxtecan-nxki (Enhertu), fedratinib hydrochloride (Inrebic), fulvestrant (Faslodex), gefitinib (Iressa), gemtuzumab ozogamicin (Mylotarg), gilteritinib (Xospata), glasdegib maleate (Daurismo), hyaluronidase-zzxf (Phesgo), ibrutinib (Imbruvica), ibritumomab tiuxetan (Zevalin), idecabtagene vicleucel (Abecma), idelalisib (Zydelig), imatinib mesylate (Gleevec), infigratinib phosphate (Truseltiq), inotuzumab ozogamicin (Besponsa), iobenguane I131 (Azedra), ipilimumab (Yervoy), isatuximab-irfc (Sarclisa), ivosidenib (Tibsovo), ixazomib citrate (Ninlaro), lanreotide acetate (Somatuline Depot), lapatinib (Tykerb), larotrectinib sulfate (Vitrakvi), Lenvatinib mesylate (Lenvima), letrozole (Femara), lisocabtagene maraleucel (Breyanzi), loncastuximab tesirine-lpyl (Zynlonta), lorlatinib (Lorbrena), lutetium Lu 177-dotatate (Lutathera), margetuximabcmkb (Margenza), midostaurin (Rydapt), mobocertinib succinate (Exkivity), mogamulizumab-kpkc (Poteligeo), moxetumomab pasudotox-tdfk (Lumoxiti), naxitamab-gqgk (Danyelza), necitumumab (Portrazza), neratinib maleate (Nerlynx), nilotinib (Tasigna), niraparib tosylate monohydrate (Zejula), nivolumab (Opdivo), obinutuzumab (Gazyva), ofatumumab (Arzerra), olaparib (Lynparza), olaratumab (Lartruvo), osimertinib (Tagrisso), palbociclib (Ibrance), panitumumab (Vectibix), panobinostat (Farydak), pazopanib (Votrient), pembrolizumab (Keytruda), pemigatinib (Pemazyre), pertuzumab (Perjeta), pexidartinib hydrochloride (Turalio), polatuzumab vedotin-piiq (Polivy), ponatinib hydrochloride (Iclusig), pralatrexate (Folotyn), pralsetinib (Gavreto), radium 223 dichloride (Xofigo), ramucirumab (Cyramza), regorafenib (Stivarga), ribociclib (Kisqali), ripretinib (Qinlock), rituximab (Rituxan), rituximab and hyaluronidase human (Rituxan Hycela), romidepsin (Istodax), rucaparib camsylate (Rubraca), ruxolitinib phosphate (Jakafi), sacituzumab govitecanhziy (Trodelvy), seliciclib, selinexor (Xpovio), selpercatinib (Retevmo), selumetinib sulfate (Koselugo), siltuximab (Sylvant), sipuleucel-T (Provenge), sirolimus protein-bound particles (Fyarro), sonidegib (Odomzo), sorafenib (Nexavar), sotorasib (Lumakras), sunitinib (Sutent), tafasitamab-cxix (Monjuvi), tagraxofusp-erzs (Elzonris), talazoparib tosylate (Talzenna), tamoxifen (Nolvadex), tazemetostat hydrobromide (Tazverik), tebentafusp-tebn (Kimmtrak), temsirolimus (Torisel), tepotinib hydrochloride (Tepmetko), tisagenlecleucel (Kymriah), tisotumab vedotin-tftv (Tivdak), tocilizumab (Actemra), tofacitinib (Xeljanz), tositumomab (Bexxar), trametinib (Mekinist), trastuzumab (Herceptin), tretinoin (Vesanoid), tivozanib hydrochloride (Fotivda), toremifene (Fareston), tucatinib (Tukysa), umbralisib tosylate (Ukoniq), vandetanib (Caprelsa), vemurafenib (Zelboraf), venetoclax (Venclexta), vismodegib (Erivedge), vorinostat (Zolinza), zanubrutinib (Brukinsa), ziv-aflibercept (Zaltrap), and combinations thereof. Examples of cancer therapies also include targeted antibody-based therapies (antibody-drug conjugates, antibody-radioisotope conjugates, and targeted immune cell therapies (e.g., immune effector cells genetically modified to express a chimeric antigen receptor (CAR).

[0072] As used herein, the term “treatment-responsive,” and its equivalents, may refer to a type of cancer cells that can be substantially killed using a predetermined type of therapy. For example, cancer cells of a subject may be responsive to a particular treatment if, after the subject is administered the treatment, the cancer cells are diminishedFMI Docket No.0032-P L&H Docket No.: F171-0008PCT by a particular progression level (e.g., radiographic progression level, marker-based progression level, such as prostate-specific antigen (PSA) progression, etc.). Accordingly, the responsiveness of the cells to the type of therapy may indicate the effectiveness of that therapy.

[0073] As used herein, the term “treatment-resistant,” and its equivalents, may refer to a type of cancer that cannot be substantially killed using a predetermined type of therapy.

[0074] As used herein, the term “metastasis profile,” and its equivalents, may refer to a propensity of a type of cancer to metastasize into one or more differentiated tumor types besides the cancer’s tissue origin. In some implementations, the metastasis profile can further indicate the type of tissue in which the cancer can or is likely to metastasize.

[0075] As used herein, the term “clinical trial,” and its equivalents, may refer to a research study used to evaluate a hypothesis based on participation by one or more subjects. In various examples, a clinical trial can be used to assess the efficacy and / or safety of a proposed therapy. A clinical trial may be performed in furtherance of approval of a treatment by a regulatory authority (e.g., the United States Food & Drug Administration (FDA)).

[0076] As used herein, the terms “image,” “image data,” “digital image,” and their equivalents, may refer to a multidimensional data object including multiple picture elements with associated values. A two-dimensional (2D) image, for instance, includes a matrix of picture elements arranged in one or more rows and one or more columns. The picture elements of a 2D image may be referred to as “pixels.” A three-dimensional (3D) image, for example, includes a matrix of picture elements arranged in one or more first rows, one or more second rows, and one or more columns. The picture elements of a 3D image may be referred to as “voxels.” Each picture element within an image is defined according to one or more channels. For example, a picture element defined by multiple channels includes multiple values that respectively correspond to the multiple channels.

[0077] As used herein, the term “channel,” and its equivalents, may refer to a component of a picture element of an image that defines the intensity or magnitude of a color component of the picture element. In grayscale images, each picture element may be defined according to single value (e.g., a number between 0 and 255) that represents the intensity of the picture element along a scale between black (e.g., 0) and white (e.g., 255). In color images, each picture element may be defined according to multiple values. For instance, each picture element of a red-green-blue (RGB) image is defined according to a first value that represents the red intensity of the picture element, a second value that represents the green intensity of the picture element, and a third value that represents the blue intensity of the picture element. Picture elements may be optionally defined according to any of various color spaces, such as cyan-magenta-yellow (CMY), YUV, and the like.

[0078] As used herein, the terms “genomic region,” “region of a genome,” and their equivalents, may refer to a continuous portion of a genome that is shorter than a chromosome within the genome.

[0079] As used herein, the term “genomic position,” and its equivalents, may refer to a location of one or more DNA base pairs relative to a reference genome. In some cases, a genomic position is defined according to a locus. In some cases, the genomic position may refer to a chromosome on which the base pair(s) are located, an arm (e.g., a p arm or a q arm) of the chromosome on which the base pair(s) are located, a distance (e.g., in terms of bases or megabases) from a centromere of the chromosome, a distance from a telomere on the p arm of the chromosome, a distance from a telomere on the q arm of the chromosome, a distance from an end of the p arm of the chromosome, a distance from an end of the q arm of the chromosome, or any combination thereof.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0080] As used herein, the term “start position,” and its equivalents, may refer to a location of a first end of a fragment in a genome, such as a location of the first end of the fragment in a chromosome. As used herein, the term “end position,” and its equivalents, may refer to a location of a second end of the fragment in the genome, such as a location of the second end of the fragment in the chromosome. As used herein, the term “middle position,” and its equivalents, may refer to a location of a center base of a fragment in the genome, such as a location of the center base of the fragment in the chromosome. If the fragment includes an even number of bases, the middle position may correspond to either center base of the fragment. Description of Example Implementations

[0081] Various implementations of the present disclosure will now be described with reference to the accompanying Figures.

[0082] FIG.1 illustrates an example environment 100 for predicting cancer types by converting fragmentomic (e.g., methylation information) into images. A subject 102, for instance, may present to a clinical environment with a lesion 104. In various cases, the lesion 104 may be a tumor that includes cancer cells. According to various examples, the subject 102 has one or more types of cancer, such as adrenal cancer, bladder cancer, blood cancer, bone cancer, brain cancer, breast cancer, carcinoma, cervical cancer, colon cancer, colorectal cancer, corpus uterine cancer, ear, nose and throat (ENT) cancer, endometrial cancer, esophageal cancer, gastrointestinal cancer, head and neck cancer, Hodgkin's disease, intestinal cancer, kidney cancer, larynx cancer, leukemia, liver cancer, lymph node cancer, lymphoma, lung cancer, melanoma, mesothelioma, myeloma, nasopharynx cancer, a neuroblastoma, non-Hodgkin's lymphoma, oral cancer, ovarian cancer, pancreatic cancer, penile cancer, pharynx cancer, prostate cancer, rectal cancer, sarcoma, seminoma, skin cancer, stomach cancer, a teratoma, testicular cancer, thyroid cancer, uterine cancer, vaginal cancer, a vascular tumor, or combinations or metastases thereof.

[0083] In some embodiments, the subject 102 has a B cell cancer (multiple myeloma), a melanoma, breast cancer, lung cancer, bronchus cancer, colorectal cancer, prostate cancer, pancreatic cancer, stomach cancer, ovarian cancer, urinary bladder cancer, brain cancer, central nervous system cancer, peripheral nervous system cancer, esophageal cancer, cervical cancer, uterine cancer, endometrial cancer, cancer of an oral cavity, cancer of a pharynx, liver cancer, kidney cancer, testicular cancer, biliary tract cancer, small bowel cancer, appendix cancer, salivary gland cancer, thyroid gland cancer, adrenal gland cancer, osteosarcoma, chondrosarcoma, a cancer of hematological tissue, an adenocarcinoma, an inflammatory myofibroblastic tumor, a gastrointestinal stromal tumor (GIST), colon cancer, multiple myeloma (MM), myelodysplastic syndrome (MDS), myeloproliferative disorder (MPD), acute lymphocytic leukemia (ALL), acute myelocytic leukemia (AML), chronic myelocytic leukemia (CML), chronic lymphocytic leukemia (CLL), polycythemia Vera, Hodgkin lymphoma, non-Hodgkin lymphoma (NHL), soft-tissue sarcoma, fibrosarcoma, myxosarcoma, liposarcoma, osteogenic sarcoma, chordoma, angiosarcoma, endotheliosarcoma, lymphangiosarcoma, lymphangioendotheliosarcoma, synovioma, mesothelioma, Ewing's tumor, leiomyosarcoma, rhabdomyosarcoma, squamous cell carcinoma, basal cell carcinoma, adenocarcinoma, sweat gland carcinoma, sebaceous gland carcinoma, papillary carcinoma, papillary adenocarcinomas, medullary carcinoma, bronchogenic carcinoma, renal cell carcinoma, hepatoma, bile duct carcinoma, choriocarcinoma, seminoma, embryonal carcinoma, Wilms' tumor, bladder carcinoma, epithelial carcinoma, glioma, astrocytoma, medulloblastoma, craniopharyngioma, ependymoma, pinealoma, hemangioblastoma, acoustic neuroma, oligodendroglioma, meningioma, neuroblastoma,FMI Docket No.0032-P L&H Docket No.: F171-0008PCT retinoblastoma, follicular lymphoma, diffuse large B-cell lymphoma, mantle cell lymphoma, hepatocellular carcinoma, thyroid cancer, gastric cancer, head and neck cancer, small cell cancer, essential thrombocythemia, agnogenic myeloid metaplasia, hypereosinophilic syndrome, systemic mastocytosis, familiar hypereosinophilia, chronic eosinophilic leukemia, neuroendocrine cancers, or a carcinoid tumor.

[0084] In some embodiments, the subject 102 has acute lymphoblastic leukemia (Philadelphia chromosome positive), acute lymphoblastic leukemia (precursor B-cell), acute myeloid leukemia (FLT3+), acute myeloid leukemia (with an IDH2 mutation), anaplastic large cell lymphoma, basal cell carcinoma, B-cell chronic lymphocytic leukemia, bladder cancer, breast cancer (HER2 overexpressed / amplified), breast cancer (HER2+), breast cancer (HR+, HER2-), cervical cancer, cholangiocarcinoma, chronic lymphocytic leukemia, chronic lymphocytic leukemia (with 17p deletion), chronic myelogenous leukemia, chronic myelogenous leukemia (Philadelphia chromosome positive), classical Hodgkin lymphoma, colorectal cancer, colorectal cancer (dMMR / MSI-H), colorectal cancer (KRAS wild type), cryopyrin- associated periodic syndrome, a cutaneous T-cell lymphoma, dermatofibrosarcoma protuberans, a diffuse large B-cell lymphoma, fallopian tube cancer, a follicular B-cell non-Hodgkin lymphoma, a follicular lymphoma, gastric cancer, gastric cancer (HER2+), gastroesophageal junction (GEJ) adenocarcinoma, a gastrointestinal stromal tumor, a gastrointestinal stromal tumor (KIT+), a giant cell tumor of the bone, a glioblastoma, granulomatosis with polyangiitis, a head and neck squamous cell carcinoma, a hepatocellular carcinoma, Hodgkin lymphoma, juvenile idiopathic arthritis, lupus erythematosus, a mantle cell lymphoma, medullary thyroid cancer, melanoma, a melanoma with a BRAF V600 mutation, a melanoma with a BRAF V600E or V600K mutation, Merkel cell carcinoma, multicentric Castleman's disease, multiple hematologic malignancies including Philadelphia chromosome-positive ALL and CML, multiple myeloma, myelofibrosis, a non-Hodgkin’s lymphoma, a nonresectable subependymal giant cell astrocytoma associated with tuberous sclerosis, a non-small cell lung cancer, a non-small cell lung cancer (ALK+), a non-small cell lung cancer (PD-L1+), a non-small cell lung cancer (with ALK fusion or ROS1 gene alteration), a non-small cell lung cancer (with BRAF V600E mutation), a non-small cell lung cancer (with an EGFR exon 19 deletion or exon 21 substitution (L858R) mutations), a non-small cell lung cancer (with an EGFR T790M mutation), a non-small cell lung cancer KRAS (+ / - G12C), a non-small cell lung cancer TMB-H, a non-small cell lung cancer MET exon 14 skipping, a non-small cell lung cancer ERBB2 inframe indel, a non-small cell lung cancer EGFR exon 20 indel, ovarian cancer, ovarian cancer (with a BRCA mutation), pancreatic cancer, a pancreatic, gastrointestinal, or lung origin neuroendocrine tumor, a pediatric neuroblastoma, a peripheral T-cell lymphoma, peritoneal cancer, prostate cancer, a renal cell carcinoma, a small lymphocytic lymphoma, a soft tissue sarcoma, a solid tumor (MSI-H / dMMR), a squamous cell cancer of the head and neck, a squamous non-small cell lung cancer, thyroid cancer, a thyroid carcinoma, urothelial cancer, a urothelial carcinoma, or Waldenstrom's macroglobulinemia.

[0085] In various cases, a care provider 105 (also referred to as a “healthcare provider”) is responsible for diagnosing and / or treating the subject 102. According to some implementations, the lesion 104 may be initially identified using a noninvasive technique. For example, the lesion 104 may be visualized using an imaging modality, such as ultrasound, x-ray, computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), single photon emission CT (SPECT), or any combination thereof. Using the noninvasive technique, the care provider 105 may identify the presence of the lesion 104, but may be unable to determine whether the lesion 104 is a cancerous tumor using noninvasive diagnostic methodologies. In some cases in which the lesion 104 is a tumor, the care providerFMI Docket No.0032-P L&H Docket No.: F171-0008PCT 105 may be unable to identify whether the tumor is metastatic or benign. In some examples, the care provider 105 is unable to determine a therapy for treating the tumor effectively. For instance, a particular therapy (e.g., radiotherapy) may be effective at treating some types of cancers, but not other types of cancers. Therefore, it may be beneficial for the care provider 105 to identify the type of cancer cells present in the lesion 104 or other parts of the body of the subject 102.

[0086] The care provider 105 could determine a type of cancer cell present in the lesion 104 by initiating a tissue biopsy on the subject 102. For instance, the care provider 105 could surgically remove a tissue sample from the lesion 104 and / or review the tissue sample using histochemistry and / or immunohistochemistry in order to classify the lesion 104. Tumor classifications, for instance, can be indicative of responsiveness to anticancer therapies. However, attempting to classify the lesion 104 using tissue biopsy has several drawbacks. First, the tissue biopsy could be a highly invasive surgical procedure, which can cause significant discomfort to the subject 102. Second, the tissue biopsy may require the subject 102 to undergo general anesthesia, which could be dangerous to the subject 102. Third, even if the tissue biopsy was performed and a tissue sample was obtained, it may not be classifiable using conventional histological techniques, such as conventional immunohistochemical staining and review. Fourth, it is unlikely that the single care provider 105 would be trained to perform the tissue biopsy (which would be performed by a surgeon), to administer anesthesia to the subject 102 during the tissue biopsy (which would be performed by an anesthesiologist), and the analysis of the tissue biopsy (which would be performed by a trained pathologist), such that the classification would utilize multiple highly trained care providers. Even if the cells in the lesion 104 could be analyzed by these means, the coordinated efforts of these care providers could delay diagnosis and treatment of the lesion 104, and could cause significant expense to the subject 102. In various examples, the delay in diagnosis and treatment could cause significant emotional hardship to the subject 102. Further, the delay in diagnosis and treatment could delay a therapy of the lesion 104, which could cause lasting harm to the subject 102, particularly in cases in which the lesion 104 is representative of an aggressive form of cancer. Notably, if the subject 102 is located in a low- resource setting or rural clinical environment, the subject 102 may be unable to participate in the tissue biopsy without traveling to a clinical environment that is capable of performing and analyzing the tissue biopsy, causing further delays and disruptions.

[0087] In various implementations of the present disclosure, cancer cells in the lesion 104 and / or other regions of the body of the subject 102 can be classified without requiring a tissue biopsy. For instance, a sample 106 is obtained from the subject 102. The sample 106, for instance, is a liquid biopsy sample including pleural lavage, blood, plasma, cerebrospinal fluid, sputum, stool, urine, lymphatic fluid, saliva, or some other fluid obtained from the body of the subject 102. In some cases, a blood sample is obtained intravenously from the subject 102. The sample 106, according to various examples, is a plasma sample obtained from the blood of the subject 102. The sample 106 can be obtained in a minimally invasive procedure, which could be performed by a medical technician rather than a surgeon. In some cases, the sample 106 includes a tissue sample.

[0088] The sample 106 includes nucleic acid molecules in the form of cfDNA. In examples in which the subject 102 has cancer (e.g., the lesion 104 is a cancerous tumor), the cfDNA, for instance, includes circulating tumor DNA (ctDNA) 108 as well as non-ctDNA 110. In cases wherein the lesion 104 is a tumor, cancer cells within the lesion 104 will lyse and release the ctDNA 108 into the bloodstream of the subject 102. Further, other cells additionally release non-ctDNAFMI Docket No.0032-P L&H Docket No.: F171-0008PCT into the bloodstream of the subject 102. In general, the cfDNA includes fragments with lengths that are in a range of 1 to 500, 3 to 500, or 100 to 500 bases long. For instance, the cfDNA includes fragments that are about 170 bases long and / or fragments that are about 340 bases long. For example, the cfDNA includes fragments that are 100 to 240 bases long and / or fragments that are 270 to 410 bases long. As will be described in further detail with respect to FIG.2, the features of the ctDNA 108 are indicative of the expression of the cancer cells within the lesion 104. That is, the features of the ctDNA 108 may be indicative of one or more genes that are expressed by the cancer cells.

[0089] In various cases, the sample 106 is transported to a location that is remote from the subject 102 for further processing. For example, the sample 106 is removed from the subject 102 in a clinical environment (e.g., a hospital) and is then transported to a remote laboratory for further testing and analysis. In some cases, the sample 106 is transported via at least one car, van, plane, train, or the like. The sample 106, for instance, is refrigerated and / or frozen during transport.

[0090] A sequencer 112 is configured to generate sequence read data 114 indicating the sequences of the ctDNA 108 and, optionally, the non-ctDNA 110. Non-ctDNA sequencing is considered optional, for example, when pre- analytical means to enrich for the ctDNA component of cfDNA are used to generate the sequencing library (e.g., oversampling of shorter cfDNA fragments for inclusion in the sequencing library). The sequencer 112, for instance, includes one or more devices that are configured to generate the sequence read data 114 by processing at least a portion of the sample 106. In some cases, the cfDNA including the ctDNA 108 and the non-ctDNA 110 is extracted from the sample 106. The extraction can be performed by the sequencer 112, by another device, manually (e.g., by a laboratory technician), or any combination thereof. Any appropriate extraction method known to those of ordinary skill in the art can be utilized.

[0091] In various cases, the sequencer 112 is configured to perform one or more processes (e.g., chemical reactions) on the cfDNA in order to prepare the cfDNA for sequencing. For instance, the sequencer 112 may ligate adapters onto the cfDNA and / or amplify the cfDNA, such that numerous copies of the ligated cfDNA are available for sequencing. Examples of the adapters include, for example, amplification primers, flow cell adapter sequences, substrate adapter sequences, or sample index sequences. The cfDNA (e.g., the ligated cfDNA) may be amplified by generating multiple copies of the cfDNA using one or more techniques such as polymerase chain reaction (PCR), a non-PCR amplification technique, or an isothermal amplification technique.

[0092] The sequencer 112 and any additional downstream data analysis may identify the length, position, and identity of the bases in the cfDNA by sequencing the cfDNA (e.g., the amplified and / or ligated cfDNA). In various implementations, the sequencer 112 utilizes first-generation sequencing (e.g., Sanger sequencing), second-generation sequencing (e.g., massive parallel sequencing), third-generation sequencing (e.g., nanopore sequencing), or a combination thereof. In some cases, the sequencer 112 is configured to sequence substantially all of the nucleotides of all of the cfDNA fragments obtained from the sample 106. In some examples, the sequencer 112 is configured to perform targeted sequencing. For instance, the sequencer 112 may determine whether the cfDNA fragments contain one or more predetermined sequences.

[0093] In various cases, the sequencer 112 includes one or more sensors that are configured to detect physical signals (also referred to as “detection signals”) that are indicative of the nucleotide sequences of the cfDNA fragments. The sequencer 112 may perform sequencing-by-synthesis. For example, the sequencer 112 may include one or moreFMI Docket No.0032-P L&H Docket No.: F171-0008PCT optical sensors configured to detect optical signals emitted from fluorescently tagged dNTPs that are joined together in a synthesized DNA strand using the ligated cfDNA as templates. The optical signals detected by the optical sensor(s), for instance, are indicative of the sequences of the cfDNA. The sequencer 112 may perform nanopore sequencing. In various cases, the sequencer 112 includes one or more electrical sensors configured to measure an electrical signal (e.g., an electrical current) across a substrate as the ligated cfDNA fragments are directed through a nanopore extending through the substrate. The electrical signal over time, in various cases, is indicative of the sequences of the cfDNA in the sample 106. The sequencer 112, in various implementations, is configured to generate the sequence read data 114 as digital data based on the analog signals detected by the sensor(s). For instance, the sequencer 112 includes one or more analog to digital converters (ADCs). In various cases, the sequencer 112 includes at least one processor configured to generate the sequence read data 114.

[0094] In some examples, the sequencer 112 performs methylation sequencing. For example, the sequencer 112 may expose the cfDNA to a reagent (e.g., including bisulfite, TET2, T4-BGT, APOBEC, etc.) that causes a portion of the cytosines in the cfDNA to be converted to uracils. In some cases (e.g., traditional methylation sequencing (bisulfite sequencing as an example or enzymatic-methyl sequencing (EM-seq) as another example), the portion converted to uracils includes unmethylated cytosines. In some examples (e.g. TET-assisted pyridine borane sequencing (TAPS)), the portion converted to uracils includes methylated cytosines. The remaining portion of the cytosines in the cfDNA remain as cytosines, for instance. The methylation sequencing, for instance, indicate the presence of hydroxymethylated cytosines.

[0095] When the converted cfDNA is amplified, the amplified cfDNA includes thiamines at the positions of the converted uracils and includes cytosines at the positions of the unconverted cytosines. When the converted cfDNA is subsequently sequenced by the sequencer 112 to generate the sequence read data 114, the sequencer 112, for instance, may compare the sequences of the cfDNA indicated in the sequence read data 114 to at least one reference sequence (e.g., a reference genome) in order to identify which of the cytosines have been converted (e.g., to uracil). In various implementations, the sequencer 112 may identify which of the cytosines in the cfDNA in the sample 106 were methylated based on the comparison of the cfDNA sequences in the sequence read data 114 to the reference sequence(s). In various implementations, indications of the position, order, and amount of the methylated cytosines is further indicated in the sequence read data 114. This information may be referred to as “methylation data.” DNA methylation is an epigenetic factor that can impact genetic expression. For instance, higher methylation at promoters can result in lowered expression of associated genes. In some cases, methylation is indicative of cancer cell type.

[0096] For instance, methylation data may be used to classify types of colorectal cancer (CRC). In various cases, global hypomethylation, which can lead to chromosomal instability (CIN), is an indicator of CFC type. Further, CpG island methylator phenotype (CIMP) may extensively display multiple cancer specific genes promoter DNA hypermethylation at some specific set of CpG islands, which is associated with CRC type.

[0097] In some implementations, sequences representing the ctDNA 108 and sequences representing the non- ctDNA 110 in the sequence read data 114 are differentiated from one another. For instance, the sequences representing the non-ctDNA 110 may be removed from the sequence read data 114. In some examples, the sequencer 112 and / or another computing device removes the sequences representing the non-ctDNA 110 from the sequence read data 114.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0098] Various features can be used to identify sequences corresponding to the ctDNA 108 rather than the non- ctDNA. In various implementations, the sequencer 112 identifies the sequences corresponding to the ctDNA 108 based on the lengths of the sequences indicated by the sequence read data 114. For instance, sequences with lengths over a predetermined threshold may be defined as corresponding to the ctDNA 108. In various examples, the sequencer 112 identifies sequences corresponding to the ctDNA 108 based on the presence of one or more predetermined variants associated with cancer. In various implementations, the sequencer analyzes the sequences of the fragments represented by the sequence read data 114 in order to determine which of the sequences correspond to the ctDNA 108. However, in some implementations, the sequence read data 114 indicates the sequences and methylation statuses of various cfDNA including the ctDNA 108 and the non-ctDNA 110.

[0099] An image generator 116 is configured to generate a fragment image 118 based on the sequence read data 114. In various implementations, the image generator 116 translates relevant portions of the sequence read data 114 into the single fragment image 118. The fragment image 118 represents various characteristics of the sequences of the cfDNA including the ctDNA 108 and / or the non-ctDNA 110. The cfDNA, in various cases, includes ssDNA.

[0100] In various implementations, at least one dimension of the fragment image 118 represents the collection of unique, fragment sequences (e.g., the sequences of the ctDNA 108 and / or other fragments indicated in the non-ctDNA 110) that overlap a predetermined genomic region and are represented in the sequence read data 114. In some cases, individual rows of the fragment image 118 respectively represent the individual fragments. In some examples, rows of the fragment image 118 include duplicated and / or interpolated rows based on one or more fragments indicated in the sequence read data 114. The number of rows within the fragment image 118 may be dependent on a type of genetic locus being portrayed by the fragment image 118. For instance, the number of rows may be dependent on molecular biology and alignment conditions associated with the genetic locus.

[0101] In some cases, one dimension of the fragment image 118 represents the genomic position (e.g., the “unambiguous placement” or “unambiguous sort order”) of various bases within the relevant fragment sequences. For instance, this dimension of the fragment image 118 may represent the genomic positions of the predetermined genomic region. In some cases, the columns of the fragment image 118 represent the location at which the bases within the fragment sequences originated. Since fragments in the ctDNA 108 and / or non-ctDNA 110 tend to have a length of 170 bases, the number of genomic positions represented in the fragment image 118 may be greater than 170. For instance, the fragment image 118 represents a genomic region having a length in a range of 200 to 1,000 bp.

[0102] In various cases, the fragment image 118 is a multi-channel image. The multiple channels of the fragment image 118, for instance, represent different characteristics of the bases of the fragment sequences. In some cases, one channel indicates the presence or absence of bases of various fragments at specified genomic positions. For instance, a channel of an example pixel, representing an example genomic position and an example fragment, may have a first value (e.g., a null value, zero, or the like) if the example fragment lacks a base at the example genomic position. In contrast, the channel of the example pixel may have a second value (e.g., a maximum value, 255, 128, 64, or the like) if the example fragment has a base at the example genomic position. In some examples, the second value (e.g., the maximum value) is less than a maximum, saturated value of the individual channel. For instance, it has been observed that limiting a maximum value of a red channel of an RGB image to 128 or 64, rather than 255, can improve visualization and performance.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0103] One or more additional channels of the fragment image 118, for instance, indicate one or more additional characteristics of the sequence read data 114 that is relevant to the type of lesion 104 in the subject 102. In some cases, an additional channel of the example pixel has a first value (e.g., a null value, zero, or the like) if the example fragment lacks a CG base pair (e.g., either a C or a G) at the example genomic position or has a second value (e.g., a maximum value, 255, 128, 64, or the like) if the example fragment has a CG base pair at the example genomic position. Thus, the example pixel may have a first color if the example fragment lacks a base at the example genomic position, a second color if the example fragment has a non-CG base pair at the example position, or a third color if the example fragment has a CG base pair at the example position.

[0104] In some cases, the image generator 116 creates the fragment image 118 based, at least in part, on the methylation data. According to some examples, a channel of the example pixel has a first value (e.g., a null value, zero, or the like) if the example fragment lacks a methylated CG base pair at the example genomic position or has a second value (e.g., a maximum value, 255, 128, 64, or the like) if the example fragment has a methylated CG base pair at the example genomic position. Thus, in some cases, the example pixel has one of at least one fourth color if the example fragment has a methylated CG base pair at the example position.

[0105] According to some examples, the image generator 116 creates the fragment image 118 based, at least in part, on comparing the sequence read data 114 to one or more reference sequences. In some cases, the image generator 116 identifies the presence of one or more variants in the sequence read data 114. For example, the image generator 116 may compare the fragment sequences indicated by the sequence read data 114 to a reference genome or pangenome and may identify the presence of variants in the fragment sequences by identifying bases in the fragment sequences that are different than the reference genome or pangenome. According to various implementations, at least one channel of the fragment image 118 indicates the presence of the one or more variants in the sequence read data 114. In some examples, a channel of the example pixel has a first value (e.g., a null value, zero, or the like) if the example fragment lacks a variant at the example genomic position or has a second value (e.g., a maximum value, 255, 128, 64, or the like) if the example fragment has the variant at the example genomic position. In some cases, the channel of the example pixel has the first value if the example fragment lacks a specific type of variant at the example genomic position (e.g., the example fragment has a base that matches a reference genome or pangenome at the example genomic position or has another type of variant at the example genomic position) or has the second value if the example fragment has the specific type of variant at the example genomic position. Specific variant types may include, for instance, SNVs, indels, rearrangements, and CNAs. For instance, the example pixel has one of at least one fifth color if the example fragment has a variant at the example genomic position.

[0106] In some cases, the image generator 116 indicates one or more additional characteristics of the sequence read data 114 in the fragment image 118. In some cases, at least one channel of the fragment image 118 indicates the presence of at least one predetermined sequence in the relevant cfDNA of the sequence read data 114. The predetermined sequence(s), for instance, may be indicative of whether the lesion 104 is cancerous and / or a type of cancer represented by the lesion 104. Examples of predetermined sequences, for instance, include promoters, enhancers, hotspots, CpG islands, exons, repeat elements (e.g., LINE1s), or genes. In specific diseases there may be specific genes-of-interest, noted by the fragment image 118. For instance, the fragment image 118 may indicate whether CRC tumor-suppressors (e.g., CDKN2A and APC) are hypomethylated, LINE1 elements are hypomethylated,FMI Docket No.0032-P L&H Docket No.: F171-0008PCT MLH1 promoters are hypermethylated (e.g., which can lead to mismatch repair deficiencies). In some cases, the fragment image 118 indicates CIMP molecular subtype, which based on hypermethylation of tumor-suppressor gene promoters. For instance, the fragment image 118 may indicate the presence and / or type of one or more of the following genes: CACNA1G, IGF2, NEUROG1, RUNX3, SOCS1, CRABP1, MLH1, p16, MINT1, MINT2, or MINT31.

[0107] In some instances, a channel of the example pixel has a first value (e.g., a null value, zero, or the like) if the example fragment lacks a base that is part of a predetermined sequence at the example genomic position or has a second value (e.g., a maximum value, 255, 128, 64, or the like) if the example fragment has a base that is part of the predetermined sequence at the example genomic position. Thus, the example pixel has one of at least one sixth color if the example fragment has a base that is part of the predetermined sequence at the example genomic position.

[0108] The fragment image 118 includes pixels with different colors depending on the characteristics of the sequence read data 114. The number of different colors represented by the pixels in the fragment image 118 depends on the number of characteristics indicated by the fragment image 118. For instance, if the fragment image 118 tracks a single characteristic (e.g., the presence of bases in the cfDNA at the various bases in a genomic region), then the fragment image 118 may be a binary image representing two colors (e.g., white and black). If the fragment image 118 tracks two characteristics (e.g., the presence of bases in the cfDNA at the various basis in the genomic region and the presence of CG dinucleotides), then the pixels of the fragment image 118 may be represented using two channels and three colors. If the fragment image 118 tracks three characteristics (e.g., the presence of bases in the cfDNA at the various bases in the genomic region, the presence of CG dinucleotides, and the presence of variants), then the pixels of the fragment image 118 may be represented using three channels and seven colors. In some cases, the complexity of the fragment image 118 is dependent on features relevant to a classification (e.g., a cancer type classification) using the fragment image 118.

[0109] According to some examples, the image generator 116 arranges the rows of the fragment image 118 in an unambiguous order based on the lengths and / or genomic positions of the cfDNA fragments indicated by the sequence read data 114. In some cases, the image generator 116 arranges the rows according to a start position, a middle position, or an end position of the relevant fragments. For instance, a first row of the fragment image 118 corresponds to a first fragment in the sample 106 and a second row of the fragment image 118 corresponds to a second fragment in the sample 106. In various cases, the first base of the first fragment corresponds to a higher position in a reference genome than the first base of the second fragment. That is, the start position of the first fragment is higher than the start position of the second fragment. In some cases, a first genomic position on a chromosome is “higher” or “earlier” than a second genomic position on a chromosome if the first genomic position is located closer to a telomere of a p arm of the chromosome than the second genomic position. In some examples, the final base of the first fragment corresponds to an higher position in the reference genome than the final base of the second fragment. For instance, the end position of the first fragment is higher than the end position of the second fragment. In some implementations, a base at a center of the first fragment is higher than a base at a center of the second fragment. If the first fragment has an even number of bases, the base at the center may be either one of bases in the middle of the first fragment. Similarly, if the second fragment has an even number of bases, the base at the center may be either one of the bases in the middle of the second fragment. Thus, the middle position of the first fragment is higher than the middle position of the second fragment. In various implementations, the first row is arranged above the second row in the fragmentFMI Docket No.0032-P L&H Docket No.: F171-0008PCT image 118, indicating the higher start position, middle position, or end position of the first fragment with respect to the second fragment. In some examples, the second row is arranged above the first row in the fragment image 118.

[0110] In some implementations, the image generator 116 adds additional rows to the fragment image 118. That is, some rows of the fragment image 118 may be upsampled. In some cases, rows are duplicated, such that multiple rows of the fragment image 118 reflect the same fragment sequence. In some examples, rows are interpolated, such that the fragment image 118 includes a row that is a combination of two other rows. Rows can be interpolated in at least one fashion, such as by combining multiple rows. According to some cases, the image generator 116 adds to the number of rows in the fragment image 118 in order to achieve a predetermined pixel dimension (e.g., a predetermined number of rows and columns) of the fragment image 118.

[0111] Although the fragment image 118 has been described as being a 2D image, implementations are not so limited. In some cases, the fragment image 118 can be a 3D image that includes a third dimension representing time. For instance, the fragment image 118 may represent fragments obtained from multiple samples obtained at different times. The changes in cfDNA states, for instance, may be relevant for predicting therapy responses and / or may be early indications of relapse. In some examples, the fragment image 118 can be an n-dimensional image, wherein n is a positive integer that is greater than one. In various implementations, the fragment image 118 is defined in a radial space.

[0112] Further, although a single fragment image 118 has been described with reference to FIG.1, implementations are not so limited. According to some implementations, multiple fragment images including the fragment image 118 are generated based on multiple regions of the genome of the subject 102. In some cases, the multiple fragment images correspond to genomic regions that are overlapping, non-overlapping, adjacent, non-adjacent, or any combination thereof.

[0113] The fragment image 118 may be input into a predictive model 120. In various cases, the predictive model 120 generates one or more disease indicators 122 based on the fragment image 118. The disease indicator(s) 122, for instance, include data that specifies one or more types of diseases of the subject 102. For example, the disease indicator(s) 122 may indicate one or more types of cancer cells present in the body of the subject 102. In some examples, the disease indicator(s) 122 indicate a cancer type of cells in the lesion 104 (e.g., a tumor type of the lesion 104). In some cases, the subject 102 includes multiple tumors including the lesion 104, such that the disease indicator(s) 122 specify one or more types of cancers represented by the multiple tumors. In some cases, the disease indicator(s) 122 provide a predicted effectiveness of a therapy in addressing the disease(s) of the subject 102. In some examples, the disease indicator(s) 122 indicate a progression of the disease(s) of the subject 102. For instance, the disease indicator(s) 122 may indicate a metastasis state of a cancer of the subject 102, or a stage (e.g., stage I, stage II, etc.) of the cancer of the subject 102.

[0114] In some cases, the predictive model 120 further analyzes additional biomarker data in order to generate the disease indicator(s) 122. For instance, the predictive model 120 may receive input data including the fragment image 118 as well as data indicating at least one of a genomic alteration, a mutational signature, an MSI status, a TMB, or a viral status of the subject 102 and / or lesion 104. The additional biomarker data may be generated based on the sample 106, medical images, or other samples obtained from the subject 102.

[0115] In various implementations, the predictive model 120 includes at least one trained ML model configured to output the disease indicators 122 in response to receiving the fragment image 118 in input data. For example,FMI Docket No.0032-P L&H Docket No.: F171-0008PCT parameters of the ML model(s) may have been previously optimized based on training data including the fragment image of regions in genomes of individuals within a population omitting the subject 102. For instance, the ML model(s) was trained using an unsupervised or semi-supervised learning technique, wherein the parameters were optimized to categorize (e.g., cluster) fragment images of the population. In some cases, the ML model(s) was trained using a supervised learning technique, wherein the training data further included ground truth categorizations of the expression of the sequence(s) of cancer cells of the individuals in the population, such that the parameters were optimized to minimize a loss between predicted disease indicators generated by the ML model(s) based on the fragment images of the population and the ground truth disease indicators of the individuals in the population. To increase training robustness, the population represented by the training data may include individuals without cancer, as well as individuals with a variety of cancer types and metastasis states. Various types of ML models can be included in the predictive model 120, such as a neural network (e.g., a convolutional neural network (CNN)), a nearest-neighbor model, a regression analysis model, a clustering model, a principal component analysis model, a gradient boosting model, a random forest, or any combination thereof.

[0116] In particular cases, the predictive model 120 includes a CNN. The CNN, for instance, may have multiple hidden layers. When the CNN receives an input (e.g., an input image), the layers may respectively perform operations to transform the input into an output (e.g., an output image). An individual hidden layer may be composed of neurons. An individual neuron within a particular layer may be disconnected from other neurons in the particular layer, but may be connected to multiple (e.g., all) of the neurons in the previous layer of the CNN. In some cases, the CNN includes at least one fully-connected layer that receives an output of the hidden layers and transforms the output into an output of the CNN.

[0117] In various cases, the CNN is a type of neural network that performs at least one convolution (or cross- correlation, dot product, or the like) operation on its input image. The CNN may generate an output image based on the convolved (or cross-correlated) input image. In various cases, each layer of the CNN is defined according to one or more parameters. The parameters, for instance, correspond to one or more filters. Each filter may include a digital image filter that, itself, can be represented as an image composed of multiple pixels. A filter of a CNN hidden layer may also be referred to as a “kernel.” Each filter, for instance, corresponds to a neuron of the layer. When a layer receives an input image (e.g., from a previous layer), a filter of the layer is convolved, cross-correlated, or otherwise used to transform the input image into an output image. In some cases, the subset of neurons in the previous layer may be defined according to a receptive field of the neuron in a subsequent layer, which may also correspond to the filter size of the neuron. Examples of CNNs include, for instance, residual neural networks (e.g., ResNet, see He et al., arXiv:1512.03385v1, 2015), U-Net (see, e.g., Ronneberger, et al., arXiv:1505.04597v1, 2015), and the like.

[0118] In some examples, the kernel and / or receptive field of at least one CNN hidden layer have various shapes. For instance, in some cases, the kernel and / or receptive field are defined as 2D squares (e.g., 2x2 pixels, 3x3 pixels, 4x4 pixels, 5x5 pixels, or the like). In some cases, the kernel and / or receptive field are defined as 2D rectangles. For instance, a dimension of the kernel that corresponds to the lengths of the cfDNA fragments (e.g., a horizontal dimension) may be longer than a dimension of the kernel that corresponds to the genomic positions (e.g., a vertical dimension).

[0119] According to various implementations, the parameters of the CNN in the predictive model 120 are optimized to enable the predictive model 120 to infer whether the subject 102 has cancer based on the fragment image 118. InFMI Docket No.0032-P L&H Docket No.: F171-0008PCT some cases, the optimized parameters of the CNN are used to identify the type of cancer cells present in the lesion 104 and / or another portion of the body of the subject 102. Visual characteristics of the fragment image 118, in various cases, are correlated to cancer type. For instance, certain methylation patterns, variants, fragment lengths, and other characteristics apparent in the fragment image 118 are associated with the presence of cancer and / or cancer type. In various cases, these characteristics are subtle and would be impossible for a clinician (e.g., the care provider 105) to identify by manually reviewing the fragment image 118. However, in various implementations of the present disclosure, the predictive model 120 may identify cancer-relevant characteristics of the fragment image 118, such that the predictive model 120 can accurately categorize a potential cancer of the subject 102. For instance, if the lesion 104 is a tumor, the predictive model 120 may indicate the type of cancer cells in the lesion 104.

[0120] In some examples, the predictive model 120 includes another type of DL model. For example, the predictive model 120 may include a multiple-instance learning (MIL) model (also referred to as a “multi-instance learning model) that is trained to perform anomaly (e.g., outlier) detection on one or more characteristics of the fragment image 118. For example, the MIL model may include one or more autoencoders.

[0121] Cancer type, in various cases, can be defined in any of various ways. In some cases, the predictive model 120 determines a tissue-of-origin (also referred to as “tissue origin”) of the cancer cells in the body of the subject 102. For instance, the predictive model 120 may determine, based on the fragment image 118, a type of differentiated tissue type from which cancer cells in the body of the subject 102 began to divide uncontrollably. In some cases, the predictive model 120 determines, based on the fragment image 118, whether the tissue origin is an epithelial tissue of the subject 102. For instance, the predictive model 120 determines whether the lesion 104 is a carcinoma. In particular cases, the predictive model 120 identifies the tissue origin at a further level of granularity, such as whether the tissue origin of the lesion 104 or other cancer cells in the body of the subject 102 include squamous cells (e.g., squamous cell carcinoma), glandular cells, adenomatous cells (e.g., adenocarcinoma), transitional cells (e.g., transitional cell carcinoma), or basal cells (e.g., basal cell carcinoma). In various examples, the predictive model 120 determines whether the tissue origin is a connective tissue of the subject 102 (e.g., whether the cancer is a type of sarcoma), such as osteocytes (e.g., bone sarcoma), chondroblasts (e.g., chondrosarcoma), or muscle cells (e.g., rhabdomyosarcoma, leiomyosarcoma, etc.). In various examples, the predictive model 120 determines, based on the fragment image 118, whether the tissue origin includes a glial cell (e.g., glioma). In some cases, the predictive model 120 identifies, based on the fragment image 118, whether the tissue origin is a blood cell of the subject 102 (e.g., whether the cancer is a type of leukemia). In various cases, the predictive model 120 determines whether the tissue origin includes lymphocytes (e.g., lymphoma) or plasma cells (e.g., myeloma). According to various examples, the predictive model 120 determines, based on the fragment image 118, whether the tissue origin includes multiple cell types, such as whether the subject 102 has adenosquamous carcinoma, carcinosarcoma, teratocarcinoma, or the like. The disease indicator(s) 122, for instance, indicate a tissue origin of cancer of the subject 102.

[0122] In some cases, the tissue-of-origin is indicative of a histological tissue type (also referred to as a “histological cancer type”) of the lesion 104. For instance, the predictive model 120 may determine, based on the fragment image 118, whether the lesion 104 has one or more histological characteristics that are indicative of cancer type. Examples of these characteristics include relative nucleus size, regularity of shape, regularity of size, cytoplasm color, prominence of nucleoli, and the like.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0123] In some examples, the predictive model 120 determines, based on the fragment image 118, a site of a primary tumor of the subject 102. In various implementations, the primary site may be an organ or anatomical site in which the first tumor developed within the body of the subject 102. For example, the primary site may include an adrenal gland, a bladder, blood, a bone, brain, a breast, a cervix, a colon, a rectum, an ear, a nose, a throat, endometrial tissue, an esophagus, a gastrointestinal tract, head, neck, intestine, a kidney, a larynx, bone marrow, liver, a lymph node, a lung, a nasopharynx, a mouth, an ovary, pancreas, pharynx, prostate, rectum, skin, stomach, testicle, thyroid, uterus, vasculature, or the like. The predictive model 120, for instance, may identify the primary site of the cancer of the subject 102 based on the fragment image 118. The disease indicator(s) 122, for instance, indicate a primary site of cancer of the subject 102. In some cases, the subject 102 has one or more secondary tumors in addition to the primary tumor.

[0124] In various implementations, the predictive model 120 determines a predicted subtype of the cancer cells of the subject 102. For instance, the subtype of the cancer cells is indicative of one or more characteristics of the cancer cells, such as a physical or morphological characteristic of the cells (e.g., a shape), a physical or morphological characteristic of at least one portion of the cells (e.g., relative size of the nucleus of a cell), the presence of a substance or structure in the cell, the presence of a substance or structure on the cell (e.g., the presence of a receptor on the cell), expression of the cells, epigenetic features of the cells (e.g., whether a particular promoter is highly methylated), a division rate of the cells, or the like. In various cases, the subtype of the cancer cells of the subject 102 are relevant for diagnosing and treating the cancer of the subject 102. For example, the predictive model 120 may determine, based on the fragment image 118, whether the cancer cells are positive for a particular receptor (e.g., HER2), negative for the particular receptor, or a mixture of the two (e.g., 40% positive for the particular receptor). The disease indicator(s) 122, for instance, indicate one or more cancer subtypes of the subject 102.

[0125] In various cases, the predictive model 120 identifies at least one type of cancer of the subject 102. For instance, the disease indicator(s) 122 may include at least one likelihood (e.g., probability) that the subject 102 has at least one predetermined type of cancer, an indication that the likelihood that the subject 102 has a predetermined type of cancer exceeds a threshold (e.g., 90% certainty, 95% certainty, 99% certainty, or the like), or some other indication of the determined cancer type. In some cases, the predictive model 120 is a binary classifier, such that the disease indicator(s) 122 include a Boolean (e.g., true or false) indication of a predetermine cancer type. In some examples, the disease indicator(s) 122 includes predictions of whether the subject 102 has one or more types of cancers. In some instances, the disease indicator(s) 122 may include a vector with multiple elements respectively corresponding to a likelihood or prediction that the subject 102 has multiple types of cancers. For example, the disease indicator(s) 122 may be

[0010] if the predictive model 120 predicts that the subject 102 does not have a first type of cancer or a second type of cancer, but does have a second type of cancer. In other examples, the disease indicator(s) 122 may be [0.010.050.95] if the predictive model 120 predicts that the subject 102 has a 1% likelihood of having a first type of cancer, a 5% likelihood of having a second type of cancer, and a 95% likelihood of having a third type of cancer.

[0126] The cancer type of the subject 102 may be indicative of the efficacy of various treatments. The disease indicator(s) 122 may, in some cases, indicate whether the cancer of the subject 102 is resistant or responsive to one or more predetermined therapies. In various cases, the fragment image 118 is indicative of whether the cancer cells are resistant (e.g., at least partially unharmed) if a particular therapy (e.g., an anticancer agent) is administered, or whether the cancer cells are responsive (e.g., at least partially killed or otherwise destroyed) if a particular therapy isFMI Docket No.0032-P L&H Docket No.: F171-0008PCT administered. In various implementations, the predictive model 120 determines whether each of one or more therapies is likely to successfully treat the cancer of the subject 102.

[0127] According to some cases, the predictive model 120 is configured to determine whether the subject 102 qualifies for a study, such as a clinical trial. For example, the predictive model 120 may determine that the subject 102 has a predetermined type of cancer and may therefore qualify for enrollment in a clinical trial to investigate the efficacy of a new therapy (e.g., a new immunotherapy) that targets the predetermined type of cancer. The disease indicator(s) 122, for instance, indicate whether the subject 102 qualifies for the clinical trial.

[0128] In some implementations, the predictive model 120 is unable to conclusively categorize the cancer cells of the subject 102. For example, the predictive model 120 may determine that, based on the fragment image 118, the certainty that the subject 102 has a predetermined type of cancer is below a threshold certainty. In various cases, the disease indicator(s) 122 may indicate that a cancer type of the subject 102 is inconclusive.

[0129] A report generator 124 is configured to generate a report 126 based on the disease indicator(s) 122. The report 126, for example, includes consumable data that can inform the care provider 105 about the at least one determined category of the cancer of the subject 102. Further, in some cases, the report 126 indicates whether the lesion 104 of the subject 102 is cancerous by reporting whether the ctDNA 108 has been identified in the sample 106. In various implementations, the report 126 may indicate the results of additional analyses, such as the results of a histological study, whole transcriptome sequencing, cell-free RNA (cfRNA) sequencing, whole exome sequencing, whole genome sequencing, a cancer (e.g., DNA) hotspot panel test, a DNA methylation test, a tumor mutational burden (TMB) test, a DNA fragmentation test, an RNA fragmentation test, a microsatellite instability (MSI) test, a tumor mutational burden (TMB) test, or a viral status test. The performance of such tests is within the ordinary skill of the art, with additional detail provided elsewhere herein. The report 126, for example, may include a genomic profile of the subject 102 based on various combinations of the above analyses and tests.

[0130] In some implementations, the report 126 indicates that a follow-up test of the subject 102 is indicated. For instance, in response to determining that the categorization of the cancer is inconclusive, the report generator 124 may generate the report 126 to indicate that one or more additional tests (e.g., a histological study, genome sequencing, exome sequencing, additional DNA sequencing, RNA sequencing, transcriptome sequencing, etc.) should be performed in order to identify whether the cancer cells of the subject 102 express the sequence(s).

[0131] In various cases, the report 126 is output to a clinical device 128. For example, the report generator 124 transmits the report 126 to the clinical device 128. In various implementations, the clinical device 128 is a computing device that is operated by, owned by, or otherwise associated with the care provider 105. For instance, the clinical device 128 may be a desktop computer, a laptop computer, a smart phone, or some other computing device associated with the care provider 105. The clinical device 128, in various cases, outputs the report 126 to the care provider 105. In some cases, the clinical device 128 includes a display (e.g., a screen) that visually presents the report 126. In various cases, the clinical device 128 includes a speaker that outputs a sound indicative of the report 126. The clinical device 128, in various cases, may output the information in the report 126 using one or more output mechanisms or devices.

[0132] The care provider 105 may review the report 126 by interacting with the clinical device 128. The report 126, in various cases, may enhance the clinical decision-making of the care provider 105. For instance, the care provider 105 may prepare and / or administer a therapy (e.g., a therapeutic agent, such as an anticancer agent) to the subjectFMI Docket No.0032-P L&H Docket No.: F171-0008PCT 102 based on the report 126. According to various implementations, the care provider 105 may initiate the therapy and / or refer the subject 102 to another care provider to receive the therapy.

[0133] In various implementations, the care provider 105 may develop a diagnosis and / or prognosis of the subject 102 based on the report 126. In various implementations, the care provider 105 may communicate information in the report 126 to the subject 102.

[0134] FIG.1 illustrates various elements that can be embodied in one or more computing devices. For example, at least a portion of the functions of the sequencer 112, the image generator 116, the predictive model 120, the report generator 124, the clinical device 128, or any combination thereof, are performed by one or more processors in at least one computing device. Examples of computing devices include server computers, desktop computers, laptop computers, tablet computers, mobile phones, wearable devices, Internet of Things (IoT) devices, and the like. In various cases, instructions for performing at least a portion of the functions of these elements are stored in memory and / or in a non-transitory computer readable medium. The instructions, for instance, are executed by the processor(s).

[0135] FIG.1 also illustrates various types of data. For example, the sequence read data 114, the fragment image 118, the disease indicator(s) 122, the report 126, or any combination thereof, includes data. The various types of data illustrated in FIG.1 may be stored, such as in memory or in non-transitory computer readable media. In various implementations, at least a portion of the data is transmitted or otherwise output by one or more computing devices. For example, a computing device may transmit one or more communication signals to another computing device, wherein the communication signal(s) encode at least a portion of the data. Examples of communication signals include electromagnetic signals, optical signals, ultrasonic signals, optical signals, and electrical signals. For example, communication signals can be transmitted wirelessly and / or in a wired fashion. The communication signals, for instance, are transmitted over one or more wireless channels and / or one or more wired channels (e.g., optical cabling, electrical cabling, etc.). In various cases, the communication signal(s) are transmitted over one or more communication networks. A communication network, for instance, may be defined according to one or more physical channels, such as one or more frequency spectra. In some cases, a communication network is defined according to one or more communication protocols and / or standards. Examples of communication networks include fiber optic networks, Institute of Electrical and Electronics Engineers (IEEE) networks (e.g., WI-FI™ networks, WiMAX networks, BLUETOOTH™ networks, etc.), cellular networks (e.g., a 3rdGeneration Partnership Project (3GPP) radio network, such as a Long Term Evolution (LTE) network, a New Radio (NR) network; or a cellular core network such as a 3rdGeneration (3G) core, a 4thGeneration (4G) core, a 5thGeneration (5G) core, etc.), ultrasonic networks, and the like. In some cases, the data is broadcasted from one device to multiple other devices. In some cases, the data is unicasted from one device to another device. For instance, various forms of data described herein may be transmitted via a peer-to-peer (P2P) connection.

[0136] FIG.2 illustrates an example environment 200 illustrating ctDNA 202, which can be utilized to analyze cancer cells of a subject. For instance, the ctDNA 202 may be the ctDNA 108 described above with reference to FIG.1.

[0137] In various implementations, a cancer cell 204 within the subject includes genomic DNA (gDNA) that is expressed by the cancer cell 204. For example, the gDNA 206 may include various sequences, such as a gene 208, a promoter 210, an enhancer 212, and a variant 214. For example, the variant 214 is part of the gene 208. In addition, various epigenetic factors impact expression of the gene 208 as well as other genes within the gDNA 206. ForFMI Docket No.0032-P L&H Docket No.: F171-0008PCT example, the gDNA 206 may be packaged within the nucleus of the cancer cell 204 with various histones 216. When the gene 208 is expressed, a portion of the gDNA 206 including the gene 208, the promotor 210, the enhancer 212, and the variant 214 may be exposed to proteins within the nucleus, such as RNA transcriptase. In various cases, the portion of the gDNA 206 is unwrapped or otherwise unpackaged from the histones 216. Thus, the expression of the gene 208 (e.g., the amount of mRNA generated by RNA transcriptase based on the gene 208 within the cancer cell 204) is linked to the frequency or time at which the portion of the gDNA 206 is exposed.

[0138] The cancer cell 204, for example, may die. The contents of the cancer cell 204, including the gDNA 206, may be released. In various cases, the gDNA 206 is released into blood 218 that flows through a blood vessel 220 of the subject. When the gDNA 206 is released from the nucleus of the cancer cell 204, the gDNA 206 is degraded due to various biophysical and / or biochemical factors. For example, the blood 218 may include various enzymes that cut the gDNA 206 into the ctDNA 202. In various cases, other mechanical, chemical, or thermal conditions in the blood 218 divide the gDNA 206 into the ctDNA 202. For example, these conditions divide the gDNA 206 into fragments at various breakpoints 222.

[0139] Notably, the presence and location of the histones 216 may impact the sequences of the ctDNA 202 that are observed in the blood 218. The breakpoints 222, for example, are more likely to occur at edges of a sequence of the gDNA 206 that is exposed by the histones 216. Therefore, the sequence of the ctDNA 202, in various examples, is indicative of the expression of mRNA and other functional RNA in the cancer cell 204. By reviewing the ctDNA 202, the expression of the cancer cell 204 can be determined without performing RNA sequencing (e.g., whole transcriptome sequencing), in some cases.

[0140] However, the methylation status of various regions within the ctDNA 202 are also indicative of the expression of the cancer cell 204. For example, even though the promoter 210 is operatively coupled to the gene 208 (e.g., the promoter 210 includes a transcription start site of the gene 208), and the promoter 210 is present in the ctDNA 202, certain epigenetic factors of the promoter 210 may indicate that the gene 208 is not expressed by the cancer cell 204. For instance, the promoter 210 may include various cytosines that are methylated. The methylated cytosines, in various cases, may have prevented RNA polymerase from binding to the promoter 210 when the promoter 210 was part of the gDNA 206, thereby preventing expression of the gene 208 in the cancer cell 204. In these cases, an anticancer therapy that targets cells expressing the gene 208 may be ineffective at targeting the cancer cell 204. Thus, in various implementations of the present disclosure, the methylation status of various regions (including, e.g., the gene 208, the promoter 210, the enhancer 212, the variant 214, etc.) in the ctDNA 202 may be determined.

[0141] In various implementations, the ctDNA 202 is obtained from a sample of plasma 232 in the blood 218 of the subject. The plasma 232, for example, includes various DNA fragments 234 including the ctDNA 202. In some cases, the DNA fragments 234 include various cfDNA, such as cfDNA released from non-cancerous cells (non-ctDNA, for instance).

[0142] By sequencing the ctDNA 202, expression of the cancer cell 204 can be inferred. Expression of the cancer cell 204 is associated with the type of the cancer cell 204. Further, using methylation sequencing, the methylation status of the ctDNA 202 can be determined. The genetic sequences of the DNA fragments 234, as well as the methylation statuses, are relevant to a categorization of the cancer cell 204 into one or more types. A predictive model, for instance, can classify the cancer cell 204 by analyzing a fragment image representing the sequences and methylation statuses of the DNA fragments 234.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0143] FIG.3 illustrates an example fragment image 300 that is generated based on fragments 302 obtained from a sample of a subject. In various cases, m unique fragments 302 are sequenced from the sample. For example, the fragments 302 include fragment 1 to fragment m. In this example, m is a positive integer that is greater than 1. According to some cases, the sample is a liquid biopsy sample. The fragments 302 may be released from cancerous and / or non-cancerous cells within the body of the subject.

[0144] In various cases, the fragments 302 are representative of a genomic region 304 of the subject. The genomic region 304, for instance, is defined according to a locus of a chromosome 306 of the subject. The genomic region 304 includes n positions within the chromosome, spanning from a position 1 to a position n. In this example, n is a positive integer that is greater than 1. In particular examples, n is in a range of 200 to 1,000. In some cases, n is within 50% to 200% of m.

[0145] The fragment image 300 is a 2D image. In various implementations, the fragment image 300 includes m fragment rows 308 corresponding to the m fragments 302 sequenced from the sample. That is, each row of the fragment image corresponds to a unique sequence obtained by sequencing the fragments 302 of the sample. The fragments 302, in various cases, include ctDNA and / or non-ctDNA. In some cases, the fragment image 300 includes additional rows representing duplicated and / or interpolated rows among the m fragment rows 308.

[0146] In addition, the fragment image 300 includes n position columns 310 corresponding to the n positions along the genomic region 304 of the chromosome 306. Accordingly, each column among the position columns 310 corresponds to a particular base pair position along the genomic region 304 of the chromosome 306.

[0147] The fragment image 300 itself includes m by n pixels. In some cases, m and n are independent of one another. In some cases, m is related to the local sequencing depth of the fragments 302. In some cases, n is limited to a predefined number in order to optimize a downstream analysis of the fragment image 300. For instance, m and / or n may be set based on specific genomic and / or epigenetic features of interest.

[0148] The pixels, for instance, range from a1,1 to an,m. The fragment image 300 is defined according to l channels 312, wherein l is a positive integer. Each pixel is defined according to l channels 312. In some cases, the channels 312 are color channels, but implementations are not so limited. For example, the fragment image 300 may be an RGB image and the l channels 312 include red, green, and blue channels.

[0149] In various implementations, each of the l channels 312 corresponds to a characteristic of the fragments 302. In some cases, one of the l channels 312 indicates the presence or absence of bases in the fragments 302 at corresponding positions in the genomic region 304. In some examples, one of the l channels 312 indicates the presence or absence of a CG bases in the fragments 302 at corresponding positions in the genomic region 304. In some cases, one of the l channels 312 indicates the presence or absence of methylated bases in the fragments 302 at corresponding positions in the genomic region 304. In some cases, one of the l channels 312 indicates the presence or absence of variants (e.g., point mutations) in the fragments 302 at corresponding positions in the genomic region 304. In some examples, one of the l channels 312 indicates the presence of one or more predetermined sequences in the fragments 302 at corresponding positions in the genomic region 304. For example, one of the l channels 312 indicates the presence of a promoter, an enhancer, a hotspot, a CpG island, an exon, or a gene at the corresponding positions. In some examples, the characteristics indicated by the l channels 312 of the fragment image 300 are relevant to a presence and / or type of cancer in the body of the subject.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0150] In various cases, the values of the l channels 312 of a pixel in the fragment image 300 reflect the position and fragment corresponding to that pixel. In some cases, possible values of a given channel in a given pixel can include a minimum or null value (e.g., 0) indicating the absence of the characteristic corresponding to the channel. In some examples, possible values of a given channel in a given pixel can include a maximum value (e.g., 255) indicating the presence of the characteristic corresponding to the channel. Thus, each of the channels 312 may be defined by Boolean values indicating the presence or absence of a characteristic.

[0151] Due to the values of the channels 312, the colors of various pixels within the fragment image 300, for instance, may indicate one or more types of cancer cells within the body of the patient. Furthermore, by translating the sequences of the fragments 302 into the fragment image 300, an image-based predictive model can be used to analyze the sequences of the fragments 302. In some cases, the fragment image 300 is input into one or more CNNs configured to identify one or more types of cancer indicated by the fragments 302. According to some examples, the CNN(s) are able to parse the types of cancers more accurately than other types of predictive models.

[0152] FIG. 4 illustrates an example environment for training and utilizing a predictive model to using fragment images. In some cases, the predictive model is used to determine the methylation status of various cells of interest. The methylation status may be used to detect diseases (e.g., cancer types and subtypes), select therapies, and to monitor disease progression.

[0153] The predictive model 402, for instance, is the predictive model 120 described above with reference to FIG.1. In various implementations, the predictive model 402 includes one or more ML models 404. A trainer 406, for instance, is configured to optimize various parameters 408 of the ML model(s) 404 based on training data 410.

[0154] The training data 410 includes example fragment images 412 and example disease indicators 414. The example fragment images 412, in various cases, are obtained based on DNA fragments (e.g., cfDNA) of individuals within a population 416. For instance, the fragment images 412 are indicative of one or more regions of interest within the DNA fragments. The example disease indicators 414 may indicate diseases (e.g., cancer type or subtype) of the individuals in the population 416, a progression or state of the diseases of the individuals in the population 416, the effectiveness of one or more therapies in treating the diseases of the individuals in the population 416, or any combination thereof of the individuals within the population 416. For example, the example disease indicators 414 may be generated based on samples obtained from the individual that are not limited to DNA fragments. In some cases, the example disease indicators 414 are obtained by performing whole genome sequencing, whole exome sequencing, RNA sequencing, immunohistochemical studies, post-immunotherapy treatment analyses, or other types of analyses. In various cases, the population 416 includes individuals with different types of cancers, different types of cancer severities, and the like. In some cases, the population 416 includes individuals that are not expressing any type of known cancer.

[0155] The ML model(s) 404 include one or more model types. For instance, the ML model(s) 404 include an artificial neural network. An artificial neural network includes various layers that respectively process input data. For example, an artificial neural network includes an input layer, one or more hidden layers, and an output layer. The input layer performs a pre-processing operation on the input data. The hidden layer(s) may perform various processing operations on the output from the input layer. The output layer, in various cases, processes the output from the hidden layer(s). Each layer, in some cases, includes one or more nodes, which are defined by individual operations. In various cases,FMI Docket No.0032-P L&H Docket No.: F171-0008PCT the hidden layer(s) include nodes that are connected to each other in parallel and / or series. Examples of artificial neural networks include feedforward neural networks, multi-layer perceptrons (MLPs), convolutional neural networks (CNNs), and backpropagation models. In various implementations, the operations performed by the layers and / or nodes within an artificial neural network included in the ML model(s) 404 is defined according to the parameters 408. For example, the parameters 408 may include weights, thresholds, filters, kernels, or other data objects that are utilized to perform operations of the ML model(s) 404.

[0156] In particular cases, the ML model(s) 404 include one or more CNNs. Each hidden layer of an example CNN in the ML model(s) 404 includes a convolutional layer including at least one neuron. According to various cases, each neuron is defined by an image filter (also referred to as a “kernel”) that is convolved and / or cross-correlated with an input image to the hidden layer, thereby producing an output image of the hidden layer. The image filter, in various cases, includes an image having pixels. Values of the pixels of the image filter, according to some examples, are defined by the parameters 408. In some cases, individual hidden layers include other components. For example, the hidden layers further apply activation functions that introduce nonlinearity into the CNN. Examples of activation functions include a linear activation function, a rectified linear unit (ReLU) function, a Heaviside activation function, a logistic activation function, a Gaussian error linear unit (GELU), a softplus function, or the like. In some cases, the CNN includes one or more pooling layers, such as a max pooling layer.

[0157] CNNs may include hidden layers arranged in series, in parallel, or a combination thereof. For instance, in some cases, a CNN includes an encoder and a decoder. An encoder, in various cases, includes one or more hidden layers configured to reduce one or more dimensions of an input image. In some examples, the encoder is configured to extract relevant features from the input image. In various cases, a decoder includes one or more hidden layers configured to increase one or more dimensions of an input image (e.g., an output image of an encoder). For example, a decoder can be configured to build a segmentation mask based on features extracted by an encoder. U-Net is an example of a CNN architecture that includes an encoder and a decoder.

[0158] In some implementations, the ML model(s) 404 include a nearest-neighbor model. One example of a nearest- neighbor model includes a k-nearest neighbor model. For example, a nearest-neighbor model defines various “neighbors,” which are points within a feature space, with associated class labels. When a new data point is mapped to the feature space, the new data point is classified based on the proximity (e.g., Euclidian distance, Manhattan distance, Minkowski distance, etc.) of its “neighbors” to the new data point as well as their associated classes. In some cases, the new data point is classified as belonging to a particular class if greater than a threshold number of neighbors within a threshold distance of the new data point are members of the class. For instance, the parameters 408 may include k (e.g., the number of neighbors compared to the new data point), the threshold distance, and so on.

[0159] In various cases, the ML model(s) 404 include a regression analysis model. The regression analysis model, for example, is defined by a regression function that defines relationships between one or more independent variables and one or more dependent variables. The regression function may further define one or more unknown parameters that define a relationship between the independent and dependent variables. In various implementations, the unknown parameters and / or the type of regression function (e.g., linear, quadratic, etc.), is defined according to the parameters 408.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0160] In some cases, the ML model(s) 404 include a clustering model. In various cases, a clustering model maps various data points (e.g., training data) to a feature space. Based on the proximity of groups of those data points in the features pace, one or more “clusters” are defined. An additional data point may be classified according to one or more of the clusters based on its proximity to the clusters (e.g., a center of the clusters, a boundary of the cluster, etc.). Examples of clustering models include k-means clustering, mean-shift clustering, expectation-maximization (EM) clustering, and agglomerative hierarchical clustering. The parameter(s) 408, for example, include a threshold proximity within which a new data point is classified within a cluster, a density of points used to define a cluster, and the like.

[0161] In some instances, the ML model(s) 404 include an autoencoder model. An autoencoder, for instance, includes an encoder and a decoder. The encoder maps input data to a code, and the decoder generates output data based on the code. Each of the encoder and the decoder may include one or more layers. Various parameters of the layers in the autoencoder model may be defined by the parameter(s) 408. In some cases, a multiple-instance learning (MIL) model may be included in the ML model(s) 404. The autoencoder and / or MIL model may be used to perform outlier detection.

[0162] In various examples, the ML model(s) 404 include a principal component analysis model. In various implementations, a principal component analysis defines a collection principal components of unit vectors within a coordinate space based on a data set (e.g., training data). The model, for example, is an orthogonal linear transformation of the data set. Various weights of the model, for example, are included in the parameter(s) 408.

[0163] The ML model(s) 404, in some implementations, includes a gradient boosting model. For example, the gradient boosting model is defined as a collection of prediction models (e.g., decision trees) that iteratively classify observed data. In various cases, the type of prediction model, weights in the prediction models, and the like, are defined by the parameter(s) 408.

[0164] The ML model(s) 404, for example, includes a random forest. The random forest, for instance, includes multiple decision trees that classify data in an ensemble fashion. In various implementations, the decision trees are defined by the parameter(s) 408.

[0165] In various implementations of the present disclosure, the trainer 406 is configured to optimize the parameters 408 based on the training data 410. For example, the trainer 406 may input first example disease indicators 414 (corresponding to a first individual among the population 416) among the example fragment images 412 into the predictive model 402, and may receive a predicted category. The trainer 406 may compute a loss (e.g., determine a discrepancy) between a first example expression indicator (corresponding to the first individual) among the example disease indicators 414 and the predicted category. Further, the trainer 406 may alter the parameters 408 in order to minimize the loss. In various cases, the trainer 406 optimizes the parameters 408 iteratively based on the entire set of the training data 410.

[0166] In various implementations, the optimization of the parameters 408 enables the predictive model 402 to identify predictive attributes of the example fragment images 412 that are correlated to or otherwise associated with the example disease indicators 414. For instance, the predictive attributes may correspond to one or more characteristics that are preferentially found on samples from individuals with predetermined diseases (e.g., cancers) and are preferentially not observed on samples from individuals without the predetermined diseases. Examples of predictive attributes include, for instance, aberrant hypomethylation patterns in fragments that are otherwise mappingFMI Docket No.0032-P L&H Docket No.: F171-0008PCT to a region that is ordinarily hypermethylated in subjects without the disease. In some cases, predictive attributes correspond to combinations of indications, such as a combination of methylation states, fragment lengths, and fragment endpoint locations.

[0167] Once the parameters 408 are optimized, the predictive model 402 may be ready to classify a new set of data. For example, the predictive model 402 may receive input data including a fragment image 418 of a subject. The fragment image 418, for instance, may include one or more of the predictive attributes. The predictive model 402 may perform various operations on the input data based on the trained ML model(s) 404 and the optimized parameters 408. In various cases, the predictive model 402 outputs output data including one or more disease indicators 420 based on the fragment image 418. The disease indicator(s) 420, for instance, indicate a disease (e.g., cancer type or subtype) of the subject, a progression or state of the disease, whether a particular therapy is predicted to be effective in treating the disease of the subject, or any combination thereof.

[0168] Although FIG.4 is primarily described as referring to supervised learning, implementations are not so limited. In various cases, the training data 410 omits the example disease indicators 414 and the trainer 406 is configured to optimize the parameters 408 using the example fragment images 412 and an unsupervised learning technique.

[0169] FIG.5 illustrates an example of training data 500 utilized to train one or more ML models. For example, the training data 500 may be the training data 410 described above with reference to FIG.4.

[0170] The training data 500, in various cases, may represent m samples, wherein m is a positive integer. In some cases, the m samples are respectively obtained from m individuals within a population, although implementations are not so limited. For example, in some cases, multiple samples may be obtained from the same individual at different times.

[0171] The training data 500 includes first to mth example fragment images 502-1 to 502-m. For example, the first to mth example fragment images 502-1 to 502-m include fragment images of one or more regions in cfDNA (e.g., ctDNA) obtained from the respective m samples.

[0172] The training data 500 may further include first to mth example disease indicators 504-1 to 504-m. The first to mth example disease indicators 504-1 to 504-m, for instance, include indications of whether cancer cells represented by the m samples are classified into one or more predetermined cancer types.

[0173] FIG.6 illustrates an example report 600 summarizing predicted categories of a cancer of a subject. In various cases, the report 600 is the report 126 described above with reference to FIG.1. The report 600, for instance, may be displayed to a patient and / or care provider. In some cases, the report 600 is generated based on a fragment image of one or more regions in a sample (e.g., a liquid biopsy sample) obtained from the subject.

[0174] The report 600 includes a disease indicator 602 that indicates a classification of at least one disease (e.g., cancer). In some cases, the disease indicator 602 indicates whether cancer cells of the subject are within one or more predetermined types of cancer.

[0175] In various cases, the report 600 includes one or more therapy indicators 608. For instance, the therapy indicator(s) 608 convey whether the cancer is predicted to be resistant to one or more predetermined therapies and / or whether the cancer is predicted to be responsive to one or more predetermined therapies. In some examples, the therapy indicator(s) 608 include a suggested treatment decision, which may specify at least one therapy in which the disease is predicted to be responsive to. In some cases, the therapy indicator(s) 608 convey a specific dosage of one or more predetermined therapies that is predicted to treat the disease.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0176] In some examples, the report 600 includes one or more prognostic indicators 610. The prognostic indicator(s) 610, for instance, indicate a prognosis of the subject. For example, the prognostic indicator(s) 610 may indicate a survivability, a recoverability, a quality-of-life indicator, or other information indicative of the prognosis of the subject.

[0177] The report 600 may include a trial qualification 612 of the subject. The trial qualification 612, for instance, indicates whether the subject is predicted to be eligible or otherwise qualify for a predetermined clinical trial.

[0178] The report 600, in various implementations, includes a metastasis profile 614 of the subject. The metastasis profile 614, for instance, indicates a likelihood (e.g., probability) that the cancer will metastasize (e.g., at a particular point in time), one or more tissues in which the cancer is predicted to metastasize, or the like.

[0179] In various cases, the report 600 includes recommended follow-up tests 616. For example, the report 600 may include a recommendation to perform whole genome sequencing on the subject, particularly in cases if the cancer cannot be categorized above a threshold certainty.

[0180] The report 600 may include a genomic profile 618 of the subject. In various cases, the genomic profile 618 includes or is generated based on the results of non-methylation analyses of the subject.

[0181] FIG.7 illustrates an example process 700 for predicting a type of cancer cell by converting fragmentomic data into an image. In various instances, the process 700 is performed by an entity including at least one processor, at least one computing device (e.g., one or more server computers), a sequencer (e.g., the sequencer 112), an image generator (e.g., the image generator 116), a predictive model (e.g., the predictive model 120), a report generator (e.g., the report generator 124), a clinical device (e.g., the clinical device 128), or any combination thereof.

[0182] At 702, the entity generates a fragment image based on data indicative of DNA fragments in a sample obtained from a subject. According to various cases, the subject is a human. According to some cases, the subject has at least one type of disease. For instance, the subject may have adrenal cancer, bladder cancer, blood cancer, bone cancer, brain cancer, breast cancer, carcinoma, cervical cancer, colon cancer, colorectal cancer, corpus uterine cancer, ear, nose and throat (ENT) cancer, endometrial cancer, esophageal cancer, gastrointestinal cancer, head and neck cancer, Hodgkin's disease, intestinal cancer, kidney cancer, larynx cancer, leukemia, liver cancer, lymph node cancer, lymphoma, lung cancer, melanoma, mesothelioma, myeloma, nasopharynx cancer, a neuroblastoma, non- Hodgkin's lymphoma, oral cancer, ovarian cancer, pancreatic cancer, penile cancer, pharynx cancer, prostate cancer, rectal cancer, sarcoma, seminoma, skin cancer, stomach cancer, a teratoma, testicular cancer, thyroid cancer, uterine cancer, vaginal cancer, a vascular tumor, or combinations or metastases thereof. In some examples, the sample is a liquid biopsy sample. For instance, the sample may include blood, plasma, cerebrospinal fluid, sputum, stool, urine, lymphatic fluid, saliva, or any combination thereof.

[0183] The data may be obtained by sequencing the DNA fragments in the sample. In various cases, the DNA fragments include cfDNA molecules present in the sample. The cfDNA may include ctDNA and / or non-ctDNA. The DNA fragments may be extracted from the sample and prepared for sequencing. In some examples, adapters are ligated onto the DNA fragments, the DNA fragments are amplified, and the DNA fragments are captured. The captured DNA fragments, in various cases, are sequenced to obtain sequence reads that represent the sequences of the captured DNA fragments. The DNA fragments may be amplified using PCR, a non-PCR amplification technique, or an isothermal amplification technique. In some examples, the DNA fragments can be sequenced via a sequencer.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0184] In some examples, the DNA fragments are sequenced using NGS, MPS, WGS, whole exome sequencing, targeted sequencing, direct sequencing, Sanger sequencing, or any combination thereof. For example, a sequencer may identify the sequence of bases in the DNA fragments by detecting signals using the amplified DNA fragments. In some cases, the detection signals include optical signals (e.g., such as in the case of sequencing-by-synthesis techniques) and / or electrical signals (e.g., such as in the case of nanopore sequencing techniques). According to some cases, the presence of methylated cytosines in the DNA fragments are also determined. For example, the entity may perform methylation sequencing on the DNA fragments. According to various implementations, the entity obtains sequence read data indicating the type, order, and length of bases in the DNA fragments and / or methylation data indicating the location of methylated cytosines in the DNA fragments.

[0185] The entity may identify a genomic region of interest. For instance, the genomic region may be a continuous segment of a reference genome. The entity may identify a genomic region. In some cases, the genomic region includes an exon of the genome. The bases representing the genomic region may correspond to columns of the fragment image. For example, the columns respectively correspond to genomic positions within the genomic region. According to various cases, the genomic region has a length in a range of 200 to 1,000 bases.

[0186] In various cases, the entity determines at least a portion of the DNA fragments whose sequences correspond, at least in part, to the genomic region. Rows of the fragment image correspond to the portion of the DNA fragments. In various cases, the rows are arranged in order of a start position, a middle position, or an end position of the individual DNA fragments in the portion. In some cases, one or more rows of the fragment image are duplicated and / or interpolated. The fragment image, in some examples, is a 2D image.

[0187] In various implementations, pixels of the fragment image correspond to genomic positions (corresponding to the columns in which the pixels are located) represented by the DNA fragments in the portion (corresponding to the rows in which the pixels are located). The fragment images may include multiple channels. For example, a channel of an example pixel indicates a presence of a base in an example fragment at an example genomic position. A channel of the example pixel may indicate the presence of a CG dinucleotide in the example fragment at the example genomic position. A channel may indicate the presence of a methylated CG dinucleotide in the example fragment at the example genomic position. In some cases, a channel may indicate the presence of a variant in the example fragment at the example genomic position. For instance, the variant is a difference between a dinucleotide at the example genomic position and a dinucleotide in at least one reference sequence (e.g., a reference genome and / or pangenome). In some cases, a channel of an example pixel indicates a presence of a dinucleotide among a predetermined sequence in the example fragment at the example genomic position. For instance, the predetermined signal includes a promoter, an enhancer, a hotspot, a CpG island, an exon, a gene, or any combination thereof.

[0188] At 704, the entity predicts whether the subject has at least one type of disease based on the fragment image. According to various cases, the entity applies a predictive model configured to identify predictive features in the fragment image that are indicative of whether the subject has at least one type of disease (e.g., cancer). The predictive model, for instance, includes at least one ML model. In some cases, the ML model(s) include at least one CNN.

[0189] According to various cases, the entity determines a tissue origin or subtype of at least one cancer cell in the body of the subject. In some cases, the entity predicts an anatomic location of a tumor in the subject based on the fragment image. The anatomic location, for instance, is an organ or other type of differentiated tissue type in the bodyFMI Docket No.0032-P L&H Docket No.: F171-0008PCT of the subject. In some examples, the entity predicts a histological cancer type of the tumor of the subject. For instance, the entity may determine whether the subject has at least one of a carcinoma, a sarcoma, a myeloma, a leukemia, or a lymphoma. In some cases, the entity determines a primary site of a tumor of the subject. The primary site may be defined in terms of an anatomic location of a primary tumor of the subject. In various examples, the entity determines whether the subject has cancer cells that are responsive to a predetermined therapy or resistant to the predetermined therapy. In some cases, the entity determines whether the subject qualifies for a clinical trial. The entity, in some examples, determines whether the subject is part of a predetermined prognostic group and / or is likely to have a particular metastasis profile.

[0190] According to some implementations, the entity generates a vector corresponding to the likelihood (e.g., probability) that the subject has different types of predetermined cancers. For instance, the elements of the vector correspond to the likelihoods that the subject has the different types of predetermined cancers. In some cases, the elements of the vector correspond to one value (e.g., one or true) when the likelihoods exceed a predetermined threshold (e.g., 90%) or another value (e.g., zero or false) when the likelihoods are equal to or below the predetermined threshold. In various cases, the entity outputs a report based on the predicted cancer(s) of the subject.

[0191] FIG.8 illustrates an example process 800 for predicting a type of a cancer cell present in the body of a subject using a CNN. In various instances, the process 800 is performed by an entity including at least one processor, at least one computing device (e.g., one or more server computers), a sequencer (e.g., the sequencer 112), an image generator (e.g., the image generator 116), a predictive model (e.g., the predictive model 120), a report generator (e.g., the report generator 124), a clinical device (e.g., the clinical device 128), or any combination thereof.

[0192] At 802, the entity generates an image based on sequencing data of a sample obtained from a subject. According to various cases, the subject is a human. According to some cases, the subject has at least one type of cancer. For instance, the subject may have adrenal cancer, bladder cancer, blood cancer, bone cancer, brain cancer, breast cancer, carcinoma, cervical cancer, colon cancer, colorectal cancer, corpus uterine cancer, ear, nose and throat (ENT) cancer, endometrial cancer, esophageal cancer, gastrointestinal cancer, head and neck cancer, Hodgkin's disease, intestinal cancer, kidney cancer, larynx cancer, leukemia, liver cancer, lymph node cancer, lymphoma, lung cancer, melanoma, mesothelioma, myeloma, nasopharynx cancer, a neuroblastoma, non-Hodgkin's lymphoma, oral cancer, ovarian cancer, pancreatic cancer, penile cancer, pharynx cancer, prostate cancer, rectal cancer, sarcoma, seminoma, skin cancer, stomach cancer, a teratoma, testicular cancer, thyroid cancer, uterine cancer, vaginal cancer, a vascular tumor, or combinations or metastases thereof. In some examples, the sample is a liquid biopsy sample. For instance, the sample may include blood, plasma, cerebrospinal fluid, sputum, stool, urine, lymphatic fluid, saliva, or any combination thereof.

[0193] The data may be obtained by sequencing the nucleic acid molecules in the sample. In various cases, the nucleic acid molecules include cfDNA molecules present in the sample. The cfDNA may include ctDNA and / or non- ctDNA. The nucleic acid molecules may be extracted from the sample and prepared for sequencing. In some examples, adapters are ligated onto the nucleic acid molecules, the nucleic acid molecules are amplified, and the nucleic acid molecules are captured. The captured nucleic acid molecules, in various cases, are sequenced to obtain sequence reads that represent the sequences of the captured nucleic acid molecules. The nucleic acid molecules may be amplified using PCR, a non-PCR amplification technique, or an isothermal amplification technique. In some examples, the nucleic acid molecules can be sequenced via a sequencer.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0194] In some examples, the nucleic acid molecules are sequenced using NGS, MPS, WGS, whole exome sequencing, targeted sequencing, direct sequencing, Sanger sequencing, or any combination thereof. For example, a sequencer may identify the sequence of bases in the nucleic acid molecules by detecting signals using the amplified nucleic acid molecules. In some cases, the detection signals include optical signals (e.g., such as in the case of sequencing-by-synthesis techniques) and / or electrical signals (e.g., such as in the case of nanopore sequencing techniques). According to some cases, the presence of methylated cytosines in the nucleic acid molecules are also determined. For example, the entity may perform methylation sequencing on the nucleic acid molecules. According to various implementations, the entity obtains sequence read data indicating the type, order, and length of bases in the nucleic acid molecules and / or methylation data indicating the location of methylated cytosines in the nucleic acid molecules.

[0195] Various techniques may be utilized to convert the sequence read data into an image space. In some cases, the image is a fragment image, as described herein. One dimension (e.g., rows) of the image may correspond to various sequence reads in the sequence read data. Another dimension (e.g., columns) of the image may correspond to various genomic positions indicated by the sequence reads. In various cases, one or more channels of the pixels in the image correspond to characteristics of various bases identified by the sequence reads. For example, channels may indicate the presence, identity, methylation state, or any combination thereof, of bases in the sequence read data at various genomic positions. In some cases, one or more channels indicate whether variants or predetermined sequences (e.g., promoters, CpG islands, or the like) are present at various genomic positions.

[0196] At 804, the entity predicts, by inputting the image into a CNN, a type of cancer cell present in the body of the subject. The CNN may be defined by one or more hidden layers. The hidden layers are respectively defined by kernels. In some cases, the CNN includes at least one kernel defined by a rectangular set of pixels, such that a height of the kernel (in pixels) is different than a length of the kernel (in pixels). The values defining the pixels of the kernel may be optimized based on training data. In some cases, the training data includes images corresponding to sequence read data obtained from a population of individuals omitting the subject, as well as ground truth labels indicating the types of cancers experienced by the population. The population, for instance, includes people with different cancer states (e.g., at least one cancer-free person), different types of cancers, and different cancer severities. In response to receiving the image, the CNN may output an indication of one or more predicted cancer types of the subject.

[0197] According to various cases, the entity determines a tissue origin or subtype of at least one cancer cell in the body of the subject. In some cases, the entity predicts an anatomic location of a tumor in the subject based on the fragment image. The anatomic location, for instance, is an organ or other type of differentiated tissue type in the body of the subject. In some examples, the entity predicts a histological cancer type of the tumor of the subject. For instance, the entity may determine whether the subject has at least one of a carcinoma, a sarcoma, a myeloma, a leukemia, or a lymphoma. In some cases, the entity determines a primary site of a tumor of the subject. The primary site may be defined in terms of an anatomic location of a primary tumor of the subject. In various examples, the entity determines whether the subject has cancer cells that are responsive to a predetermined therapy or resistant to the predetermined therapy. In some cases, the entity determines whether the subject qualifies for a clinical trial. The entity, in some examples, determines whether the subject is part of a predetermined prognostic group and / or is likely to have a particular metastasis profile.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0198] According to some implementations, the entity generates a vector corresponding to the likelihood that the subject has different types of predetermined cancers. For instance, the elements of the vector correspond to the likelihoods that the subject has the different types of predetermined cancers. In some cases, the elements of the vector correspond to one value (e.g., one or true) when the likelihoods exceed a predetermined threshold (e.g., 90%) or another value (e.g., zero or false) when the likelihoods are equal to or below the predetermined threshold. In various cases, the entity outputs a report based on the predicted cancer(s) of the subject.

[0199] FIG.9 illustrates an example environment 900 for sequencing various nucleic acid molecules 902. In various implementations, the nucleic acid molecules 902 include cfDNA and / or gDNA. For instance, the nucleic acid molecules 902 may include ctDNA. The nucleic acid molecules 902, in various cases, are extracted from a sample, such as a biological sample obtained from a subject. In some implementations, the nucleic acid molecules 902 include DNA that is complementary to RNA present in the sample.

[0200] The nucleic acid molecules 902, in various cases, are ligated with adapters 904. For examples, the adapters 904 are hybridized to the nucleic acid molecules 902. The adapters 904, for example, include additional nucleic acid molecules. In various implementations, the adapters 904 have a shorter length than the nucleic acid molecules 902 being sequenced. For instance, the adapters 904 include amplification primers, flow cell adapter sequences, substrate adapter sequences, or sample index sequences. Although FIG.9 illustrates adapters 904 being ligated to one end of each of the nucleic acid molecules 902, implementations are not so limited. For example, the adapters 904 may be ligated to both ends of each of the nucleic acid molecules 902.

[0201] In various examples, the nucleic acid molecules 902 ligated with the adapters 904 are amplified in order to generate amplified molecules 906. Various amplification techniques can be performed. For instance, the amplified molecules 906 are generated using PCR, a non-PCR amplification technique, an isothermal amplification technique, or any combination thereof.

[0202] Amplified molecules 906 may be captured by bait molecules 910 and sequenced. In some implementations, the amplified molecules 906 are sequenced via sequencing-by-synthesis. In various cases, fluorescently tagged deoxyribonucleotide triphosphates (dNTP) 912 (a type of fluorescently tagged nucleotide triphosphates (NTPs)) are utilized to synthesize a strand that is complementary to DNA strands bound to the substrate 908. When a dNTP 912 is added to the strand (e.g., by an enzyme), the dNTP 912 emits an optical signal 914. In various implementations, the frequency of the optical signal 914 is dependent on the type of dNTP 912 from which the optical signal 914 is emitted. By detecting the optical signals 914 as the strand is being synthesized, the sequence of the original nucleic acid molecules 902 can be derived.

[0203] In some implementations, the amplified molecules 906 are sequenced via nanopore sequencing. For instance, the amplified molecules 906 are directed through a nanopore 916 extending through a substrate 918. In various cases, the amplified molecules 906 are negatively charged, such that they can be directed through the nanopore 916 by imposing an electrical field across the substrate 918. In various cases, the amplified molecules 906 and the nanopore 916 are in the presence of a charged solution. Thus, charged solutes traveling through the nanopore 916 can be monitored by reviewing an electrical signal (e.g., a current) sensed between electrodes 920 disposed on either side of the substrate 918. As an amplified molecule 906 is directed through the nanopore 916, the individual bases within the amplified molecule 906 will block the nanopore 916, which may decrease the amount of chargedFMI Docket No.0032-P L&H Docket No.: F171-0008PCT solutes traveling through the nanopore 916 and consequently, the magnitude of the electrical signal detected by the electrodes 920. Each of the four types of bases within the amplified molecules 906, may block the nanopore 916 to a different extent. Therefore, the sequence of the nucleic acid molecules 902 can be derived by analyzing the measured electrical signal with respect to time as the amplified molecules 906 are directed through the nanopore 916.

[0204] FIG.10 illustrates one or more devices 1000 configured to perform various operations described herein. The device(s) 1000 include one or more processor(s) 1002. In some implementations, the processor(s) 1002 includes a central processing unit (CPU), a graphics processing unit (GPU), both CPU and GPU, or other processing unit or component known in the art.

[0205] The processor(s) 1002 is operably connected to memory 1004. In various implementations, the memory 1004 is volatile (such as random access memory (RAM)), non-volatile (such as read only memory (ROM), flash memory, etc.) or some combination of the two. The memory 1004 stores instructions that, when executed by the processor(s) 1002, causes the processor(s) 1002 to perform various operations. In various examples, the memory 1004 stores methods, threads, processes, applications, objects, modules, any other sort of executable instruction, or a combination thereof. In some cases, the memory 1004 stores files, databases, or a combination thereof. In some examples, the memory 1004 includes, but is not limited to, RAM, ROM, electrically erasable programmable read-only memory (EEPROM), flash memory, or any other memory technology. In some examples, the memory 1004 includes one or more of CD-ROMs, digital versatile discs (DVDs), content-addressable memory (CAM), or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by the processor(s) 1002. For instance, the memory 1004 stores instructions that, when executed by the processor(s) 1002, causes the processor(s) 1002 to perform operations of the image generator 116, the predictive model 120, and the report generator 124.

[0206] The processor(s) 1002 is operably connected to one or more input devices 1006 and one or more output devices 1008. Collectively, the input device(s) 1006 and the output device(s) 1008 function as an interface between at least one user and the device(s) 1000. The input device(s) 1006 is configured to receive an input from a user and includes at least one of a keypad, a cursor control, a touch-sensitive display, a voice input device (e.g., a microphone), a haptic feedback device (e.g., a gyroscope), or any combination thereof. The output device(s) 1008 includes at least one of a display, a speaker, a haptic output device, a printer, or any combination thereof. In various examples, the processor(s) 1002 causes a display among the input device(s) 1006 to visually output various data described herein. In some implementations, the input device(s) 1006 includes one or more touch sensors, the output device(s) 1008 includes a display screen, and the touch sensor(s) are integrated with the display screen.

[0207] In various implementations, the processor(s) 1002 is operably connected to one or more transceivers 1010 that transmit and / or receive data over one or more communication networks 1012. For example, the transceiver(s) 1010 includes a network interface card (NIC), a network adapter, a local area network (LAN) adapter, or a physical, virtual, or logical address to connect to the various external devices and / or systems. In various examples, the transceiver(s) 1010 includes any sort of wireless transceivers capable of engaging in wireless communication (e.g., radio frequency (RF) communication). For example, the communication network(s) 1012 includes one or more wireless networks that include a 3rd Generation Partnership Project (3GPP) network, such as a Long Term Evolution (LTE) radio access network (RAN) (e.g., over one or more LTE bands), a New Radio (NR) RAN (e.g., over one or more NRFMI Docket No.0032-P L&H Docket No.: F171-0008PCT bands), or a combination thereof. In some cases, the transceiver(s) 1010 includes other wireless modems, such as a modem for engaging in WI-FI®, WIGIG®, WIMAX®, BLUETOOTH®, or infrared communication over the communication network(s) 1012.

[0208] The device(s) 1000 may further include the sequencer 112. In various implementations, the sequencer 112 includes one or more fluidic circuits 1014 configured to receive a sample 1016 derived from a subject 1018. The sequencer 112, in various cases, may be configured to generate data indicative of one or more sequences of nucleic acid molecules (e.g., DNA and / or RNA) present in the sample 1016. In various cases, the sequencer 112 introduces one or more reagents 1019 to the fluidic circuit(s) 1014 in order to prepare for and perform sequencing of the nucleic acid molecules. Further, the sequencer 112 may include one or more sensors 1020 configured to measure or otherwise detect detection signals from the fluidic circuit(s) 1014, which may be indicative of the sequences of the nucleic acid molecules. According to various implementations, the sensor(s) 1020 may further include one or more ADCs. The sequencer 112, in various cases, outputs sequence read data to the processor(s) 1002 for additional processing. Experimental Example

[0209] FIGS.11A to 12D illustrate various examples of fragment images, according to some implementations of the present disclosure. Each of the fragment images illustrated in FIGS.11A to 12D is an RGB image. Rows of the images correspond to sequenced DNA fragments, and columns of the images correspond to genomic positions. The red channel of pixels in each image indicates the presence (e.g., red value = 128) or absence (e.g., red value = 0) of base pairs in the sequenced DNA fragments. The green channel of pixels in each image indicates the presence (e.g., green value = 255) or absence (e.g., green value = 0) of CG dinucleotides in the sequenced DNA fragments. The blue channel of pixels in each image indicates the presence (e.g., blue value = 255) or absence (e.g., blue value = 0) of methylated CG dinucleotides in the sequenced DNA fragments. The rows of each image are arranged according to the middle positions of the fragments indicated by the rows.

[0210] FIGS.11A to 11C illustrate examples of fragment images generated based on a sample obtained from a subject without cancer. FIG.11A illustrates a fragment image corresponding to a relatively large genomic region. FIGS.11B and 11C illustrate fragment images corresponding to relatively small genomic regions that are included in the large genomic region. In other words, FIGS.11B and 11C are cropped from FIG.11A.

[0211] FIGS.12A to 12D illustrate examples of fragment images generated based on a sample obtained from a subject with cancer. FIGS.12A and 12B illustrate fragment images corresponding to relatively large genomic regions. FIGS.12C and 12D illustrate fragment images corresponding to relatively small genomic regions that are included in the large genomic regions, such that FIGS.12C and 12D can represent cropped versions of FIGS.12A and 12B.

[0212] FIG.13 illustrates results from an example CNN trained to differentiate between fragment images generated based on samples from subjects without cancer and fragment images generated based on samples from subjects with cancer. For instance, the fragment images illustrated in FIGS.11A to 12D, as well as other training images (e.g., 219 non-cancer images, 214 colorectal cancer images, and 170 lung cancer images) were used as training images for the CNN. The example CNN has two hidden layers and one flatten layer with 128 nodes. A final layer in the CNN utilizes a sigmoid activation function. FIG.13 (top) illustrates a loss of the CNN over a number of training images used to optimize parameters of the CNN. FIG.13 (bottom) illustrates an accuracy of the CNN over the number of training images used to optimize the parameters of the CNN. As shown, the loss and accuracy level off at around 10 trainingFMI Docket No.0032-P L&H Docket No.: F171-0008PCT images. FIG.13 illustrates that the CNN is able to identify distinctive predictive attributes in fragment images that are indicative of the presence of cancer cells in subjects corresponding to the fragment images. Example Clauses 1. A method, including: providing a plurality of nucleic acid molecules obtained from a sample from a subject; extracting, from the sample, a plurality of cell-free DNA (cfDNA) molecules; ligating one or more adapters onto one or more of the cfDNA molecules; amplifying the one or more ligated cfDNA molecules; capturing all or a subset of the amplified cfDNA molecules; sequencing, by a sequencer, all or a subset of the captured cfDNA molecules to obtain a plurality of sequence reads that represent the captured cfDNA molecules; receiving, at one or more processors, sequence read data for the plurality of sequence reads; generating, using the one or more processors, a fragment image indicating at least a portion of the cfDNA corresponding to a region of a genome, wherein an example row of the fragment image corresponds to a fragment among the cfDNA represented in the sequence read data, a first channel of an example pixel of the example row indicating a presence of a base in the fragment at a particular position in the region, a second channel of the example pixel indicating whether the base includes a cytosine, a third channel of the example pixel indicating whether the base includes a methylated cytosine; inputting input data including the fragment image into at least one model configured to generate a probability that the subject has at least one cancer; and generating, using the one or more processors, a report based on the probability that the subject has at least one cancer. 2. The method of clause 1, wherein the fragment image includes a plurality of rows respectively corresponding to a plurality of fragments among the cfDNA, and wherein the plurality of rows in the fragment image are ordered based on middle positions of the plurality of fragments in the region. 3. The method of clause 1 or 2, wherein the region includes a genomic interval whose methylation status is indicative of one or more of a tissue-of-origin of the at least one cancer, a prognosis of the at least one cancer, or a therapy response of the at least one cancer. 4. The method of any of clauses 1 to 3, wherein the region includes an exon of the genome. 5. The method of any of clauses 1 to 4, wherein the at least one model includes at least one machine learning (ML) model. 6. The method of any of clauses 1 to 5, wherein the at least one model includes at least one convolutional neural network (CNN). 7. A method, including: identifying data indicative of DNA from a sample derived from a subject; generating, based on the data, a fragment image indicating a plurality of fragments among the DNA corresponding to a region of a genome; inputting input data including the fragment image into a model configured to generate a prediction of whether the subject has at least one cancer; and generating a report based on the prediction. 8. The method of clause 7, wherein the DNA includes cell-free DNA (cfDNA). 9. The method of clause 7 or 8, wherein the DNA includes at least one fragment having a length in a range of about 1 base to about 500 bases. 10. The method of any of clauses 7 to 9, wherein the sample includes a liquid biopsy sample. 11. The method of any of clauses 7 to 10, wherein the sample includes a blood sample. 12. The method of any of clauses 7 to 11, wherein the sample includes plasma.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT 13. The method of any of clauses 7 to 12, wherein the subject has adrenal cancer, bladder cancer, blood cancer, bone cancer, brain cancer, breast cancer, carcinoma, cervical cancer, colon cancer, colorectal cancer, corpus uterine cancer, ear, nose and throat (ENT) cancer, endometrial cancer, esophageal cancer, gastrointestinal cancer, head and neck cancer, Hodgkin's disease, intestinal cancer, kidney cancer, larynx cancer, leukemia, liver cancer, lymph node cancer, lymphoma, lung cancer, melanoma, mesothelioma, myeloma, nasopharynx cancer, a neuroblastoma, non-Hodgkin's lymphoma, oral cancer, ovarian cancer, pancreatic cancer, penile cancer, pharynx cancer, prostate cancer, rectal cancer, sarcoma, seminoma, skin cancer, stomach cancer, a teratoma, testicular cancer, thyroid cancer, uterine cancer, vaginal cancer, a vascular tumor, or combinations or metastases thereof. 14. The method of any of clauses 7 to 13, wherein the data indicative of the DNA includes sequence read data of the DNA. 15. The method of clause 14, further including: ligating one or more adapters onto one or more DNA molecules in the sample; amplifying the one or more ligated DNA molecules; capturing all or a subset of the amplified DNA molecules; and sequencing, by a sequencer, the captured DNA molecules to obtain a plurality of sequence reads that represent the captured DNA molecules, wherein the sequence read data is indicative of the sequence reads. 16. The method of clause 15, further including: extracting the one or more DNA molecules from the sample. 17. The method of clause 15 or 16, wherein the one or more adapters include at least one of amplification primers, flow cell adapter sequences, substrate adapter sequences, or sample index sequences. 18. The method of any of clauses 15 to 17, wherein the captured DNA molecules are captured from the amplified DNA molecules by hybridization to one or more bait molecules. 19. The method of clause 18, wherein the one or more bait molecules include one or more additional nucleic acid molecules, each of the one or more additional nucleic acid molecules including a region that is complementary to a region of a captured DNA molecule. 20. The method of any of clauses 15 to 19, wherein amplifying the one or more ligated DNA molecules includes performing a polymerase chain reaction (PCR) amplification technique, a non-PCR amplification technique, or an isothermal amplification technique. 21. The method of any of clauses 15 to 20, wherein sequencing the captured DNA molecules includes use of a massively parallel sequencing (MPS) technique, whole genome sequencing (WGS), whole exome sequencing, targeted sequencing, direct sequencing, or Sanger sequencing. 22. The method of any of clauses 15 to 21, wherein sequencing the captured DNA molecules includes next- generation sequencing (NGS). 23. The method of clause 22, wherein sequencing the captured DNA molecules is performed by a next- generation sequencer. 24. The method of any of clauses 15 to 23, wherein sequencing the captured DNA molecules includes sequencing-by-synthesis or nanopore sequencing. 25. The method of any of clauses 15 to 24, further including: generating ligated molecules by ligating adapters onto DNA molecules of the sample; generating amplified ligated DNA molecules by amplifying the ligated DNAFMI Docket No.0032-P L&H Docket No.: F171-0008PCT molecules; generating, using the amplified ligated DNA molecules, detection signals; detecting, by at least one sensor, the detection signals; and generating the sequence read data based on the detection signals. 26. The method of clause 25, wherein the detection signals include electrical signals and / or optical signals. 27. The method of clause 25 or 26, wherein generating, using the amplified ligated DNA molecules, the detection signals includes simultaneously: synthesizing, by a polymerase using fluorescently tagged nucleotide triphosphates (NTPs), a synthesized nucleic acid molecule based on one of the amplified ligated DNA molecules, and wherein detecting, by the at least one sensor, the detection signals includes: detecting, by at least one optical sensor, optical signals emitted by the fluorescently tagged NTPs upon binding to the synthesized nucleic acid molecule, the optical signals being indicative of at least one sequence of the DNA. 28. The method of any of clauses 25 to 27, wherein generating, using the amplified ligated DNA molecules, the detection signals includes simultaneously: directing the amplified ligated DNA molecules through a nanopore extending from a first space to a second space through a substrate, and wherein detecting, by the at least one sensor, the detection signals includes: detecting, by sensors disposed in the first space and the second space, an electrical signal over time, the electrical signal being indicative of at least one sequence of the DNA. 29. The method of any of clauses 7 to 28, further including: receiving the sample. 30. The method of clause 29, wherein the sample includes blood, plasma, cerebrospinal fluid, sputum, stool, urine, lymphatic fluid, or saliva. 31. The method of clause 29 or 30, wherein the sample includes genomic DNA. 32. The method of any of clauses 29 to 31, further including: extracting the DNA from the sample, wherein identifying the data indicative of the DNA includes sequencing the DNA. 33. The method of any of clauses 7 to 32, wherein the DNA includes circulating tumor DNA (ctDNA). 34. The method of any of clauses 7 to 33, wherein the fragment image includes a plurality of rows respectively corresponding to the plurality of fragments in the DNA. 35. The method of clause 34, wherein the plurality of rows in the fragment image are ordered based on start positions, middle positions, or end positions of the plurality of fragments within the genomic region. 36. The method of any of clauses 7 to 35, wherein the fragment image includes a plurality of columns respectively corresponding to a plurality of positions within the genomic region. 37. The method of any of clauses 7 to 36, wherein an example pixel is in a row of the fragment image corresponding to an example fragment among the plurality of fragments at an example genomic position among a plurality of positions within the genomic region. 38. The method of clause 37, wherein a channel of the example pixel indicates a presence of a base pair in the example fragment at the example genomic position. 39. The method of clause 37 or 38, wherein a channel of the example pixel indicates a presence of a CG dinucleotide in the example fragment at the example genomic position. 40. The method of any of clauses 37 to 39, wherein a channel of the example pixel indicates a presence of a methylated CG dinucleotide in the example fragment at the example genomic position. 41. The method of any of clauses 37 to 40, wherein a channel of the example pixel indicates a presence of a variant in the example fragment at the example genomic position.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT 42. The method of clause 41, wherein the variant includes a difference between a dinucleotide at the example genomic position and a dinucleotide in at least one reference sequence. 43. The method of any of clauses 37 to 42, wherein a channel of the example pixel indicates a presence of a dinucleotide among a predetermined sequence in the example fragment at the example genomic position. 44. The method of clause 43, wherein the predetermined sequence includes at least one of a promoter, an enhancer, a hotspot, a CpG island, an exon, or a gene. 45. The method of any of clauses 7 to 44, wherein the fragment image includes a red-green-blue (RGB) image. 46. The method of any of clauses 7 to 45, wherein the fragment image includes a two-dimensional (2D) image. 47. The method of any of clauses 7 to 46, wherein the fragment image includes at least one upsampled row including a duplicated row or an interpolated row. 48. The method of any of clauses 7 to 47, wherein the fragment image includes a number of columns in a range of about 200 to about 1000. 49. The method of any of clauses 7 to 48, wherein the model includes at least one machine learning (ML) model. 50. The method of clause 49, wherein the at least one ML model includes at least one convolutional neural network (CNN) including a plurality of convolutional layers. 51. The method of clause 49 or 50, wherein the at least one ML model includes at least one multi-instance learning (MIL) model. 52. The method of any of clauses 49 to 51, further including: training the at least one ML model by optimizing parameters of the at least one ML model based on training data, the training data including example fragment images identified from example samples of a population. 53. The method of clause 52, wherein the population omits the subject. 54. The method of clause 52 or 53, wherein the population includes at least one first individual with cancer and at least one second individual without cancer. 55. The method of clause 54, wherein the at least one first individual includes: at least one third individual having a first type of cancer; and at least one fourth individual having a second type of cancer, wherein the prediction of whether the subject has at least one cancer includes a prediction of whether the subject has the first type of cancer and a prediction of whether the subject has the second type of cancer. 56. The method of clause 55, wherein the first type of cancer corresponds to a first tissue origin and the second type of cancer corresponds to a second tissue origin. 57. The method of any of clauses 54 to 56, wherein the at least one first individual includes individuals with different cancer stages. 58. The method of any of clauses 52 to 57, wherein the training data further includes labels indicating whether the example samples are obtained from at least one individual having cancer, and wherein training the at least one ML model includes identifying, using supervised ML based on pairs of the labels and corresponding instances of the example fragment images, predictive attributes of the example fragment images that are indicative of the labels.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT 59. The method of clause 58, wherein generating the prediction of whether the cancer has cancer includes: identify instances of the predictive attributes associated with the fragment image; and generate the prediction of whether the subject has cancer based on the instances of the predictive attributes. 60. The method of any of clauses 49 to 59, wherein the at least one ML model includes: a first ML model configured to generate a first prediction that the subject has a first type of cancer; and a second ML model configured to generate a second prediction that the subject has a second type of cancer. 61. The method of clause 60, further including: identifying example fragment images of example DNA in example samples obtained from a population; identifying first labels indicating whether the population has the first type of cancer; identifying second labels indicating whether the population has the second type of cancer; training the first ML model based on first training data including: the example fragment images; and the first labels; and training the second ML model based on second training data including: the example fragment images; and the second labels. 62. The method of any of clauses 7 to 61, the prediction of whether the subject has at least one cancer includes a prediction of whether the subject has a type of cancer. 63. The method of clause 62, wherein the type of cancer indicates an anatomic location of a tumor. 64. The method of clause 63, wherein the anatomic location includes an organ and / or differentiated tissue type. 65. The method of any of clauses 62 to 64, wherein the type of cancer includes a histological cancer type of a tumor. 66. The method of clause 65, wherein the histological cancer type includes at least one of a carcinoma, a sarcoma, a myeloma, a leukemia, or a lymphoma. 67. The method of any of clauses 62 to 66, wherein the type of cancer indicates a primary site of a primary tumor. 68. The method of clause 67, wherein the DNA includes ctDNA from the primary tumor. 69. The method of clause 67 or 68, wherein the DNA includes ctDNA from a secondary tumor. 70. The method of any of clauses 67 to 69, wherein the primary site includes an anatomical location of the primary tumor. 71. The method of clause 70, wherein the anatomical location includes an organ and / or a differentiated tissue type. 72. The method of any of clauses 62 to 71, wherein the type of cancer indicates a tissue origin of a tumor of the subject. 73. The method of any of clauses 62 to 71, wherein the type of cancer indicates a subtype of at least one cancer cell of the subject. 74. The method of clause 73, wherein at least a portion of the DNA includes ctDNA originating from the at least one cancer cell. 75. The method of any of clauses 62 to 74, wherein the type of cancer indicates whether the subject has a tumor that is resistant to a predetermined therapy or responsive to a predetermined therapy. 76. The method of clause 75, wherein the predetermined therapy includes at least one of an immunotherapy, a chemotherapy, or a radiotherapy.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT 77. The method of any of clauses 62 to 76, wherein the type of cancer indicates whether the subject qualifies for a clinical trial. 78. The method of any of clauses 62 to 77, wherein the type of cancer indicates whether the subject is in a predetermined prognostic group. 79. The method of any of clauses 62 to 78, wherein the type of cancer indicates a metastasis profile. 80. The method of any of clauses 7 to 79, wherein the model is configured to generate the prediction of whether the subject has at least one cancer by: determining a probability that the subject has at least one cancer; and determining whether the probability exceeds a threshold. 81. The method of any of clauses 7 to 80, wherein the report includes an instruction to perform a follow-up test on the subject. 82. The method of clause 81, wherein the follow-up test includes at least one of: a histological study; whole transcriptome sequencing; cfRNA sequencing; whole exome sequencing; whole genome sequencing; a cancer hotspot panel test; a DNA methylation test; a DNA fragmentation test; an RNA fragmentation test; a microsatellite instability (MSI) test; a tumor mutational burden (TMB) test; or a viral status test. 83. The method of clause 81 or 82, wherein the follow-up test includes at least one of whole transcriptome sequencing; cfRNA sequencing; or an RNA fragmentation test. 84. The method of any of clauses 7 to 83, wherein the report includes a recommendation to administer a predetermined therapy to the subject. 85. The method of clause 84, wherein the predetermined therapy includes at least one of chemotherapy, radiotherapy, immunotherapy, targeted therapy, or surgery. 86. The method of any of clauses 7 to 85, further including: generating, based on the prediction of whether the subject has cancer, a genomic profile of the subject, the report including the genomic profile. 87. The method of clause 86, wherein the genomic profile includes results from at least one of: a histological study; whole transcriptome sequencing; cfRNA sequencing; whole exome sequencing; whole genome sequencing; a cancer hotspot panel test; a DNA methylation test; a DNA fragmentation test; an RNA fragmentation test; a microsatellite instability (MSI) test; a tumor mutational burden (TMB) test; or a viral status test. 88. The method of clause 86 or 87, wherein the genomic profile of the subject includes: results from a nucleic acid sequencing-based test. 89. The method of any of clauses 86 to 88, further including: selecting, based on the genomic profile and / or the prediction, an anticancer agent for administration to the subject. 90. The method of clause 89, further including: administering the anticancer agent to the subject. 91. The method of any of clauses 86 to 90, further including: applying, based on the genomic profile, an anticancer therapy to the subject. 92. The method of clause 91, wherein the anticancer therapy includes at least one of chemotherapy, radiotherapy, immunotherapy, a targeted therapy, or surgery. 93. The method of any of clauses 7 to 91, further including: identifying, based on the prediction, a suggested treatment decision for the subject, the report including the suggested treatment decision.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT 94. The method of clause 93, wherein the suggested treatment decision includes chemotherapy, radiotherapy, immunotherapy, targeted therapy, or surgery. 95. The method of any of clauses 7 to 94, further including: outputting the report. 96. The method of clause 95, wherein outputting the report includes: transmitting data indicating the report to an external device. 97. The method of clause 96, wherein the external device is associated with the subject or a healthcare provider. 98. The method of clause 96 or 97, wherein the data indicating the report is transmitted over one or more communication networks. 99. The method of any of clauses 96 to 98, wherein the data indicating the report is transmitted over a peer-to- peer connection. 100. The method of any of clauses 95 to 99, wherein outputting the report includes: visually presenting, by a display, the report. 101. The method of any of clauses 7 to 100, further including: generating, based on the prediction, a therapy for the subject. 102. The method of clause 101, wherein the therapy includes a dosage of one or more therapeutic agents predicted to treat a tumor of the subject. 103. The method of any of clauses 7 to 102, further including: determining, based on the prediction, whether the subject is eligible for a clinical trial, wherein the report indicates whether the subject is eligible for the clinical trial. 104. The method of any of clauses 7 to 103, further including: identifying data indicative of additional biomarkers of the subject, wherein the input data further includes the data indicative of the additional biomarkers of the subject. 105. The method of clause 104, wherein the additional biomarkers include at least one of results from: a histological study; whole transcriptome sequencing; cfRNA sequencing; whole exome sequencing; whole genome sequencing; a cancer hotspot panel test; a DNA methylation test; a DNA fragmentation test; an RNA fragmentation test; a microsatellite instability (MSI) test; a tumor mutational burden (TMB) test; or a viral status test. 106. A system, including: at least one processor; and memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations including: the method of any of clauses 7 to 105. 107. The system of clause 106, further including: a sequencer configured to generate the data by sequencing the DNA. 108. The system of clause 106 or 107, further including: a transceiver configured to receive a communication signal encoding the data. 109. The system of any of clauses 106 to 108, further including: a transceiver configured to transmit, to an external device, a communication signal encoding the report. 110. The system of any of clauses 106 to 109, further including: a display configured to visually present the report.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT 111. A non-transitory computer readable medium storing instructions for performing operations including: the method of any of clauses 7 to 105. 112. A method, including: providing a plurality of nucleic acid molecules obtained from a sample from a subject; extracting, from the sample, a plurality DNA molecules; ligating one or more adapters onto one or more of the DNA molecules; amplifying the one or more ligated DNA molecules; capturing all or a subset of the amplified DNA molecules; sequencing, by a sequencer, all or a subset of the captured DNA molecules to obtain a plurality of sequence reads that represent the captured DNA molecules; receiving, at one or more processors, sequence read data for the plurality of sequence reads; generating, using the one or more processors, a fragment image indicating at least a portion of the DNA corresponding to a region of a genome, wherein an example row of the fragment image corresponds to a fragment among the DNA represented in the sequence read data, a first channel of an example pixel of the example row indicating a presence of a base in the fragment at a particular position in the region, a second channel of the example pixel indicating whether the base includes a cytosine, a third channel of the example pixel indicating whether the base includes a methylated cytosine; inputting input data including the fragment image into at least one model configured to generate a probability that the subject has at least one cancer; and generating, using the one or more processors, a report based on the probability that the subject has at least one cancer. 113. A method, including: identifying data indicative of DNA from a sample derived from a subject; generating, based on the data, a fragment image indicating a plurality of fragments among the DNA corresponding to a region of a genome; inputting input data including the fragment image into a model configured to generate a prediction of whether the subject has at least one cancer; and generating a report based on the prediction. 114. The method of clause 113, wherein an example row of the fragment image corresponds to a fragment among the DNA represented in the data, a channel of an example pixel of the example row indicating a presence of a predetermined base in the fragment at a particular position in the region. 115. The method of clause 113 or 114, wherein an example row of the fragment image corresponds to a fragment among the DNA represented in the data, a channel of an example pixel of the example row indicating whether a predetermined variant is present in the fragment at a particular position in the region. 116. The method of clause 115, wherein the predetermined variant includes at least one of a CNV, an indel, a rearrangement, or a CNA. 117. The method of any of clauses 113 to 116, wherein an example row of the fragment image corresponds to a fragment among the DNA represented in the data, a channel of an example pixel of the example row indicating whether a methylated cytosine is present in the fragment at a particular position in the region. 118. The method of any of clauses 113 to 117, wherein an example row of the fragment image corresponds to a fragment among the DNA represented in the data, a channel of an example pixel of the example row indicating whether a hydroxymethylated cytosine is present in the fragment at a particular position in the region. Conclusion

[0213] All publications, patents, and patent applications mentioned in this specification are herein incorporated by reference in their entirety to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference in its entirety. In the event of a conflict between a term herein and a term in an incorporated reference, the term herein controls.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT

[0214] The features disclosed in the foregoing description, or the following claims, or the accompanying drawings, expressed in their specific forms or in terms of a means for performing the disclosed function, or a method or process for attaining the disclosed result, as appropriate, may, separately, or in any combination of such features, be used for realizing implementations of the disclosure in diverse forms thereof.

[0215] As will be understood by one of ordinary skill in the art, each implementation disclosed herein can comprise, consist essentially of or consist of its particular stated element, step, or component. Thus, the terms “include” or “including” should be interpreted to recite: “comprise, consist of, or consist essentially of.” The transition term “comprise” or “comprises” means has, but is not limited to, and allows for the inclusion of unspecified elements, steps, ingredients, or components, even in major amounts. The transitional phrase “consisting of” excludes any element, step, ingredient or component not specified. The transition phrase “consisting essentially of” limits the scope of the implementation to the specified elements, steps, ingredients or components and to those that do not materially affect the implementation. As used herein, the term “based on” is equivalent to “based at least partly on,” unless otherwise specified.

[0216] Unless otherwise indicated, all numbers expressing quantities, properties, conditions, and so forth used in the specification and claims are to be understood as being modified in all instances by the term “about.” Accordingly, unless indicated to the contrary, the numerical parameters set forth in the specification and attached claims are approximations that may vary depending upon the desired properties sought to be obtained by the present disclosure. At the very least, and not as an attempt to limit the application of the doctrine of equivalents to the scope of the claims, each numerical parameter should at least be construed in light of the number of reported significant digits and by applying ordinary rounding techniques. When further clarity is required, the term “about” has the meaning reasonably ascribed to it by a person skilled in the art when used in conjunction with a stated numerical value or range, i.e., denoting somewhat more or somewhat less than the stated value or range, to within a range of ±20% of the stated value; ±19% of the stated value; ±18% of the stated value; ±17% of the stated value; ±16% of the stated value; ±15% of the stated value; ±14% of the stated value; ±13% of the stated value; ±12% of the stated value; ±11% of the stated value; ±10% of the stated value; ±9% of the stated value; ±8% of the stated value; ±7% of the stated value; ±6% of the stated value; ±5% of the stated value; ±4% of the stated value; ±3% of the stated value; ±2% of the stated value; or ±1% of the stated value.

[0217] Notwithstanding that the numerical ranges and parameters setting forth the broad scope of the disclosure are approximations, the numerical values set forth in the specific examples are reported as precisely as possible. Any numerical value, however, inherently contains certain errors necessarily resulting from the standard deviation found in their respective testing measurements.

[0218] The terms “a,” “an,” “the,” and similar referents used in the context of describing implementations (especially in the context of the following claims) are to be construed to cover both the singular and the plural, unless otherwise indicated herein or clearly contradicted by context. Recitation of ranges of values herein is merely intended to serve as a shorthand method of referring individually to each separate value falling within the range. Unless otherwise indicated herein, each individual value is incorporated into the specification as if it were individually recited herein. All methods described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The use of any and all examples, or exemplary language (e.g., “such as”) provided herein isFMI Docket No.0032-P L&H Docket No.: F171-0008PCT intended merely to better illuminate implementations of the disclosure and does not pose a limitation on the scope of the disclosure. No language in the specification should be construed as indicating any non-claimed element essential to the practice of implementations of the disclosure.

[0219] Groupings of alternative elements or implementations disclosed herein are not to be construed as limitations. Each group member may be referred to and claimed individually or in any combination with other members of the group or other elements found herein. It is anticipated that one or more members of a group may be included in, or deleted from, a group for reasons of convenience and / or patentability. When any such inclusion or deletion occurs, the specification is deemed to contain the group as modified thus fulfilling the written description of all Markush groups used in the appended claims.

[0220] Unless otherwise indicated, the practice of the present disclosure can employ conventional techniques of immunology, molecular biology, microbiology, cell biology and recombinant DNA. These methods are described in the following publications. See, e.g., Sambrook, et al. Molecular Cloning: A Laboratory Manual, 2nd Edition (1989); F. M. Ausubel, et al. eds., Current Protocols in Molecular Biology, (1987); the series Methods IN Enzymology (Academic Press, Inc.); M. MacPherson, et al., PCR: A Practical Approach, IRL Press at Oxford University Press (1991); MacPherson et al., eds. PCR 2: Practical Approach, (1995); Harlow and Lane, eds. Antibodies, A Laboratory Manual, (1988); and R. I. Freshney, ed. Animal Cell Culture (1987).

[0221] Tumor mutational burden (TMB) is a measure of the number of mutations carried by tumor cells. By comparing DNA sequences from a patient’s healthy tissues and tumor cells, the number of acquired somatic mutations present in tumors, but not in normal tissues, may be determined. In some instances, driver mutations may be excluded from a TMB calculation.

[0222] In certain examples, "tumor mutational burden" or “TMB” refers to the number of somatic mutations in a tumor's genome and / or the number of somatic mutations per area of the tumor's genome. In some embodiments, TMB, as used herein, refers to the number of somatic mutations per megabase (Mb) of DNA sequenced. In some embodiments, germline (inherited) variants are excluded when determining TMB, given that the immune system has a higher likelihood of recognizing these as self. In various cases, driver mutations are excluded from a TMB calculation.

[0223] Microsatellites are highly polymorphic DNA-repeat regions. In certain examples, “microsatellite” refers to a repetitive nucleic acid having repeat units of less than about 10 base pairs or nucleotides in length. In certain examples, a microsatellite refers to a tract of tandemly repeated (i.e. adjacent) DNA motifs ranging from one to six or up to ten nucleotides, with each motif repeated 5 to 50 repeated times. “Microsatellite instability” refers to genetic instability in the microsatellite regions. Cancer patients with microsatellite instability classified as being high (MSI-H or MSI-High) frequently exhibit an accumulation of somatic mutations in tumor cells that leads to a range of molecular and biological changes including high tumor mutational burden, increased expression of neoantigens and abundant tumor-infiltrating lymphocytes. Chang et al. “Microsatellite Instability: A Predictive Biomarker for Cancer Immunotherapy,” Appl Immunohistochem Mol Morphol, 26(2):e15-e21 (2018). These changes have been linked to increased sensitivity to checkpoint inhibitor drugs, such as pembrolizumab, which is used to treat advanced melanoma, head and neck squamous cell carcinoma, non-small cell lung cancer (NSCLC), and classical Hodgkin lymphoma.

[0224] A viral status test refers to a test that identifies the presence of viral RNA or DNA in a subject. The test can identify viral load and / or viral identity. For example, the viral status test can identify the presence of viral RNA or DNAFMI Docket No.0032-P L&H Docket No.: F171-0008PCT associated with the occurrence of certain cancers. Examples of such viruses include Hepatitis B Virus (HBV) and Hepatitis C Virus (HCV), Kaposi Sarcoma-Associated Herpesvirus (KSHV), Merkel Cell Polyomavirus (MCV), Human Papillomavirus (HPV), Human Immunodeficiency Virus Type 1 (HIV-1, or HIV), Human T-Cell Lymphotropic Virus Type 1 (HTLV-1), and Epstein-Barr Virus (EBV).

[0225] Cancer “hotspot” mutations give rise to oncological outcomes. PhyloP, SIFT, Grantham, COSMIC and PolyPhen-2 are in silico tools that can be used to assess pathogenicity of identified variants. Exemplary hotspot genes and mutations include EGFR exon 19 activating mutation, EGFR exon 19 deletion, EGFR exon 19 insertion, EGFR exon 19 sensitizing mutation, EGFR exon 20 activation mutation, EGFR exon 20 insertion, EGFR G719 mutation, EGFR L858R mutation, EGFR L861 mutation, EGFR S768 mutation, EGFR T790M mutation, C797 mutation, KIT activating mutation, KRAS activating mutation, MET activating mutation, NRAS activating mutation, PMS2 promoter mutations, among many others. Hotspot mutations also occur in the following genes: AKT2, BRCA1, BRCA2, ERC1, NSD1, POLH, PPM1G, PTEN, RAD18, RAD51, RAD51B, RB1, TERT, TP53, TP53Bp1, ALK, ARMT1, ATAD5, ATG7, ATIC, AXL, BIRC6, BRD3, BRD4, CAPRIN1, CCAR2, CCDC6, CDK5RAP2, CHD9, CIT, CTNNB1, CUL1, EBF1, EIF3E, HIP1, HMGA2, IRF2BP2, NOTCH1, NOTCH4, NPM1, OFD1, TACC1, TACC3, TERF2, TMEM106B, UBE2L3, USP10, WRDR48, 7AP1, ZEB2, and ZMYND8.

[0226] A “DNA methylation test” refers to an assay, which can be commercially available, for distinguishing methylated versus unmethylated cytosine loci in DNA. Techniques for measuring cytosine methylation include bisulfite- based methylation assays. The addition of bisulfite to DNA results in the methylation of unmethylated cytosine and its ultimate conversion to the nucleotide uracil. Uracil has similar binding properties to thiamine in the DNA sequence. Previously methylated cytosine does not undergo similar chemical conversion on exposure to bisulfite. Bisulfite assays can thus be used to discriminate previously methylated versus unmethylated cytosine.

[0227] An exemplary quantitative methylation detection assay combines bisulfite treatment and restriction analysis COBRA, which uses methylation sensitive restriction endonucleases, gel electrophoresis, and detection based on labeled hybridization probes. (Ziong and Laird, Nucleic Acid Res.199725; 2532-4). Another exemplary detection assay is the methylation specific polymerase chain reaction PCR (MSPCR) for amplification of DNA segments of interest. This assay can be performed after sodium bisulfite conversion of cytosine and uses methylation sensitive probes. Other detection assays include the Quantitative Methylation (QM) assay, which combines PCR amplification with fluorescent probes designed to bind to putative methylation sites; MethyLightTM(Qiagen, Redwood City, CA) a quantitative methylation detection assay that uses fluorescence-based PCR (Eads, et al., Cancer Res.1999; 59:2302- 2306); and Ms-SNuPE, a quantitative technique for determining differences in methylation levels in CpG sites. As with other techniques, Ms-SNuPE also requires bisulfite treatment to be performed first, leading to the conversion of unmethylated cytosine to uracil while methyl cytosine is unaffected. PCR primers specific for bisulfite converted DNA are then used to amplify the target sequence of interest. The amplified PCR product is isolated and used to quantitate the methylation status of the CpG site of interest. (Gonzalgo and Jones Nuclei Acids Res1997; 25:252-31).

[0228] In particular embodiments, pyrosequencing can be used to detect marker methylation. Pyrosequencing is a method of DNA sequencing that relies on detection of the release of pyrophosphates as DNA is synthesized (and is therefore a “sequencing by synthesis” technique). To assess methylation by pyrosequencing, a DNA sample can be incubated with sodium bisulfite, converting unmethylated cytosine to uracil. The presence of uracil will result in thymineFMI Docket No.0032-P L&H Docket No.: F171-0008PCT incorporation during PCR amplification. Therefore, sequencing results that include thymine at a nucleotide position that is known to encode cytosine can be interpreted as unmethylated sites. In contrast cytosines present in the sequencing results indicate that the site was methylated in the original DNA sample, because methylation protects cytosine from conversion to uracil upon treatment. Bisulfite treatment can also be performed on control samples with known methylation patterns, to reduce or eliminate false positive results. Commercially available pyrosequencing machines include Pyro Mark Q96 (Qiagen, Hilden, Germany). For more details on methods to use pyrosequencing for measurement of methylation, see Delaney et al. Methods Mol Biol.20151343: 249-264. Pyrosequencing is especially useful for detecting methylation in the CpG sites within genes.

[0229] In particular embodiments, a protein marker is detected by contacting a sample with reagents (e.g., antibodies), generating complexes of reagent and marker(s), and detecting the complexes. Particular embodiments for detecting and measuring protein levels can use methods including agglutination, chemiluminescence, electro- chemiluminescence (ECL), enzyme-linked immunoassays (ELISA), immunoassay, immunoblotting, immunodiffusion, immunoelectrophoresis, immunofluorescence, immunohistochemistry, immunoprecipitation, mass-spectrometry, and western blot. See also, e.g., E. Maggio, Enzyme-Immunoassay (1980), CRC Press, Inc., Boca Raton, Fla; and U.S. Pat. Nos.4,727,022; 4,659,678; 4,376,110; 4,275,149; 4,233,402; and 4,230,797.

[0230] Read depth refers to the number of times that a specific genomic site is sequenced during a sequencing run.

[0231] Certain implementations are described herein, including the best mode known to the inventors for carrying out implementations of the disclosure. Of course, variations on these described implementations will become apparent to those of ordinary skill in the art upon reading the foregoing description. The inventor expects skilled artisans to employ such variations as appropriate, and the inventors intend for implementations to be practiced otherwise than specifically described herein. Accordingly, the scope of this disclosure includes all modifications and equivalents of the subject matter recited in the claims appended hereto as permitted by applicable law. Moreover, any combination of the above-described elements in all possible variations thereof is encompassed by implementations of the disclosure unless otherwise indicated herein or otherwise clearly contradicted by context.

Claims

FMI Docket No.0032-P L&H Docket No.: F171-0008PCT CLAIMS What is claimed is:

1. A method, comprising: identifying data indicative of DNA from a sample derived from a subject; generating, based on the data, a fragment image indicating a plurality of fragments among the DNA corresponding to a region of a genome; inputting input data comprising the fragment image into a model configured to generate a prediction of whether the subject has at least one cancer; and generating a report based on the prediction.

2. The method of claim 1, wherein the DNA comprises cell-free DNA (cfDNA, and wherein the DNA comprises at least one fragment having a length in a range of about 1 base to about 500 bases.

3. The method of claim 1, wherein the sample comprises a liquid biopsy sample.

4. The method of claim 1, wherein the fragment image comprises: a plurality of rows respectively corresponding to the plurality of fragments in the DNA; and a plurality of columns respectively corresponding to a plurality of positions within the region of the genome.

5. The method of claim 4, wherein the plurality of rows in the fragment image are ordered based on start positions, middle positions, or end positions of the plurality of fragments within the region of the genome.

6. The method of claim 1, wherein an example pixel is in a row of the fragment image corresponding to an example fragment among the plurality of fragments at an example genomic position among a plurality of positions within the region of the genome.

7. The method of claim 6, wherein a first channel of the example pixel indicates a presence of a base pair in the example fragment at the example genomic position, wherein a second channel of the example pixel indicates a presence of a CG dinucleotide in the example fragment at the example genomic position, and wherein a third channel of the example pixel indicates a presence of a methylated CG dinucleotide in the example fragment at the example genomic position.

8. The method of claim 6, wherein a channel of the example pixel indicates a presence of a variant in the example fragment at the example genomic position, and / or wherein the variant comprises a difference between a dinucleotide at the example genomic position and a dinucleotide in at least one reference sequence.

9. The method of claim 6, wherein a channel of the example pixel indicates a presence of a dinucleotide among a predetermined sequence in the example fragment at the example genomic position, wherein the predetermined sequence comprises at least one of a promoter, an enhancer, a hotspot, a CpG island, an exon, or a gene.

10. The method of claim 1, wherein the model comprises at least one machine learning (ML) model comprising: at least one convolutional neural network (CNN) comprising a plurality of convolutional layers; and / or at least one multi-instance learning (MIL) model.FMI Docket No.0032-P L&H Docket No.: F171-0008PCT 11. The method of claim 10, further comprising: training the at least one ML model by optimizing parameters of the at least one ML model based on training data, the training data comprising example fragment images identified from example samples of a population, wherein the population comprises at least one first individual with cancer and at least one second individual without cancer, wherein the at least one first individual comprises: at least one third individual having a first type of cancer; and at least one fourth individual having a second type of cancer, and wherein the prediction of whether the subject has at least one cancer comprises a prediction of whether the subject has the first type of cancer and a prediction of whether the subject has the second type of cancer.

12. The method of claim 10, wherein the at least one ML model comprises: a first ML model configured to generate a first prediction that the subject has a first type of cancer; and a second ML model configured to generate a second prediction that the subject has a second type of cancer, and wherein the method further comprises: identifying example fragment images of example DNA in example samples obtained from a population; identifying first labels indicating whether the population has the first type of cancer; identifying second labels indicating whether the population has the second type of cancer; training the first ML model based on first training data comprising: the example fragment images; and the first labels; and training the second ML model based on second training data comprising: the example fragment images; and the second labels.

13. The method of claim 1, wherein the prediction of whether the subject has at least one cancer comprises a prediction of whether the subject has a type of cancer comprising at least one of an anatomic location of a tumor, a histological cancer type of the tumor, a primary site of a primary tumor, a tissue origin of the tumor, a subtype of at least one cancer cell of the subject, whether the subject qualifies for a clinical trial, or a metastasis profile.

14. The method of claim 13, wherein the DNA comprises ctDNA from a secondary tumor.

15. The method of claim 13, wherein the type of cancer indicates whether the tumor is resistant to a predetermined therapy or responsive to a predetermined therapy, wherein the predetermined therapy comprises at least one of an immunotherapy, a chemotherapy, or a radiotherapy.

16. The method of claim 1, further comprising: identifying data indicative of additional biomarkers of the subject, wherein the input data further comprises the data indicative of the additional biomarkers of the subject, wherein the additional biomarkers comprise at least one of results from: a histological study; whole transcriptome sequencing; cfRNA sequencing; whole exome sequencing; whole genome sequencing; a cancer hotspot panel test;FMI Docket No.0032-P L&H Docket No.: F171-0008PCT a DNA methylation test; a DNA fragmentation test; an RNA fragmentation test; a microsatellite instability (MSI) test; a tumor mutational burden (TMB) test; or a viral status test.

17. A system, comprising: at least one processor; and memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising: identifying data indicative of DNA from a sample derived from a subject; generating, based on the data, a fragment image indicating a plurality of fragments among the DNA corresponding to a region of a genome; inputting input data comprising the fragment image into a model configured to generate a prediction of whether the subject has at least one cancer; and generating a report based on the prediction.

18. A method, comprising: providing a plurality of nucleic acid molecules obtained from a sample from a subject; extracting, from the sample, a plurality of cell-free DNA (cfDNA) molecules; ligating one or more adapters onto one or more of the cfDNA molecules; amplifying the one or more ligated cfDNA molecules; capturing all or a subset of the amplified cfDNA molecules; sequencing, by a sequencer, all or a subset of the captured cfDNA molecules to obtain a plurality of sequence reads that represent the captured cfDNA molecules; receiving, at one or more processors, sequence read data for the plurality of sequence reads; generating, using the one or more processors, a fragment image indicating at least a portion of the cfDNA molecules corresponding to a region of a genome, wherein an example row of the fragment image corresponds to a fragment among the cfDNA molecules represented in the sequence read data, a first channel of an example pixel of the example row indicating a presence of a base in the fragment at a particular position in the region, a second channel of the example pixel indicating whether the base comprises a cytosine, a third channel of the example pixel indicating whether the base comprises a methylated cytosine; inputting input data comprising the fragment image into at least one model configured to generate a probability that the subject has at least one cancer; and generating, using the one or more processors, a report based on the probability that the subject has at least one cancer.

19. The method of claim 18, wherein the fragment image comprises a plurality of rows respectively corresponding to a plurality of fragments among the cfDNA molecules, and wherein the plurality of rows in the fragment image are ordered based on middle positions of the plurality of fragments in the region.

20. The method of claim 18, wherein the region comprises a genomic interval whose methylation status is indicative of one or more of a tissue-of-origin of the at least one cancer, a prognosis of the at least one cancer, or a therapy response of the at least one cancer.

Images