Machine learning models for predicting image-like frames from sequence data

Abstract

Presented herein are systems and methods of generating frames across time from sequence data using machine learning (ML) models. A computing system coupled with memory can retrieve a sequence dataset comprising a plurality of identifiers generated using a gene segment from a subject at a first time. Each identifier of the plurality of identifiers can have an alphanumeric character indicating a base type at a respective position in the gene segment. The computing system can convert the plurality of the identifiers of the sequence dataset to generate a first frame comprising a first plurality of coordinates for the gene segment at the first time. The computing system can provide the first frame as an input to a ML model. The computing system can generate, based on providing the input to the ML model, a second frame comprising a second plurality of coordinates for the gene segment at a second time.

Description

MACHINE LEARNING MODELS FOR PREDICTING IMAGE-LIKE FRAMES FROMSEQUENCE DATACROSS REFERENCE TO RELATED APPLICATIONS

[0001] The present application claims priority to U.S. Provisional Application No. 63 / 636,578, filed April 19, 2024, which is incorporated herein by reference in its entirety.BACKGROUND

[0002] A computing device may use computer vision to process a digital image to detect various features within the image.SUMMARY

[0003] Aspects of the present disclosure are directed to systems, methods, devices, and non-transitory computer readable media for generating frames across time from sequence data using machine learning (ML) models. One or more processors coupled with memory can retrieve a sequence dataset comprising a plurality of identifiers generated using a gene segment from a subject at a first time. Each identifier of the plurality of identifiers can have an alphanumeric character indicating a base type of a plurality of base types at a respective position of a plurality of positions in the gene segment. The one or more processors can convert the plurality of the identifiers of the sequence dataset to generate a first frame comprising a first plurality of coordinates for the gene segment at the first time. The one or more processors can provide the first frame as an input to a ML model. The ML model can be established using a plurality of examples, each of the plurality of examples comprising a respective plurality of example frames for a respective gene segment across a corresponding plurality of times. The one or more processors can generate, based on providing the first frame dataset as the input to the ML model, a second frame comprising a second plurality of coordinates for the gene segment at a second time. The one or more processors can store, using one or more data structures, an association between the subject and the second frame for the gene segment at the second time.[0004| In some embodiments, the subject can be at risk of or diagnosed with cancer. The one or more processors can determine a score indicating a degree of progression of mutation associated with the cancer in the subject based on the first frame at the first time and the second frame at the second time. The one or more processors can generate a classification for the subject in accordance a comparison between the score and a threshold. In some embodiments, the one or more processors can generate the classification identifying the subject as a candidate for administration of therapy for the cancer, responsive to the score satisfying the threshold. The one or more processors can provide an output indicating the subject as the candidate for administration of the therapy for the cancer. In some embodiments, the subject can be administered with a therapeutically effective dosage of the therapy for the cancer subsequent to provision of the output. The therapy can include at least one of immunotherapy, chemotherapy, radiotherapy, or a targeted therapy. In some embodiments, the one or more processors can generate the classification identifying the subject as a non-candidate of administration of therapy for the cancer, responsive to the score not satisfying the threshold. The one or more processors can provide an output indicating the subject as the non-candidate for administration of the therapy for the cancer. The subject can be administered with at least one of no therapy or a previous therapy for the cancer subsequent to provision of the output.

[0005] In some embodiments, the one or more processors can identify, for each identifier of the plurality of identifiers of the sequence dataset, one of a presence or an absence of a mutation at the respective position of the plurality of positions in the gene segment. The plurality of mutations can include at least one of a copy number alteration, a deletion, a duplication, an inversion, an insertion, or a translocation. The one or more processors can convert the sequence dataset to generate the first frame comprising the first plurality of coordinates. Each coordinate of the first plurality of coordinates can have a respective first set of channel values identifying the base type and one of the presence or the absence of the mutation at the respective position of the plurality of positions in the gene segment at the first time. The one or more processors can generate, based on providing the input to the ML model, the second frame comprising a second plurality of coordinates for the gene segment at the second time,Each coordinate of the second plurality of coordinates can have a respective second set of channel values identifying the base type and one of the presence or the absence of the mutation at the respective position of the plurality of positions in the gene segment at the second time.

[0006] In some embodiments, the one or more processors can determine a plurality of variant allele frequencies (VAFs) corresponding to the plurality of identifiers of the sequence dataset. Each of the plurality of VAFs may identify a frequency of a variant allele at the respective position of the plurality of positions in the gene segment. The one or more processors can generate, using the plurality of VAFs, a third frame comprising a third plurality of coordinates for the gene segment at a third time. The one or more processors can provide the first frame and third frame as input to the input ML model to generate the second frame. In some embodiments, the one or more processors can generate a frame dataset comprising a plurality of frames corresponding to a plurality of times. The plurality of frames can include (i) the first frame corresponding to the first time of the plurality of times and (ii) a null frame as a placeholder for the second frame corresponding to the second time of the plurality of times. A time span between the first time and the second time ranges between 1 day and 5 years.

[0007] In some embodiments, the ML model may be established by identifying, from the respective plurality of frames of an example of the plurality of examples, a first example frame comprising a first example plurality of coordinates. Each of the first example plurality of coordinates having a corresponding first example set of channel values at a corresponding position of a plurality of positions in the respective gene segment. The ML model may be established by modifying the first example frame by masking one or more of the respective plurality of coordinates to generate a second example frame comprising a second example plurality of coordinates, one or more of the second example plurality of coordinates having null values. The ML model may be established by providing the second example frame as input to the ML model to generate a third example frame comprising a third example plurality of coordinates. The ML model may be established by determining a loss metric based on a comparison between the first example plurality of coordinates of the first frame and the thirdexample plurality of coordinates of the third frame. The ML model may be established by updating one or more weights of the ML model in accordance with the loss metric.

[0008] In some embodiments, the subject can be at risk of or diagnosed with cancer comprising at least one of skin cancer, lung cancer, brain cancer, bone cancer, leukemia, lymphoma, head and neck cancer, colon cancer, rectal cancer, uterine cancer, endometrial cancer, pancreatic cancer, thyroid cancer, stomach cancer, ovarian cancer, cervical cancer, bladder cancer, or breast cancer. The gene segment can be obtained via a sample from the subject, wherein the sample comprises at least one of a skin, lung, brain, bone, bone marrow, lymph node, head, neck, colon, rectum, uterus, endometrium, pancreas, thyroid, stomach, ovary, cervix, bladder, breast, urine, plasma, or blood.BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The foregoing and other objects, aspects, features, and advantages of the disclosure will become more apparent and better understood by referring to the following description taken in conjunction with the accompanying drawings, in which:

[0010] FIG. 1 depicts a graph of a tumor mutation profile over time.[0011 | FIG. 2 depicts a block diagram of transformation of sequence data corresponding to a TP53 gene with 19,000 base pairs to frame data over 48 states or times.

[0012] FIG. 3 depicts a block diagram of a frame dataset across genomic coordinates, with 32 known frames and 16 null frames that are to be generated.

[0013] FIGs. 4A and 4B depicts block diagram of a hierarchical multi-scale approach for predicting frames. FIG. 4A shows generating additional frames using attention maps and knowledge distillation. FIG. 4B shows selection of the most significant portion of the attention map.[0014| FIG. 5 depicts a block diagram of a two-dimensional image frame data representation of a genomic profile across time.

[0015] FIG. 6 depicts a block diagram of an encoding of each coordinate in the frame data representation of the genomic profile.

[0016] FIG. 7 depicts a block diagram of an auto-encoder model architecture for reconstructing and predicting frame data from a partially masked frame data during training.|0017] FIG. 8A depicts a block diagram of a vision transformer (ViT) model used as an encoder in the auto-encoder model for predicting frame data.10018] FIG. 8B depicts a block diagram of a transformer encoder in the vision transformer (ViT) model.

[0019] FIG. 9A depicts a block diagram of a direct scaling strategy to represent genomic profile data using image frame data.

[0020] FIG. 9B depicts a block diagram of a multi-stage scaling strategy to represent genomic profile data using image frame data.

[0021] FIG. 10 depicts a block diagram of a system of generating frames across time from sequence data using machine learning (ML) models, in accordance with an illustrative embodiment.

[0022] FIGs. 11 A and 1 IB depict block diagrams of a process for training machine learning models in the system to generate frames, in accordance with an illustrative embodiment.

[0023] FIGs. 12A and 12B depict block diagrams of a process for applying machine learning models to predict frames in the system to generate frames, in accordance with an illustrative embodiment.[0024| FIG. 13 depicts a block diagram of a process for providing outputs in the system to generate frames, in accordance with an illustrative embodiment.

[0025] FIG. 14 depicts a flow diagram of a method of generating frames across time from sequence data using machine learning (ML) models, in accordance with an illustrative embodiment.

[0026] FIG. 15 depicts a block diagram of a server system and a client computer system, in accordance with one or more implementations.DETAILED DESCRIPTION

[0027] Following below are more detailed descriptions of various concepts related to, and embodiments of, systems and methods for generating frames across time from sequence data using machine learning (ML) models. It should be appreciated that various concepts introduced above and discussed in greater detail below may be implemented in any of numerous ways, as the disclosed concepts are not limited to any particular manner of implementation. Examples of specific implementations and applications are provided primarily for illustrative purposes.

[0028] Section A describes predicting tumor genetic evolution with image-like data representation for transformer networks.[0029| Section B describes systems and methods for generating frames across time from sequence data using machine learning (ML) models.

[0030] Section C describes a network environment and computing environment which may be useful for practicing various computing related embodiments described herein.A. Predicting Tumor Genetic Evolution with Image-Like Data Representation for Transformer Networks

[0031] The present disclosure addresses the challenge of predicting future tumor genetic alterations and proposes an innovative image-like data representation to overcome theselimitations, employing a 3D array where genomic coordinates and time changes are represented similarly to image dimensions, and copy numbers and indels are encoded in five “color” channels. Transformer networks for capturing complex data dependencies are effective in DNA encoding but fall short when integrating genomic alterations, and temporal information. This approach demonstrates the potential of using single nucleotide resolution genome prediction to develop personalized immunotherapies and tailor treatment strategies, significantly advancing cancer treatment planning.

[0032] Referring to FIG. 1, depicted is a graph of a tumor mutation profile over time. As shown, tumors vary significantly in their evolution, with some being highly aggressive or evading immune detection. Predicting future genetic alterations in tumors is crucial for planning treatment strategies and developing immunotherapies to prevent immune escape. Referring to FIG. 2, depicted is a block diagram of transformation of sequence data corresponding to a TP53 gene with 19,000 base pairs to frame data over 48 states or times. TP53, a medium-length gene, is 19,000 bases long. If this is represented in a temporal framework with 48 time-frames, the equivalent representation may be equivalent to an image with a resolution of 896x896 pixels.|0033| Referring to FIG. 3, depicted is a block diagram of a frame dataset across genomic coordinates, with 32 known frames and 16 null frames that are to be generated. The 16 null frame may correspond to genomic profile data that are to be predicted. Referring to FIGs. 4A and 4B, depicted are block diagram of a hierarchical multi-scale approach for predicting frames. FIG. 4A shows generating additional frames using attention maps and knowledge distillation. FIG. 4B shows selection of the most significant portion of the attention map.Twenty -thousand base pair (bps) may be fitted into the context window. Then, using knowledge distillation, an attention map may be obtained for the selected region. This process may be repeated over the entire gene.

[0034] Referring to FIG. 5, depicted is a block diagram of a two-dimensional image frame data representation of a genomic profile across time. The data representation may be a 2D format, where the first dimension corresponds to the genomic coordinate, and the second to time(or state). Each subsequent profile may be located below the previous one. This format offers 2 advantages: capacity to encode diverse alterations across various channels for a given coordinate and ability to capture spatio-temporal dependencies through computer vision algorithms.Referring to FIG. 6, depicted is a block diagram of an encoding of each coordinate in the frame data representation of the genomic profile. The coding of DNA structural elements is 5 channels with one-hot encoding, as well as multipliers for the number of copies.

[0035] Referring to FIG. 7, depicted is a block diagram of an auto-encoder model architecture for reconstructing and predicting frame data from a partially masked frame data during training. The algorithm may be a masked autoencoder model, which is a scalable selfsupervised learner for computer vision. Its approach involves masking random patches of the image and reconstructing the missing pixels. From training, the model may be embedded with local context (e.g., nucleotide- wise context, such as single nucleotide variants or motifs) and global context (e.g., inter-genic context, such as gene networks and germline variants, and genomic level context, such pathways and chromosomal alterations). Referring to FIG. 8A depicts a block diagram of a vision transformer (ViT) model used as an encoder in the autoencoder model for predicting frame data. Referring to FIG. 8B depicts a block diagram of a transformer encoder in the vision transformer (ViT) model. The encoder may be a Visual Transformer (ViT) that treats images as sequences of non-overlapping patches represented as tokens, which are flattened and processed by self-attention mechanisms.|0036| Referring to FIG. 9A, depicted is a block diagram of a direct scaling strategy to represent genomic profile data using image frame data. As depicted, one gene may be represented with a 1000 pixel embedding in the model thereby expanding the input data proportionally to the full set of genes. However, this method may be too resource-intensive, especially for large-scale deployment. 25,000 genes may be represented with a 400,000 pixel in the embedding. Referring to FIG. 9B, depicted is a block diagram of a multi-stage scaling strategy to represent genomic profile data using image frame data. To address this, proposed is an alternative strategy that uses a multi-stage approach. In this method, the data may be compressed into a latent space using neural networks, analyzed, and then decompressed backinto the original format through up-sampling. As shown, a latent space genome map may correspond to different genes within the model. A reconstructed frame representation may be generated using the latent space genome map, and then up-sampled to nucleo-tide wise scale. This multi-stage process may allow for the extraction of key information while reducing computational load. The latent space may be represented by the model’s attention map, from which genomically important regions for reconstruction can be identified.

[0037] The method surpasses other techniques by integrating copy number variations and indel information, representing these data through a technique akin to color channels in images. A temporal aspect is introduced by stacking sequential genome sequences across 48 time frames. Employing the Visual Transformer with self-distillation with no labels (DINO) technique on the TP53 gene of 1902 patients from the PCAWG dataset resulted in an attention map used to create an “unsparse” gene sequence of the most 5% base pairs variable in terms of structure and time. The next 15 out of 33 time frames may be predicted as an image inpainting problem to identify future genomic sequences. The masked autoencoder utilized for this task was able to recover ground truth structure, achieving a structural similarity index of 0.79. This approach demonstrates the potential of using single nucleotide resolution genome prediction to develop personalized immunotherapies and tailor treatment strategies, significantly advancing cancer treatment planning.B. Systems and Methods for Generating Frames Across Time from Sequence Data Using Machine Learning (ML) Models|0038| Statistical models may be used to predict the expected gene mutation, in the context of genomics and evolutionary modeling particularly when evaluating cancer. Cancer may be characterized by the proliferation and survival of abnormal cells, particularly in oncogenes, tumor suppressor genes, and repair genes. To predict the mutations, models may rely on population-level data to deduce mutation frequencies in the genomic sequence data. While these methods can provide certain insights, they may be severely limited for the prediction of gene mutations due to a number of reasons. First, statistical models may generally operate underthe assumption of independence among mutational events and genomic locations, as these models may not factor in contextual information and relational interdependence across events and locations. Second, these approaches may lack sufficient resolution at the genomic location level and at the temporal level to capture temporal dynamics of tumor evolution. Third, statistical-based methods may be heavily dependent on existing datasets, and consequently may be prone to biased outputs and may fail to generalize to alterations not found in the original dataset.

[0039] To address these and other technical challenges, the genomic sequence data may be converted to an image-like matrix representation, including genomic coordinates and time changes, with nucleotide base type and mutation type encoded into each coordinate. The matrix representation may be used in conjunction with a machine learning (ML) model, along with a hierarchical multi-scale approach. Under this approach, the ML model may generate additional frames for the matrix representation using attention maps and knowledge distillation. The use of the matrix representation and the ML model may capture complex dependencies (both at the local and global context level) and spatio-temporal relationships that the statistics-based approaches cannot. Furthermore, the ML model may be trained to find certain latent features within the matrix representation that correlate to future mutations. This can allow the ML model to determine and predict mutations that are not found in the existing training data. The prediction of future genetic mutations using the image-like representation and the ML model may also be used to plan personalized immunotherapies and tailor treatment strategies.

[0040] Referring now to FIG. 10, depicted is a block diagram of a system 100 of generating frames across time from sequence data using machine learning (ML) models. In brief overview, the system 100 may include at least one data processing system 105, at least one genomic profiler 110, and at least one administrative device 115, among others, communicatively coupled via at least one network 120. The data processing system 105 may include at least one dataset indexer 125, at least one variation analyzer 130, at least one frame constructor 135, at least one model trainer 140, at least one model applier 145, at least one output evaluator 150, at least one machine learning (ML) model 155 including at least one encoder 160and at least one decoder 165, and at least one database 170, among others. Each of the components in the system 100 as detailed herein may be implemented using hardware (e.g., one or more processors coupled with memory), or a combination of hardware and software as detailed herein in Section C. The system 100 may be used to carry out the functionalities detailed herein in Section A.|00411 In further detail, the data processing system 105 may (sometimes herein generally referred to as a computing system or a server) be any computing device including one or more processors coupled with memory and software and capable of performing the various processes and tasks described herein. The data processing system 105 can be in communication with the genomic profiler 110, the administrative device 115, the database 170, and other devices, via the network 120. The data processing system 105 may be situated, located, or otherwise associated with at least one server group. The server group may correspond to a data center, a branch office, or a site at which one or more servers corresponding to the data processing system 105 is situated.

[0042] The data processing system 105 may include one or more subsystems, modules, or components to executing the various processes and tasks detailed herein. On the data processing system 105, the dataset indexer 125 may retrieve a sequence dataset corresponding to a gene segment in a sample from a subject via the genomic profiler 110 or the database 170. The variation analyzer 130 may identify mutations within the sequence dataset. The frame constructor 135 may convert the sequence dataset to an initial frame dataset. The model trainer 140 may initialize, train, and establish the ML model 155 using a training dataset. The model applier 145 may provide as incoming frame datasets to the ML model 155 to generate predicted frames. The output evaluator 150 may provide an output base on the frame dataset including the predicted frames to the administrative device 115.

[0043] The ML model 155 (sometimes herein referred to as a generative model) may be any type of artificial intelligence (Al) algorithm or ML architecture to generate predicted frames using the initial frame dataset. The ML model 155 may include, for example, a deep learningartificial neural network (ANN), such as a transformer architecture, an encoder-decoder model with a convolution neural network architecture, a diffusion model, or an autoencoder model, among others. The ML model 155 may be, for example, an instance of an autoencoder (e.g., as depicted in FIG. 7). In some embodiments, the ML model 155 may include the deep-learning ANN, and one or more of: a clustering algorithm, a support vector machine (SVM), a decision tree, a Bayesian model, or a regression model, among others. In general, the ML model 155 may include inputs and outputs related to one another via a set of weights. The set of weights may be in accordance with the Al algorithm or ML model (e.g., transformer models) used to implement the architecture for the ML model 155. For example, the set of weights of the ML model 155 may be in accordance with one or more transformer models, each with encoder layers, decoder layers, feed-forward layers, cross-attention layers, and activation layers, among others. The set of weights of the ML model 155 may be distributed or arranged across the encoder 160 and the decoder 165, among others.

[0044] In the ML model 155, the encoder 160 may be any type of Al algorithm or ML model to generate embeddings identifying contextual information using input tokens derived from the genomic data. The encoder 160 may include, for example, a visual transformer network (e.g., as depicted in FIGs. 8A and 8B), a self-attention encoder, a bidirectional encoder, a sparse attention encoder, or a convolutional encoder (e.g., a convolutional neural network (CNN)), among others. The encoder 160 may include inputs and outputs related to one another via a set of weights. The set of weights may be in accordance with an architecture used to implement the encoder 160. In addition, the decoder 165 may be any type of Al algorithm or ML model to generate additional predicted frames using the embeddings outputted by the encoder 160. The decoder 165 may include, for example, an autoregressive decoder, a bidirectional decoder, a non- autoregressive decoder, a denoising decoder, a sparse attention decoder, a cross-attention decoder, or a convolutional decoder (e.g., CNN), among others. The decoder 165 may include inputs and outputs related to one another via a set of weights. The set of weights may be in accordance with an architecture used to implement the decoder 165.[0045| The genomic profiler 110 may be any device to perform genetic sequencing on a gene segment (e.g., deoxyribonucleic acid (DNA) or ribonucleic acid (RNA) sample) in a sample taken from a subject and to generate sequencing datasets using the genetic sequencing. The genetic sequencing carried out may be a high throughput, massively parallel sequencing technique (sometimes herein referred to as next-generation sequencing), such as whole genome sequencing (WGS), pyrosequencing, Reversible dye-terminator sequencing, SOLiD sequencing, Ion semiconductor sequencing, Helioscope single molecule sequencing, among others. Using the genetic sequencing data, the genomic profiler 110 may generate genomic dataset profiling the subject. In generating the genomic dataset, the genomic profiler 110 may execute read alignment, variant calling, and gene expression quantification, among others. In some embodiments, the genomic profiler 110 may be part of a genomic profiling platform, such as IMPACT, GENIE, or FoundationOne platform, among others. The genomic profiler 110 may use the gene sequencing to generate a sequencing dataset. The sequencing dataset may lack identifiers for at least a portion of genes (e.g., due to differences in sequencing or acquisition protocols across the different genomic profiling platforms). The sequencing dataset may be maintained using one or more files according to a format (e.g., FASTQ, BAM, SAM, BCL, or VCF formats).

[0046] The gene segment, for which sequence data is generated, include at least one of:AKT Serine / Threonine Kinase 1 (AKT1), Anaplastic Lymphoma Kinase (ALK), Adenomatous Polyposis Coli (APC), Androgen Receptor (AR), A-Raf Proto-Oncogene, Serine / Threonine Kinase (ARAF), AT-Rich Interaction Domain 1 A (ARID1A), AT -Rich Interaction Domain 2 (ARID2), Ataxia Telangiectasia Mutated (ATM), Beta-2-Microglobulin (B2M), B-Cell CLL / Lymphoma 2 (BCL2), BCL6 Corepressor (BCOR), B-Raf Proto-Oncogene, Serine / Threonine Kinase (BRAF), Breast Cancer Type 1 Susceptibility Protein (BRCA1), Breast And Ovarian Cancer Susceptibility Protein 2 (BRCA2), Caspase Recruitment Domain Family Member 11 (CARD 11), Core-Binding Factor Subunit Beta (CBFB), Cyclin DI (CCND1), Cadherin 1 (CDH1), Cyclin Dependent Kinase 4 (CDK4), Cyclin Dependent Kinase Inhibitor 2A (CDKN2A), Capicua Transcriptional Repressor (CIC), CREB Binding Protein (CREBBP),CCCTC-Binding Factor (CTCF), Catenin Beta 1 (CTNNB1), Dicer 1, Ribonuclease III (DICER1), DIS3 Homolog, Exosome Endoribonuclease And 3'-5' Exoribonuclease (DIS3), DNA Methyltransferase 3 Alpha (DNMT3A), Epidermal Growth Factor Receptor (EGFR), Eukaryotic Translation Initiation Factor 1A X-Linked (EIF1AX), E1A Binding Protein P300 (EP300), Erb- B2 Receptor Tyrosine Kinase 2 (ERBB2), Erb-B2 Receptor Tyrosine Kinase 3 (ERBB3), ERCC Excision Repair 2, TFIIH Core Complex Helicase Subunit (ERCC2), Estrogen Receptor 1 (ESRI), Enhancer Of Zeste 2 Polycomb Repressive Complex 2 Subunit (EZH2), F-Box And WD Repeat Domain Containing 7 (FBXW7), Fibroblast Growth Factor Receptor 1 (FGFR1), Fibroblast Growth Factor Receptor 2 (FGFR2), Fibroblast Growth Factor Receptor 3 (FGFR3), Fibroblast Growth Factor Receptor 4 (FGFR4), Fms Related Receptor Tyrosine Kinase 3 (FLT3), Forkhead Box Al (F0XA1), Forkhead Box L2 (F0XL2), Forkhead Box 01 (F0X01), Far Upstream Element Binding Protein 1 (FUBP1), GATA Binding Protein 3 (GATA3), G Protein Subunit Alpha 11 (GNA11), G Protein Subunit Alpha Q (GNAQ), GNAS Complex Locus (GNAS), H3 Histone, Family 3A (H3F3A), Histone Cluster 1, H3b (HIST1H3B), HRas Proto-Oncogene, GTPase (HRAS), Isocitrate Dehydrogenase (NADP(+)) 1 (IDH1), Isocitrate Dehydrogenase (NADP(+)) 2 (IDH2), IKAROS Family Zinc Finger 1 (IKZF1), Inositol Polyphosphate Phosphatase Like 1 (INPPL1), Janus Kinase 1 (JAK1), Lysine Demethylase 6A (KDM6A), Kelch Like ECH Associated Protein 1 (KEAP1), KIT Proto-Oncogene, Receptor Tyrosine Kinase (KIT), Kinetochore Localized Astrin (SPAG5) Binding Protein (KNSTRN), KRAS Proto-Oncogene, GTPase (KRAS), Mitogen- Activated Protein Kinase Kinase 1 (MAP2K1), Mitogen- Activated Protein Kinase 1 (MAPK1), MYC Associated Factor X (MAX), Mediator Complex Subunit 12 (MED 12), MET Proto-Oncogene, Receptor Tyrosine Kinase (MET), MutL Homolog 1 (MLH1), MutS Homolog 2 (MSH2), MutS Homolog 3 (MSH3), MutS Homolog 6 (MSH6), Mechanistic Target Of Rapamycin Kinase (MTOR), MYC ProtoOncogene, BHLH Transcription Factor (MYC), MYCN Proto-Oncogene, BHLH Transcription Factor (MYCN), Myeloid Differentiation Primary Response 88 (MYD88), Myogenic Differentiation 1 (MY0D1), Neurofibromin 1 (NF1), NFE2 Like BZIP Transcription Factor 2 (NFE2L2), Notch Receptor 1 (N0TCH1), NRAS Proto-Oncogene, GTPase (NRAS), Neurotrophic Receptor Tyrosine Kinase 1 (NTRK1), Neurotrophic Receptor Tyrosine Kinase 2(NTRK2), Neurotrophic Receptor Tyrosine Kinase 3 (NTRK3), Nucleoporin 93 (NUP93), P21 (RAC1) Activated Kinase 7 (PAK7), Platelet Derived Growth Factor Receptor Alpha (PDGFRA), Phosphatidylinositol-4,5-Bisphosphate 3-Kinase Catalytic Subunit Alpha (PIK3CA), Phosphatidylinositol-4,5-Bisphosphate 3-Kinase Catalytic Subunit Beta (PZK3CB), Phosphoinositide-3-Kinase Regulatory Subunit 1 (PIK3R1), Phosphoinositide-3 -Kinase Regulatory Subunit 2 (PIK3R2), PMS1 Homolog 2, Mismatch Repair System Component (PMS2) , DNA Polymerase Epsilon, Catalytic Subunit (POLE), Protein Phosphatase 2 Scaffold Subunit Aalpha (PPP2R1A), Protein Phosphatase 6 Catalytic Subunit (PPP6C), Protein Kinase C Iota (PRKCI), Patched 1 (PTCHI), Phosphatase And Tensin Homolog (PTEN), Protein Tyrosine Phosphatase Non-Receptor Type 11 (PTPN11), Rac Family Small GTPase 1 (RAC1), Raf-1 Proto-Oncogene, Serine / Threonine Kinase (RAFI), RB Transcriptional Corepressor 1 (RBI), Ret Proto-Oncogene (RET), Ras Homolog Family Member A (RHOA), Ras Like Without CAAX 1 (RIT1), ROS Proto-Oncogene 1, Receptor Tyrosine Kinase (ROS1), RAS Related 2 (RRAS2), Retinoid X Receptor Alpha (RXRA), SET Domain Containing 2, Histone Lysine Methyltransferase (SETD2), Splicing Factor 3b Subunit 1 (SF3B1), SMAD Family Member 3 (SMAD3), SMAD Family Member 4 (SMAD4), SWI / SNF Related BAF Chromatin Remodeling Complex Subunit ATPase 4 (SMARCA4), SWI / SNF Related BAF Chromatin Remodeling Complex Subunit Bl (SMARCB1), SOS Ras / Rac Guanine Nucleotide Exchange Factor 1 (S0S1), Speckle Type BTB / POZ Protein (SPOP), Signal Transducer And Activator Of Transcription 3 (STAT3), Serine / Threonine Kinase 11 (STK11), Serine / Threonine Kinase 19 (STK19), Transcription Factor 7 Like 2 (TCF7L2), Telomerase Reverse Transcriptase (TERT), Transforming Growth Factor Beta Receptor 1 (TGFBR1), Transforming Growth Factor Beta Receptor 2 (TGFBR2), Tumor Protein P53 (TP53), Tumor Protein P63 (TP63), TSC Complex Subunit 1 (TSC1), TSC Complex Subunit 2 (TSC2), U2 Small Nuclear RNA Auxiliary Factor 1 (U2AF1), Von Hippel-Lindau Tumor Suppressor (VHL), or Exportin 1 (XPO1), among others.

[0047] The administrative device 115 (sometimes herein referred to as a computing device) may be any computing device comprising one or more processors coupled with memory and software and capable of performing the various processes and tasks described herein. Theadmini strati ve device 115 may be in communication with the data processing system 105 and the genomic profiler 110 via the network 120. The administrative device 115 may have at least one display. The administrative device 115 may be associated with an entity (e.g., a clinician) examining the subject or gene data from the subject. The display may present information about the subject provided by the data processing system 105.[0048 The database 170 may store and maintain various resources and data associated with the data processing system 105, the genomic profiler 110, and the administrative device 115, among others. The database 170 may include a database management system (DBMS) to arrange and organize the data maintained thereon. The database 170 may be in communication with the data processing system 105, the genomic profiler 110, and the administrative device 115, via the network 120. While running various operations, the data processing system 105, the genomic profiler 110, and the administrative device 115 may access the database 170 to retrieve identified data therefrom. The data processing system 105, the genomic profiler 110, and the administrative device 115 may also write data onto the database 170 from running such operations.

[0049] Referring now to FIGs. 11 A and 1 IB, depicted are block diagrams of a process 200 for training machine learning models in the system to generate frames. Starting with FIG. 11A, under the process 200, the dataset indexer 125 may retrieve, obtain, or otherwise identify training data 205 for initializing, training, and establishing the ML model 155 from the database 170. The establishment of the ML model 155 may include training, pre-training, or fine-tuning, among others. The training data 205 may identify or include a set of examples with which the ML model 155 is to be trained. In the training data 205, each example may identify or include at least one frame dataset 210 generated using at least one sequence dataset 215 derived from at least one gene segment 220. In some embodiments, at least one example of the training dataset 205 may identify or include the sequence dataset 215, and exclude the frame dataset 210. In some embodiments, at least one example of the training dataset 205 may identify or include the frame dataset 210, and exclude the sequence dataset 205. In some embodiments, at least oneexample of the training dataset 205 may identify or include both the frame dataset 210 and the sequence dataset 215.

[0050] The sequence dataset 215 (sometimes herein referred to as genomic profile) may profile, define, or otherwise characterize genetic makeup of the gene segment 220 from a subject across a set of time samples (or states). The sequence dataset 215 may include one or more sequences 225 A-N (hereinafter generally referred to as sequences 225). The set of sequences 225 may correspond to a set of sampling times at which sequencing was performed. Each sequence 225 may correspond to a respective time at which sequencing was performed on the gene segment 220. The gene segment 220 may be of the same time across the sampling times. The sampling times may range between 1 day and 5 years. The set of sequences 225 may be chronological sequence (or anti-chronological) corresponding to the sampling times. In some embodiments, the sequence dataset 215 or the individual sequences 225 may include or identify metadata about the gene segment 220 or the sequencing of the gene segment 220. For example, the metadata may identify or include a sampling time (e.g., in the form of year, month, day, and time of day) at which the gene sequencing was performed on the gene segment 220 to generate the sequence 225, a length of the set of identifiers in the sequence 225, gene type for the gene segment 220, an identifier (e.g., pseudo-anonymized identifier) for the subject, and a cancer type that the subject from which the gene segment 220 is obtained is at risk of or diagnosed with, among others.|0051| The gene segment 220 may include a set of positions 230A-N (hereinafter generally referred to as positions 230). Each position 230 may correspond to a base pair position (or coordinate) within the gene segment 220. The set of positions 230 may range anywhere from 10,000 to 300,000 base pairs (bps). In the gene segment 220, each position 230 may have at least one nucleotide base, such as adenine (A), cytosine (C), guanine (G), thymine (T), and uracil (U). The sequence 225 may include a set of identifiers 235 A-N (hereinafter generally referred to as identifiers 235) corresponding to the set of positions 230 of the gene segment 220 at the respective time. Each identifier 235 may have at least one alphanumeric character indicating a base type at the respective position 230 in the gene segment 220. For instance, the identifier 235may have the alphanumeric character “A” for the nucleotide base type adenine, “C” for cytosine, “G” for guanine, “T” for thymine, and “U” for uracil, among others.

[0052] The gene segment 220 may be acquired or obtained via a sample from a subject (e.g., via biopsy). The sample may include, for example, sample comprises at least one of a skin, lung, brain, bone, bone marrow, lymph node, head, neck, colon, rectum, uterus, endometrium, pancreas, thyroid, stomach, ovary, cervix, bladder, breast, urine, plasma, or blood, among others. The subject may have been at risk of or diagnosed with cancer. The cancer may include, for example, skin cancer, lung cancer, brain cancer, bone cancer, leukemia, lymphoma, head and neck cancer, colon cancer, rectal cancer, uterine cancer, endometrial cancer, pancreatic cancer, thyroid cancer, stomach cancer, ovarian cancer, cervical cancer, bladder cancer, or breast cancer, among others. Genetic sequencing may be performed (e.g., by the genomic profiler 110) on the gene segment 220 to derive or generate the corresponding sequence 225. Multiple samples may be obtained from the subject across a set of times. Each time, the gene segment 220 of the same type may be obtained from the sample and genetic sequencing may be performed to generate the respective sample for the sequence dataset 215.|0053| The frame dataset 210 may identify or include a set of frames 240A-N (hereinafter generally referred to as frames 240). The set of frames 240 may correspond to the set of sampling time at which sequencing was performed on the gene segment 220. The sequence for the set of frames 240 may correspond to the sequence for the set of sequences 225 of the sequence dataset 215 (e.g., in chronological or anti-chronological order). Each frame 240 may be an image or an arrayed representation of the respective sequence 225 and by extension the gene segment 220 sequenced at the respective sampling time. Each frame 240 may correspond to a respective sequence 225 in the sequence dataset 215. By extension, each frame 240 may correspond to the gene segment 220 at the respective time (e.g., the same respective time as the sequence dataset 215). Each frame 240 may identify or include a set of coordinates 245A-N (hereinafter generally referred to as coordinates 245). The set of coordinates 245 may correspond to the set of identifiers 235 in the respective sequence 225. By extension, the set of coordinates 245 may correspond to the set of positions 230 in the gene segment 220.[0054| In each frame 240, each coordinate 245 may include or have a set of values (sometimes herein referred to as channel values). At each coordinate 245, the set of values may identify or indicate the base type of the nucleotide base and at least one mutation type at the respective position 230 in the gene segment 220. Each value in the coordinate 245 may correspond to a respective channel (e.g., a color channel) to encode information about the gene segment 220 at the respective position 230, and by extension the respective identifier 235 of the corresponding sequence 225. The set of values may be numerical values (e.g., integer, double, floating value) or a Boolean value, among others, for the channels forming the coordinate 245. The set of values may have any number channels. At each coordinate 245, the set of values may include one or more of: at least one channel to indicate presence or absence of the nucleotide base type adenine (A), at least one channel to indicate presence or absence the cytosine (C), at least one channel to indicate presence or absence of guanine (G), at least one channel to indicate presence or absence of thymine (T), or at least one channel to indicate presence or absence of uracil (U), among others, at the respective position 230 of the gene segment 220. In some embodiments, the set of values may include one or more of: at least one channel to indicate presence or absence of a copy number alteration (CNA) (e.g., gain, amplification, loss, or homozygous deletion), at least one channel to indicate presence or absence of deletion, at least one channel to indicate presence or absence of duplication, at least one channel to indicate presence or absence of inversion, at least one channel to indicate presence or absence of insertion, or at least one channel to indicate presence or absence of translocation, among others, at the respective position 230 of the gene segment 220. For instance, at each coordinate 245 of the frame 240, the set of values may be of the form {A, C, G, T, U, CNA, Del., Dup., Inv., Ins., Transloc.}.[0055 In some embodiments, the variation analyzer 130 may process or parse the set of identifiers 235 in each sequence 225 to detect or identify any variants. For each identifier 235 in a given sequence 225, the variation analyzer 130 may determine or identify one of a presence or absence of a mutation at the respective position 230 in the corresponding gene segment 220. The training data 205 may include the sequence dataset 215 and may initially lack the frame dataset210. In some embodiments, the sequence 225 of the sequence dataset 215 may lack metadata indicating presence of the mutation in the corresponding gene segment 220. To derive the frame dataset 210 from the sequence dataset 215, the variation analyzer 130 may traverse through the set of sequences 225 in the sequence dataset 215 to determine or identify mutations. The mutation may include, for example, a copy number alteration (e.g., gain, amplification, loss, or homozygous deletion), a deletion, a duplication, an inversion, an insertion, or a translocation, among others.

[0056] In some embodiments, the variation analyzer 130 may identify the mutation by feeding the set of identifiers 235 of the sequence 225 via variant calling. The variant calling may be part of a detection algorithm or tool, such as a GATK HaplotypeCaller, Mutect2, Manta, Delly, or LUMPY, among others. In general, by using variant calling, the variation analyzer 130 may compare the sequence identifiers 235 of the sequence 225 derived from the gene segment 220 with a reference sequence for a reference gene segment of the same gene type. Based on the comparison, the variant analyzer 130 may identify the presence or absence of the mutation at one or more positions 230 in the gene segment 220. For example, if the base type at the respective position 230 differs from the base type at the same position in the reference gene segment, the variation analyzer 130 may identify the presence of the mutation at the respective position 230. Otherwise, if the base type at the respective position 230 is the same as the base type at the same position in the reference gene segment, the variation analyzer 130 may identify the absence of the mutation at the respective position 230.

[0057] In some embodiments, the frame constructor 135 may transform or convert the sequence dataset 215 to the frame dataset 210. The frame constructor 135 may traverse through the set of sequences 225 in the sequence dataset 215. The frame constructor 135 may transform or convert the sequence 225 to the corresponding frame 240 for the frame dataset 210. In each sequence 225, the frame constructor 135 may traverse through the set of identifiers 235 the sequence dataset 215 to generate the corresponding frame 240 in the frame dataset 210. For each identifier 235, the frame constructor 135 may create, write, or otherwise generate a corresponding coordinate 245 to include the set of values. The set of values of the coordinate245 may identify or include, for example, one or more of: at least one channel to indicate presence or absence of the nucleotide base type adenine (A), at least one channel to indicate presence or absence the cytosine (C), at least one channel to indicate presence or absence of guanine (G), at least one channel to indicate presence or absence of thymine (T), or at least one channel to indicate presence or absence of uracil (U), among others, at the respective position 230 corresponding to the identifier 235.

[0058] In some embodiments, the frame constructor 135 may generate the coordinate 245 using the identification of the mutation at the position 230 in the gene segment 220 corresponding to the identifier 235 in the sequence 225. The set of values at the coordinate 245 may identify or include, for example, at least one channel to indicate presence or absence of a copy number alteration (CNA) (e.g., gain, amplification, loss, or homozygous deletion), at least one channel to indicate presence or absence of deletion, at least one channel to indicate presence or absence of duplication, at least one channel to indicate presence or absence of inversion, at least one channel to indicate presence or absence of insertion, or at least one channel to indicate presence or absence of translocation, among others, at the respective position 230 corresponding to the identifier 235. The set of values may have any number of channels to represent and characterize the gene segment 220 at the respective position 230 associated with the coordinate 245. For instance, the set of values of each coordinate 245 in the frame 240 may be of the form {0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0} for the variables {A, C, G, T, U, CNA, Del., Dup., Inv., Ins., Transloc.}. Since the values at each coordinate 245 represent an encoding the information at the respective position 230 in the gene segment 220, at least one of the values can be set to a nonnull value (e.g., 1 or 100, as opposed to 0). The frame constructor 135 may traverse through the set of sequences 225 in the sequence dataset 215 to generate the set of frames 240 for the frame dataset 210.

[0059] Continuing onto FIG. 1 IB, the model trainer 140 may initialize, train, and establish the ML model 155. The model trainer 140 may initialize the ML model 155 by assigning or setting values (e.g., random values) to the set of weights of the ML model 155, including the encoder 160 and the decoder 165. In training, the model trainer 140 may use atleast a portion of the examples of the training data 205 to modify or update the set of weights in the ML model 155 (including the encoder 160 and the decoder 165) to embed and encode information in the weights. The model trainer 140 may retrieve, obtain, or otherwise identify the frame dataset 210 from each example of the training dataset 205. In some embodiments, the model trainer 140 may extract or identify the set of frames 240 from the frame dataset 210. In the frame dataset 210, each set of frames 240 may have the set or coordinates 245. Each of the coordinates 245 may have the set channel values, with at least one non-null value.

[0060] In some embodiments, the model trainer 140 may use masking to train the ML model 155, including the encoder 160 and the decoder 165. To train in this manner, the model trainer 140 may alter, change, or otherwise modify the frame dataset 210 include one or more masks 250A-N (hereinafter generally referred to as mask 250) to generate at least one masked frame dataset 210’. To generate the masked frame dataset 210’, the model trainer 140 may identify or select at least one frame 240 from the original frame dataset 210 (e.g. at random using a pseudo-random number generator or in accordance with a predefined mask). In the selected frame 240, the model trainer 140 may identify or select one or more coordinates 245 (e.g., at random or predefined mask). For each selected coordinate 245, the model trainer 140 may alter, change, or otherwise modify the set of channel values in the coordinate 245 to null values to add the respective mask 250.

[0061] The model trainer 140 may repeat the selection of the frames 240 and coordinates 245 and modification of the channel values to add masks 250, across the frames 240 of the frame dataset in while generating the masked frame dataset 210’ . The masked frame dataset 210’ may include the set of frames 240. Each frame 240 may include the set of coordinates 245. In the set of coordinates 245 of the frame 240, one or more of the set of coordinates 245 include null channel values (e.g., respective masks 250). Each mask 250 may correspond to the coordinate 245 in which the set of channel values are all null (e.g., {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0} for the variables {A, C, G, T, U, CNA, Del., Dup., Inv., Ins., Transloc.}). Each of the remainder of the coordinates 245 in the frame 240 of the masked frame dataset 210’ may have the same set ofchannel values as the corresponding coordinate 245 in the corresponding frame 240 in the original frame dataset 210.

[0062] To train and establish the ML model 155, the model trainer 140 may apply, feed, or otherwise provide the masked frame dataset 210’ as input to the ML model 155. In some embodiments, the model trainer 140 may feed each frame 240 of the masked frame dataset 210’ (e.g., in sequence) as input to the ML model 155. To provide the input to the ML model 155, the model trainer 140 may feed the masked frame dataset 210’ (or each individual frame 240) to the encoder 160 of the ML model 155. The model trainer 140 may process the input (e.g., masked frame dataset 210’ or each frame) in accordance with the set of weights of the ML model 155 across the encoder 160 and the decoder 165. The encoder 160 may produce, determine, or otherwise generate a set of embeddings 255A-N (hereinafter generally referred to as embeddings 255) using the input. The set of embeddings 255 (sometimes herein referred to as a latent space genome map) may represent latent features within the input, such as context and dependency information. In some embodiments, the set of embeddings 255 may be an attention map. The set of embeddings 255 for the attention map may include a set of attention values. Each attention value may define or characterize a focus (e.g., weighting indicating specified) as assigned the encoder 160 on particular coordinates 245 within the input (e.g., masked frame dataset 210’ or the individual frame 240). The focus may be based on relationship among different coordinates 245 of the input with respect to the target output of the encoder 160 and the overall ML model 155.

[0063] Continuing on, the encoder 160 may feed forward, input, or otherwise provide the set of embeddings 255 as input to the decoder 165. In some embodiments, the encoder 160 may identify or select at least a portion of the set of embeddings 255 to provide as input to the decoder 165, using the attention values of the attention map. For instance, the encoder 160 may select the portion of the set of embeddings 255 corresponding to the highest N (e.g., top 5-20%) of attention values in the attention map. With the selection, the encoder 160 may feed the selected portion of the embeddings 255 to the decoder 165. The decoder 165 may produce, create, or otherwise generate at least one reconstructed frame dataset 210” using the inputembeddings 255 fed by the encoder 160. The reconstructed frame dataset 210” may be the output of the overall ML model 155.

[0064] Based on providing the masked frame dataset 210’ as the input to the ML model 155, the model trainer 140 may produce, create, or otherwise generate the reconstructed frame dataset 210”. The reconstructed frame dataset 210” may be of the same form as the original frame dataset 210 or the masked frame dataset 210’. The reconstructed frame dataset 210” may include a set of frames 240’A-N (hereinafter generally referred to as frames 240’). In some embodiments, based on providing the individual frames 240 from the masked frame dataset 210’ as input to the ML model 155, the model trainer 140 may generate the corresponding frame 240’ for the reconstructed frame dataset 210”. Each frame 240’ may include a set of coordinates 245 ’A-N (hereinafter generally referred to as coordinates 245’). The set of coordinates 245’ in the reconstructed frame dataset 210” may lack any masks 250 or any coordinates 245’ in which all the channel values are set to null. The model trainer 140 may form, create, or otherwise generate the reconstructed frame dataset 210’ to include the set of frames 240’ outputted by the ML model 155 based on providing each individual frame 240 from the masked frame dataset 210’ as input.

[0065] Similar to the original frames 240 of the original frame dataset 210, each coordinate 245’ of the reconstructed frame dataset 210” may include or have a set of channel values. At each coordinate 245’, the set of values may identify or indicate the base type of the nucleotide base and at least one mutation type at the respective position 230 in the gene segment 220. The set of values may be numerical values (e.g., integer, double, floating value) or a Boolean value, among others, for the channels forming the coordinate 245. The set of values at each coordinate 245’ of the reconstructed frame dataset 210” may have any number channels. At each coordinate 245’, the set of values may include one or more of: at least one channel to indicate presence or absence of the nucleotide base type adenine (A), at least one channel to indicate presence or absence the cytosine (C), at least one channel to indicate presence or absence of guanine (G), at least one channel to indicate presence or absence of thymine (T), or at least one channel to indicate presence or absence of uracil (U), among others, at the respectiveposition 230 of the gene segment 220. In some embodiments, the set of values may include one or more of: at least one channel to indicate presence or absence of a copy number alteration (CNA) (e.g., gain, amplification, loss, or homozygous deletion), at least one channel to indicate presence or absence of deletion, at least one channel to indicate presence or absence of duplication, at least one channel to indicate presence or absence of inversion, at least one channel to indicate presence or absence of insertion, or at least one channel to indicate presence or absence of translocation, among others, at the respective position 230 of the gene segment 220.

[0066] The model trainer 140 may calculate, generate, or otherwise determine at least one at least one loss metric 260. The loss metric 260 may be determined by the model trainer 140 based on the comparison between the reconstructed frame dataset 210” generated by the ML model 155 and the original frame dataset 210 of the example from the training dataset 205. The loss metric 260 may identify or indicate a degree of deviation between the expected output as identified in the example of the training data 205 and the output from the ML model 155. In some embodiments, for each coordinate 245’ of the frames 240’ of the reconstructed frame dataset 210’, the model trainer 140 may compare the set of channel values of the coordinate 245’ with the corresponding set of channel values of the corresponding coordinate 245 of the frames 240 of the original frame dataset 210. Based on the comparison, the model trainer 140 may generate the loss metric 260 (e.g., a reconstruction loss). The loss metric 260 (e.g., for the reconstruction loss) may be calculated in accordance with any number of loss functions, such as a norm loss (e.g., LI or L2), mean squared error (MSE), quadratic loss, cross-entropy loss, or Huber loss, among others. In some embodiments, the loss metric 260 (e.g., for the reconstruction loss) may be a degree of similarity calculated in accordance with any number of similarity measures, such as cosine similarity, Jaro distance, Jaccard index, or Dice coefficient, among others.10067] The model trainer 140 may modify or update one or more weights of the ML model 155 using the loss metric 260. The updating of the weights may be in accordance with a back propagation and optimization function (sometimes referred to herein as an objective function) with one or more parameters (e.g., learning rate, momentum, weight decay, andnumber of iterations). The optimization function may define one or more parameters at which the weights of the ML model 155 are to be updated. The optimization function may be in accordance with stochastic gradient descent, and may include, for example, an adaptive moment estimation (Adam), implicit update (ISGD), and adaptive gradient algorithm (AdaGrad), among others. The model trainer 140 can iteratively train the ML model 155 (including the encoder 160 and the decoder 165) until convergence. Upon convergence, the model trainer 140 can store and maintain the set of weights of the ML model 155 (including the encoder 160 and the decoder 165) for use in inference. The ML model 155 may be trained to generate predicted frames representing genetic profiles corresponding to sampling times prior to or subsequent to the sampling times of any of the sequences.

[0068] Referring now to FIGs. 12A and 12B, depicted are block diagrams of a process 300 for applying machine learning models to predict frames in the system to generate frames. Starting with FIG. 12A, under the process 300, the genomic profiler 110 may be used to assess or evaluate a genomic profile for at least one subject 305. The subject 305 may be a human or animal subject. The subject 305 may be at risk of or diagnosed with cancer. At least one sample 310 containing at least one gene segment 315 (e.g., DNA or RNA sample) may be obtained from an organ associated with the cancer from the subject 305 (e.g., via biopsy) at a sampling time (Ts). The sample may include, for example, sample comprises at least one of a skin, lung, brain, bone, bone marrow, lymph node, head, neck, colon, rectum, uterus, endometrium, pancreas, thyroid, stomach, ovary, cervix, bladder, breast, urine, plasma, or blood, among others. The subject may have been at risk of or diagnosed with cancer. The cancer may include, for example, skin cancer, lung cancer, brain cancer, bone cancer, leukemia, lymphoma, head and neck cancer, colon cancer, rectal cancer, uterine cancer, endometrial cancer, pancreatic cancer, thyroid cancer, stomach cancer, ovarian cancer, cervical cancer, bladder cancer, or breast cancer, among others.

[0069] The genetic profiler 110 may execute, carry out, or otherwise perform gene sequencing on the gene segment 315 to create, produce, or otherwise generate at least one sequence dataset 320. The sequence dataset 320 may profile, define, or otherwise characterizegenetic makeup of the gene segment 315 from the sample 310 of the subject 305 at the first sampling time. The sequence dataset 320 may identify or include at least one sequence 325. In some embodiments, the sequence dataset 320 or the individual sequences 325 may include or identify metadata about the gene segment 315 or the sequencing of the gene segment 315. For example, the metadata may include a sampling time (e.g., in the form of year, month, day, and time of day) at which the gene sequencing was performed on the gene segment 315 to generate the sequence 325, a length of the set of identifiers in the sequence 325, gene type for the gene segment 315, an identifier (e.g., pseudo-anonymized identifier) for the subject 305, and a cancer type that the subject 305 is at risk of or diagnosed with, among others.

[0070] The gene segment 315 may include a set of positions 330A-N (hereinafter generally referred to as positions 330). Each position 330 may correspond to a base pair position (or coordinate) within the gene segment 315. The set of positions 330 may range anywhere from 10,000 to 300,000 base pairs (bps). In the gene segment 315, each position 330 may have at least one nucleotide base, such as adenine (A), cytosine (C), guanine (G), thymine (T), and uracil (U). The sequence 325 may include a set of identifiers 335 A-N (hereinafter generally referred to as identifiers 335) corresponding to the set of positions 330 of the gene segment 315 at the respective time. Each identifier 335 may have at least one alphanumeric character indicate a base type at the respective position 330 in the gene segment 315. For instance, the identifier 335 may have the alphanumeric character “A” for the nucleotide base type adenine, “C” for cytosine, “G” for guanine, “T” for thymine, and “U” for uracil, among others. In some embodiments, the sequence dataset 320 may include multiple sequences 325 corresponding to the multiple sequencing of the gene segments 315 from multiple sampling times.

[0071] In some embodiments, the genomic profiler 110 may generate the sequence dataset 320 to include identifications of mutations in the set of identifiers 335. The mutation may include, for example, a copy number alteration (e.g., gain, amplification, loss, or homozygous deletion), a deletion, a duplication, an inversion, an insertion, or a translocation, among others. The identification of the mutation may be performed by the genomic profiler 110 via variant calling (e.g., in a similar manner as the variant analyzer 130), such as by using a toolsuch as GATK HaplotypeCaller, Mutect2, Manta, Delly, and LUMPY, among others. With the generation of the sequence dataset 320, the genomic profiler 110 may send, transmit, or otherwise provide the sequence dataset 320 to the data processing system 105. In some embodiments, the genomic profiler 110 may send the sequence dataset 320 for storage on the database 170.100721 The dataset indexer 125 may obtain, receive, or otherwise retrieve the sequence dataset 320 from the genomic profiler 110 or the database 170. The sequence dataset 320 may include the sequence 325 generated using the gene segment 315 at the given sampling time (Ts). With the retrieval, the dataset indexer 125 may process or parse the sequence dataset 320 to extract or identify the sequence 325 corresponding to the gene segment 315. From the sequence 325, the dataset indexer 125 may extract or identify the set of identifiers 335 corresponding to the set of positions 330 in the gene segment 315. In some embodiments, the dataset indexer 125 may extract or identify the metadata associated with the sequence dataset 320, such as the sampling time at which the gene sequencing was performed on the gene segment 315 to generate the sequence 325, the length of the set of identifiers 335 in the sequence 325, gene type for the gene segment 315, the identifier corresponding the subject 305, and a cancer type that the subject 305 from which the gene segment 315 is obtained is at risk of or diagnosed with, among others.

[0073] The variation analyzer 130 may process or parse the set of identifiers 335 in each sequence 325 to detect or identify any variants. For each identifier 335 in a given sequence 325, the variation analyzer 130 may determine or identify one of a presence or absence of a mutation at the respective position 330 in the corresponding gene segment 315. In some embodiments, the sequence 325 of the sequence dataset 320 may lack metadata indicating presence of the mutation in the corresponding gene segment 315. To derive the frame dataset 210 from the sequence dataset 320, the variation analyzer 130 may traverse through the set of sequences 225 in the sequence dataset 320 to determine or identify mutations. The mutation may include, for example, a copy number alteration (e.g., gain, amplification, loss, or homozygous deletion), a deletion, a duplication, an inversion, an insertion, or a translocation, among others.[0074| In some embodiments, the variation analyzer 130 may identify the mutation by feeding the set of identifiers 335 of the sequence 325 via variant calling. The variant calling may be part of a detection algorithm or tool, such as a GATK HaplotypeCaller, Mutect2, Manta, Delly, or LUMPY, among others. In general, by using variant calling, the variation analyzer 130 may compare the sequence identifiers 335 of the sequence 325 derived from the gene segment 315 with a reference sequence for a reference gene segment of the same gene type. Based on the comparison, the variant analyzer 130 may identify the presence or absence of the mutation at one or more positions 330 in the gene segment 315. For example, if the base type at the respective position 330 differs from the base type at the same position in the reference gene segment, the variation analyzer 130 may identify the presence of the mutation at the respective position 330. Otherwise, if the base type at the respective position 330 is the same as the base type at the same position in the reference gene segment, the variation analyzer 130 may identify the absence of the mutation at the respective position 330 in the gene segment 315.

[0075] In some embodiments, the variation analyzer 130 may identify, generate, or otherwise generate at least one variant dataset 340 using the sequence 325. The variant dataset 340 may be used to generate seed frames for the initial frame dataset to account for the fact that the number of gene sequencing samples may be limited (e.g., one as depicted). The variant dataset 340 may identify or include a set of variant allele frequencies (VAFs) 345A-N (hereinafter generally referred to as VAFs 345) corresponding to the set of positions 330 in the gene segment 315 (or the identifiers 335 of the sequence 325). In the variant dataset 340, each VAF 345 may include or identify at least one frequency of a variant allele at the respective position 330 in the gene segment 315. The variant allele may include a nucleotide base type that is different from an expected nucleotide base type as identified in the gene sequence data for the reference gene of the same type as the gene segment 315, as a result of a mutation. For each identifier 335 of the sequence 325, the variation analyzer 130 may calculate, generate, or otherwise determine at least one VAF 345 via variant calling, using tools such as GATK HaplotypeCaller, Mutect2, Manta, Delly, or LUMPY, among others. For instance, the variant analyzer 130 may retrieve or identify the VAF 345 at each position of a reference gene segmentof the same gene type as the gene segment 315. The VAF 345 may be pre-defined (e.g., stored on the database 170) or may be determined as a function of one or more of the set of identifiers 335 of the sequence 325.

[0076] The frame constructor 135 may transform or convert the sequence dataset 320 to at least one frame dataset 350. The frame dataset 350 may include a set of frames in a sequence corresponding to sampling times (e.g., Ts+i, Ts+2 ...TS+N, TP, TP+I, ... as depicted) or states. The sequence may be in chronological order (e.g., as depicted) or anti -chronological order. In converting the sequence dataset 320, the frame constructor 135 may transform or convert the sequence 325 in the sequence dataset 320 to a corresponding frame 355 (also herein referred to as an original frame) for the frame dataset 350. The frame 355 may be an image or an arrayed representation of the sequence 325 and by extension the gene segment 315 sequenced at the sampling time (Ts). The frame 355 may correspond to the sequence 325 in the sequence dataset 320. By extension, each frame 355 may correspond to the gene segment 315 at the sampling time (Ts). In generating the frame 355, the frame constructor 135 may traverse through the set of identifiers 335 the sequence dataset 320 to generate the corresponding frame 355 in the frame dataset 350. The frame 355 may identify or include a set of coordinates 360A-N (hereinafter generally referred to as coordinates 360). The set of coordinates 245 may correspond to the set of identifiers 335 in the sequence 325. By extension, the set of coordinates 360 may correspond to the set of positions 330 in the gene segment 315.

[0077] For each identifier 335, the frame constructor 135 may create, write, or otherwise generate a corresponding coordinate 245 to include the set of values. Each coordinate 360 may include or have a set of values (sometimes herein referred to as channel values). At each coordinate 360, the set of values may identify or indicate the base type of the nucleotide base and at least one mutation type at the respective position 330 in the gene segment 315. Each value in the coordinate 360 may correspond to a respective channel (e.g., a color channel) to encode information about the gene segment 315 at the respective position 330, and by extension the respective identifier 335 of the corresponding sequence 325. The set of values may be numerical values (e.g., integer, double, floating value) or a Boolean value, among others, for the channelsforming the coordinate 360. The set of values may have any number channels. The set of values of the coordinate 245 may identify or include, for example, one or more of: at least one channel to indicate presence or absence of the nucleotide base type adenine (A), at least one channel to indicate presence or absence the cytosine (C), at least one channel to indicate presence or absence of guanine (G), at least one channel to indicate presence or absence of thymine (T), or at least one channel to indicate presence or absence of uracil (U), among others, at the respective position 230 corresponding to the identifier 335.

[0078] In some embodiments, the frame constructor 135 may generate the coordinate 360 using the identification of the mutation at the position 330 in the gene segment 315 corresponding to the identifier 335 in the sequence 325. The set of values at the coordinate 360 may identify or include, for example, at least one channel to indicate presence or absence of a copy number alteration (CNA) (e.g., gain, amplification, loss, or homozygous deletion), at least one channel to indicate presence or absence of deletion, at least one channel to indicate presence or absence of duplication, at least one channel to indicate presence or absence of inversion, at least one channel to indicate presence or absence of insertion, or at least one channel to indicate presence or absence of translocation, among others, at the respective position 330 corresponding to the identifier 335. The set of values may have any number of channels to represent and characterize the gene segment 315 at the respective position 330 associated with the coordinate 360. For instance, the set of values of each coordinate 360 in the frame 355 may be of the form {0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0} for the variables {A, C, G, T, U, CNA, Del., Dup., Inv., Ins., Transloc.}. Since the values at each coordinate 360 represent an encoding the information at the respective position 330 in the gene segment 315, at least one of the values can be set to a nonnull value (e.g., 1 or 100, as opposed to 0). If there are multiple sequences 325 in the sequence dataset 320, the frame constructor 135 may traverse each of the sequences 325 to generate the corresponding frames 355 for the frame dataset 350.

[0079] In some embodiments, in generating the frame dataset 350, the frame constructor135 may produce, create, or otherwise generate one or more seed frames 355’A-N (hereinafter generally referred to as seed frames 355) using the variant dataset 340. Each seed frame 355’may be an image or an arrayed representation of an expected sequence and by extension the gene segment 315 sequenced at respective sampling times (e.g., Ts+i, Ts+2 ...TS+N as depicted). The sampling times corresponding to the seed frames 355’ may be prior to or subsequent (e.g., as depicted) to the sampling time at which the gene segment 315 was sequenced to generate the sequence 325. To generate the seed frame 355’, the frame constructor 135 may generate a copy of the original frame 355 including the set of coordinates 360, each with the original set of channel values. For each coordinate 360, the frame constructor 135 may identify or determine a modification to apply, in accordance with the VAF 345 at the respective position 330 in the gene segment 315.

[0080] In determining, the frame constructor 135 may generate a seed value (e.g., using a pseudo-random number generator) and compare the value with the VAF 345 with a threshold. The threshold may delineate or define a value for the VAF 345 at which to apply the specified modification to the set of channel values of the coordinate 360. If the seed value satisfies (e.g., greater than or equal to), the frame constructor 135 may apply the specified modification to the set of channel values in the coordinate 360. Otherwise, if the seed value does not satisfy (e.g., less than), the frame constructor 135 may refrain from applying the specified modification to the set of channel values in the coordinate 360. With the generation of the seed frame 355’, the frame constructor 135 may insert, add, or otherwise include the seed frame 355’ to the frame dataset 350. The frame constructor 135 may include the seed frame 355’ corresponding to the sampling time within the frame dataset 350. For example, if the sampling time corresponding to the seed frame 355’ is prior to the sampling time associated with the original frame 355, the frame constructor 135 may include the generated seed frame 355’ prior to the original frame 355 (e.g., by pre-pending) within the frame dataset 350. Otherwise, if the sampling time corresponding to the seed frame 355’ is subsequent to the sampling time associated with the original frame 355, the frame constructor 135 may include the generated seed frame 355’ subsequent to the original frame 355 (e.g., by appending as depicted) within the frame dataset 350.[00811 In addition, the frame constructor 135 may produce, create, or otherwise generate one or more null frames 355” A-N (hereinafter generally referred to as null frames 355”). Each null frame 355” (sometimes herein referred to as placeholder frames) may function or serve as a respective placeholder for a predicted frame corresponding to a sampling time (e.g., Tp, Tp+i, . .. as depicted) (or state) different from the sampling time (e.g., Ts) (or state) of the original frame 355. For example, the sampling time for the null frame 355” may correspond to a time in the future relative to the sampling time corresponding the original frame 355 and by extension the sequence 325 and the gene segment 315. A time span between the sampling time (e.g., Ts) associated with the original frame 355 and the time (e.g., Tp, Tp+i, . . .) associated with the null frame 355” may range between 1 day and 5 years, such as 1 day, 3 days, 5 days, 1 week, 2 weeks, 3 weeks, 1 month, 3 month, 6 months, 9 months, 1 year, 1.5 years, 2 years, 2.5 years, 3 years, 3.5 years, 4 years, 4.5 years, or 5 years, among others. The null frame 355” may include the set of coordinates 360. Each of the coordinates 360 of the null frame 355” may include the set of channel values that are all set a predefined value (e.g., all null in the form of {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0} for the variables {A, C, G, T, U, CNA, Del., Dup., Inv., Ins., Transloc.}).

[0082] With the generation of the null frame 355”, the frame constructor 135 may insert, add, or otherwise include the null frame 355” to the frame dataset 350. The frame constructor 135 may include the null frame 355” corresponding to the sampling time within the frame dataset 350. For instance, if the sampling time corresponding to the null frame 355” is prior to the sampling time associated with the original frame 355, the frame constructor 135 may include the generated null frame 355” prior to the original frame 355 (e.g., by prepending) within the frame dataset 350. If the sampling time corresponding to the null frame 355” is prior to the sampling times associated with the original frame 355 and the seed frames 355’, the frame constructor 135 may include the null frame 355” prior to the seed frame 355’ and the original frame 355. Otherwise, if the sampling time corresponding to the null frame 355” is subsequent to the sampling time associated with the original frame 355, the frame constructor 135 may include the generated null frame 355” subsequent to the original frame 355 (e.g., by appending) within the frame dataset 350. If the sampling time corresponding to the null frame 355” issubsequent to the sampling times associated with the original frame 355 and the seed frames 355’ (e.g., as depicted), the frame constructor 135 may include the generated null frame 355” subsequent to the original frame 355 and the seed frames 355’s (e.g., by appending) within the frame dataset 350.100831 Moving onto FIG. 12B, the model applier 145 may may apply, feed, or otherwise provide the frame dataset 530 as input to the ML model 155. In some embodiments, the model applier 145 may feed each frame (e.g., the original frame 355, the seed frames 355’, or the null frames 355”) of the frame dataset 530 (e.g., in sequence) as input to the ML model 155. To provide the input to the ML model 155, the model applier 145 may feed the frame dataset 530 (or each individual frame) to the encoder 160 of the ML model 155. The model applier 145 may process the input (e.g., frame dataset 530 or each frame) in accordance with the set of weights of the ML model 155 across the encoder 160 and the decoder 165 to generate an output.

[0084] The encoder 160 may produce, determine, or otherwise generate a set of embeddings 365A-N (hereinafter generally referred to as embeddings 365) using the input. The set of embeddings 365 (sometimes herein referred to as a latent space genome map) may represent latent features within the input, such as context and dependency information. In some embodiments, the set of embeddings 365 may be an attention map. The set of embeddings 365 for the attention map may include a set of attention values. Each attention value may define or characterize a focus (e.g., weighting indicating specified) as assigned the encoder 160 on particular coordinates 360 within the input (e.g., frame dataset 530 or the individual frame 355). The focus may be based on relationship among different coordinates 360 of the input with respect to the target output of the encoder 160 and the overall ML model 155.

[0085] Continuing on, the encoder 160 may feed forward, input, or otherwise provide the set of embeddings 365 as input to the decoder 165. In some embodiments, the encoder 160 may identify or select at least a portion of the set of embeddings 365 to provide as input to the decoder 165, using the attention values of the attention map. For example, the encoder 160 may select the portion of the set of embeddings 365 corresponding to the highest N (e.g., top 5-20%)of attention values in the attention map. With the selection, the encoder 160 may feed the selected portion of the embeddings 365 to the decoder 165. The decoder 165 may produce, create, or otherwise generate at least one predicted frame set 350’ using the input embeddings 365 fed by the encoder 160. The predicted frame set 350’ from the decoder 165 may be the output of the overall ML model 155.

[0086] Based on providing the frame dataset 350 as the input to the ML model 155, the model applier 145 may produce, create, or otherwise generate the predicted frame dataset 350’. The predicted frame dataset 350’ may include the original frame 355 with the same set of coordinates 360 and corresponding set of channel values from the input frame dataset 350. The original frame 355 may also be in the same time in the sequence in the predicted frame dataset 350’ as in the input frame dataset 350. For instance, the original frame 355 may be at the same sampling time (Ts) in the predicted frame dataset 350’ as in the time (Ts) of the original frame 355 in the input frame dataset 350. In some embodiments, the predicted frame dataset 350’ may include the one or more seed frames 355’ with the same sets of coordinates 360 and corresponding set of channel values from the input frame dataset 350. The seed frames 355’ may also be in the same times in the sequence in the predicted frame dataset 350’ as in the input frame dataset 350. For example, the seed frames 355’ may be at the same respective sampling times (Ts+i, Ts+2 ...TS+N) in the predicted frame dataset 350’ as in the times (Ts+i, Ts+2 ...TS+N) of the times of the seed frame 355’ in the input frame dataset 350.

[0087] In addition, the predicted frame dataset 350’ may include one or more predicted frames 370A-N (hereinafter generally referred to as predicted frames 370). The one or more predicted frames 370 may substitute or replace the null frames 355” in the original frame dataset 350. The predicted frames 370 may also be in the same times in the sequence in the predicted frame dataset 350’ as the times of the null frames 355” in the input frame dataset 350. For instance, each predicted frame 370 may be at the same respective sampling time (e.g., Tp, Tp+i, . . .) in the predicted frame dataset 350’ as in the time (e.g., Tp, Tp+i, . ..) of the null frame 355” that the predicted frame 370 is replacing in the input frame dataset 350. Each predicted frame 370 may identify or include a corresponding set of coordinates 360’A-N (hereinafter generallyreferred to as a set of coordinates 360). The set of coordinates 360’ may correspond to the gene segment 315 at the respective sampling time (e.g., Tp, Tp+i, ... as depicted) (or state). The set of coordinates 360 may lack the coordinates 360 with the set of channel values set to null (or another predefined value). The set of coordinates 360’ for the predicted frame 370 may be of the same number as the set of coordinate 360 of the original frame 355 and the seed frame 355’.|0088| Each coordinate 360’ of the reconstructed frame dataset 210” may include or have a set of channel values. At each coordinate 360’, the set of values may identify or indicate the base type of the nucleotide base and at least one mutation type at the respective position 330 in the gene segment 315. The set of values may be numerical values (e.g., integer, double, floating value) or a Boolean value, among others, for the channels forming the coordinate 360’. The set of values at each coordinate 360’ of the predicted frame 370 may have any number channels. At each coordinate 360’, the set of values may include one or more of: at least one channel to indicate presence or absence of the nucleotide base type adenine (A), at least one channel to indicate presence or absence the cytosine (C), at least one channel to indicate presence or absence of guanine (G), at least one channel to indicate presence or absence of thymine (T), or at least one channel to indicate presence or absence of uracil (U), among others, at the respective position 330 of the gene segment 315. In some embodiments, the set of values may include one or more of: at least one channel to indicate presence or absence of a copy number alteration (CNA) (e.g., gain, amplification, loss, or homozygous deletion), at least one channel to indicate presence or absence of deletion, at least one channel to indicate presence or absence of duplication, at least one channel to indicate presence or absence of inversion, at least one channel to indicate presence or absence of insertion, or at least one channel to indicate presence or absence of translocation, among others, at the respective position 330 of the gene segment 315.

[0089] Referring now to FIG. 13, depicted is a block diagram of a process 400 for providing outputs in the system to generate frames. Under the process 400, the output evaluator 150 may calculate, generate, or otherwise determine at least one score 405, using the frames (e.g., the original frame 355, the seed frames 355’, and the predicted frames 370) of the predicted frame dataset 350’. The score 405 may identify or indicate a degree of progression of themulation associated with the cancer in the subject 305 over time (e.g., Ts-i, Ts+2 ...TS+N, TP, TP+I, . . . ). The score 405 may be based on a comparison the original frame 355 (and the seed frames 355’) relative to the predicted frames 370. In determining the score 405, the output evaluator 150 may compare the set of channel values in each coordinate 360 in the original frame 355 and the seed frames 355’ with the set of channel values in the corresponding coordinate 360’ in the predicted frame 370. Based on the comparison, the output evaluator 150 may identify or determine whether the set of channel values across the coordinates 360 and 360’ differ. Based on a number of differences in the set of channel values across the times, the output evaluator 150 may determine the value for the score 405. In general, the higher the number of differences in the original frame 355 (and the seed frames 355’) relative to the predicted frames 370, the higher the score 405 may be. Conversely, the lower the number of differences in the original frame 355 (and the seed frames 355’) relative to the predicted frames 370, the lower the score 405 may be. The progression of mutation may be correlated with a high progression risk, indicating that the subject 305 likely has a highly aggressive tumor. The subject 305 may be a candidate for more intensive treatment, closer monitoring, or a preventive approach, among others.

[0090] The output evaluator 150 may determine or generate at least one classification 410 for the subject 305, in accordance with a comparison between the score 405 and a threshold. The classification 410 may indicate or identify whether the subject 305 is a candidate or noncandidate for administration of therapy 415. The therapy 415 may include, for example, immunotherapy, chemotherapy, radiotherapy, or a targeted therapy, among others. The threshold may delineate or define a value for the score 405 at which to identify the subject 305 as the candidate. The threshold may depend on any number of factors, such as the sampling time of the predicted frame 370, the gene type of the gene segment, the type of cancer, or type of therapy 415, among others. To identify, the output evaluator 150 may compare the score 405 for the subject 305 with the threshold. If the score 405 satisfies (e.g., greater than or equal to) the threshold, the output evaluator 150 may generate the classification 410 to identify the subject 305 as the candidate for the administration of the therapy 415. In some embodiments, the output evaluator 150 may determine the classification 410 to indicate the mutation associated with the sicancer as progressively worse. In some embodiments, the output evaluator 150 may determine the classification 410 to identify the subject 305 as a first risk category (e.g., high-risk category). If the score 405 does not satisfy (e.g., less than) the threshold, the output evaluator 150 may generate the classification to identify the subject 305 as the non-candidate for the administration of the therapy 415. In some embodiments, the output evaluator 150 may determine the classification 410 to indicate the mutation associated with the cancer as not progressively worse (e.g., steady). In some embodiments, the output evaluator 150 may determine the classification 410 to identify the subject 305 as a second category (e.g., low-risk category).

[0091] The output evaluator 150 may store and maintain an association between the subject 305 (e.g., using an identifier) and the predicted frame dataset 350’ including the predicted frames 370 for the gene segment 315 at the respective times (e.g., Tp, Tp+i, .. .) on the database 170, using one or more data structures or files. In some embodiments, the association may be among one or more of: the subject 305, the sequence dataset 320, the predicted frame dataset 350’, the score 405, and the classification 410, among others. The data structures may include, for example, an array, matrix, table, linked list, binary tree, heap, stack, queue, class object, or data file, among others. The files may include, for example, extensible markup language (XML), comma-separated values (CSV), structured query language (SQL), or JavaScript Object Notation (JSON), among others.

[0092] The output evaluator 150 may produce, create, or otherwise generate at least output 420. The output 420 may include information based on one or more of the sequence dataset 320, the predicted frame dataset 350’, the score 405, and the classification 410, among others. The output 420 may include instructions defining rendering or presentation of the information included therein. In some embodiments, the output evaluator 150 may generate the output 420 to identify or include at least one indication 425 based on the classification 410. When the classification to identify the subject 305 as the candidate for the administration of the therapy 415, the output evaluator 150 may generate the indication 425 to identify the subject 305 as the candidate for the administration of the therapy 415. In some embodiments, the indication 425 may include the identification of the therapy 415, for which the subject 305 is identified asthe candidate. Conversely, when the classification to identify the subject 305 as the noncandidate for the administration of the therapy 415, the output evaluator 150 may generate the indication 425 to identify the subject 305 as the non-candidate for the administration of the therapy 415.

[0093] The administrative device 115 may retrieve, obtain, or otherwise receive the output 420 from the data processing system 105. With receipt, the administrative device 115 may display, render, or otherwise present the information of the output 420, including the indication 425. For example, the administrative device 115 may display the sequence dataset 320, the predicted frame dataset 350’, the score 405, and the classification 410 via a user interface (e.g., a graphical user interface) of a web application hosted on the data processing system 105. A user (e g., a clinician examining the subject 305) of the administrative device 115 may use the information presented via the user interface in a clinical decision. When the indication 425 identifies the subject 305 as the candidate for administration of therapy 415, the subject 305 may be administered (e.g., by the clinician) with therapy 415 for the cancer at a therapeutically effective dosage or amount. When the indication 425 identifies the subject 305 as the non-candidate for administration of therapy 415, the subject 305 may be not be administered with therapy for the cancer. In some embodiments, the subject 305 may be administered with a previously selected therapy for the cancer.

[0094] In this manner, the data processing system 105 may use the image-like frame datasets and the ML model 155 to predict the mutation and its progression across time in the subject. The frame dataset may define spatio-temporal relationships across genomic coordinates and time for the genomic data. Unlike statistical-based methods, the ML model 155, integrated with attention maps and knowledge distillation, may allow capturing of data dependencies (including local and global context) and focusing on relevant portions of a very sparse frame dataset. This can allow for the ML model 155 to generate predicted mutations with high resolution and accuracy. By leveraging attention mechanisms and hierarchical multi-scale approaches, the ML model 155 can focus on relevant portions of the data, reducing the computational load on processors and memory on the part of the data processing system 105.Furthermore, the multi-stage process of reducing and compressing the data into a latent feature space and then up-sampling and decompressing while retaining relevant information for the frame prediction can optimize memory utilization.

[0095] From a clinical perspective, the generation of the predicted frame datasets with high-resolution and high accuracy can allow for identification of subjects for candidates for additional anti-cancer therapy. Identifying such subjects for the administration of therapy can lead to improved clinical outcomes by assisting clinicians to make better-informed decisions about the administration of therapies. For instance, in anticipating future mutations for therapy planning, by predicting likely future mutations, the ML model 155 can help clinicians design a treatment strategy that considers when and which targeted therapies will be effective. In clinical practice, cancer often can recur not because the initial mutations were not treated, but because new, untreated mutations emerge after initial therapy. This method can enable early identification of such potential resistance mutations, allowing a preemptive selection of drugs to target them before they appear. In addition, by identifying high-risk patients, the ML model 155 can also help detect subjects with highly aggressive tumors, characterized by a high number and rapid accumulation of mutations. These individuals may benefit from more intensive treatment regimens or even preventive therapies, improving outcomes by addressing tumor evolution earlier in its course.

[0096] Referring now to FIG. 14, depicted is a flow diagram of a method 500 of generating frames across time from sequence data using machine learning (ML) models. The method 500 may be implemented or performed by any of the components detailed herein, such as the system 100 or the system 600. Under the method 500, a computing system may retrieve a sequence dataset including a set of identifiers derived from a gene sequence (505). The computing system may identify mutations in the sequence dataset (510). The computing system may convert the sequence dataset to a frame (515). The computing system may provide the frame as an input to a machine learning (ML) model (520). The computing system may generate a predicted frame based on providing the input to the ML model (525). The computing system may determine a score indicating progression based on the frames (530). The computing systemmay generate a classification in accordance with the score (535). The computing system may determine whether the subject is a candidate for therapy based on the classification (540). If the subject is determined to be a candidate, the computing system may identify the subject as a candidate for therapy (545). Otherwise, if the subject is determined to be a non-candidate, the computing system may identify the subject as a non-candidate for therapy (550). The computing system may provide an output in accordance with the identification (555).C. Computing and Network Environment[0097| Various operations described herein can be implemented on computer systems. FIG. 15 shows a simplified block diagram of a representative server system 600, client computing system 614, and network 626 usable to implement certain embodiments of the present disclosure. In various embodiments, server system 600 or similar systems can implement services or servers described herein or portions thereof. Client computing system 614 or similar systems can implement clients described herein. The system 100 described herein can be similar to the server system 600. Server system 600 can have a modular design that incorporates a number of modules 602 (e.g., blades in a blade server embodiment); while two modules 602 are shown, any number can be provided. Each module 602 can include processing unit(s) 604 and local storage 606.

[0098] Processing unit(s) 604 can include a single processor, which can have one or more cores, or multiple processors. In some embodiments, processing unit(s) 604 can include a general-purpose primary processor as well as one or more special-purpose co-processors such as graphics processors, digital signal processors, or the like. In some embodiments, some or all processing units 604 can be implemented using customized circuits, such as application specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs). In some embodiments, such integrated circuits execute instructions that are stored on the circuit itself. In other embodiments, processing unit(s) 604 can execute instructions stored in local storage 606. Any type of processors in any combination can be included in processing unit(s) 604.[0099| Local storage 606 can include volatile storage media (e g., DRAM, SRAM, SDRAM, or the like) and / or non-volatile storage media (e.g., magnetic or optical disk, flash memory, or the like). Storage media incorporated in local storage 606 can be fixed, removable or upgradeable as desired. Local storage 606 can be physically or logically divided into various subunits such as a system memory, a read-only memory (ROM), and a permanent storage device. The system memory can be a read-and-write memory device or a volatile read-and-write memory, such as dynamic random-access memory. The system memory can store some or all of the instructions and data that processing unit(s) 604 need at runtime. The ROM can store static data and instructions that are needed by processing unit(s) 604. The permanent storage device can be a non-volatile read-and-write memory device that can store instructions and data even when module 602 is powered down. The term “storage medium” as used herein includes any medium in which data can be stored indefinitely (subject to overwriting, electrical disturbance, power loss, or the like) and does not include carrier waves and transitory electronic signals propagating wirelessly or over wired connections.|0100| In some embodiments, local storage 606 can store one or more software programs to be executed by processing unit(s) 604, such as an operating system and / or programs implementing various server functions such as functions of the system 100 of FIG. 15 or any other system described herein, or any other server(s) associated with system 100 or any other system described herein.|0101| Software” refers generally to sequences of instructions that, when executed by processing unit(s) 604 cause server system 600 (or portions thereof) to perform various operations, thus defining one or more specific machine embodiments that execute and perform the operations of the software programs. The instructions can be stored as firmware residing in read-only memory and / or program code stored in non-volatile storage media that can be read into volatile working memory for execution by processing unit(s) 604. Software can be implemented as a single program or a collection of separate programs or program modules that interact as desired. From local storage 606 (or non-local storage described below), processing unit(s) 604can retrieve program instructions to execute and data to process in order to execute various operations described above.

[0102] In some server systems 600, multiple modules 602 can be interconnected via a bus or other interconnect 608, forming a local area network that supports communication between modules 602 and other components of server system 600. Interconnect 608 can be implemented using various technologies including server racks, hubs, routers, etc.[01031 A wide area network (WAN) interface 610 can provide data communication capability between the local area network (interconnect 608) and the network 626, such as the Internet. Technologies can be used, including wired (e.g., Ethernet, IEEE 602.3 standards) and / or wireless technologies (e.g., Wi-Fi, IEEE 602.11 standards).

[0104] In some embodiments, local storage 606 is intended to provide working memory for processing unit(s) 604, providing fast access to programs and / or data to be processed while reducing traffic on interconnect 608. Storage for larger quantities of data can be provided on the local area network by one or more mass storage subsystems 612 that can be connected to interconnect 608. Mass storage subsystem 612 can be based on magnetic, optical, semiconductor, or other data storage media. Direct attached storage, storage area networks, network-attached storage, and the like can be used. Any data stores or other collections of data described herein as being produced, consumed, or maintained by a service or server can be stored in mass storage subsystem 612. In some embodiments, additional data storage resources may be accessible via WAN interface 610 (potentially with increased latency).

[0105] Server system 600 can operate in response to requests received via WAN interface 610. For example, one of modules 602 can implement a supervisory function and assign discrete tasks to other modules 602 in response to received requests. Work allocation techniques can be used. As requests are processed, results can be returned to the requester via WAN interface 610. Such operation can generally be automated. Further, in some embodiments, WAN interface 610 can connect multiple server systems 600 to each other, providing scalable systems capable of managing high volumes of activity. Other techniques formanaging server systems and server farms (collections of server systems that cooperate) can be used, including dynamic resource allocation and reallocation.

[0106] Server system 600 can interact with various user-owned or user-operated devices via a wide-area network such as the Internet. An example of a user-operated device is shown in FIG. 15 as client computing system 614. Client computing system 614 can be implemented, for example, as a consumer device such as a smartphone, other mobile phone, tablet computer, wearable computing device (e.g., smart watch, eyeglasses), desktop computer, laptop computer, and so on.

[0107] For example, client computing system 614 can communicate via WAN interface 610. Client computing system 614 can include computer components such as processing unit(s) 616, storage device 618, network interface 620, user input device 622, and user output device 624. Client computing system 614 can be a computing device implemented in a variety of form factors, such as a desktop computer, laptop computer, tablet computer, smartphone, other mobile computing device, wearable computing device, or the like.

[0108] Processing unit(s) 616 and storage device 618 can be similar to processing unit(s) 604 and local storage 606 described above. Suitable devices can be selected based on the demands to be placed on client computing system 614; for example, client computing system 614 can be implemented as a “thin” client with limited processing capability or as a high- powered computing device. Client computing system 614 can be provisioned with program code executable by processing unit(s) 616 to enable various interactions with server system 600.

[0109] Network interface 620 can provide a connection to the network 626, such as a wide area network (e.g., the Internet) to which WAN interface 610 of server system 600 is also connected. In various embodiments, network interface 620 can include a wired interface (e.g., Ethernet) and / or a wireless interface implementing various RF data communication standards such as Wi-Fi, Bluetooth, or cellular data network standards (e.g., 3G, 4G, LTE, etc.).

[0110] User input device 622 can include any device (or devices) via which a user can provide signals to client computing system 614. The client computing system 614 can interpret the signals as indicative of particular user requests or information. In various embodiments, user input device 622 can include any or all of a keyboard, touch pad, touch screen, mouse or other pointing device, scroll wheel, click wheel, dial, button, switch, keypad, microphone, and so on.[Olli] User output device 624 can include any device via which client computing system 614 can provide information to a user. For example, user output device 624 can include a display to display images generated by or delivered to client computing system 614. The display can incorporate various image generation technologies, e.g., a liquid crystal display (LCD), lightemitting diode (LED) including organic light-emitting diodes (OLED), projection system, cathode ray tube (CRT), or the like, together with supporting electronics (e g., digital-to-analog or analog-to-digital converters, signal processors, or the like). Some embodiments can include a device such as a touchscreen that function as both input and output device. In some embodiments, other user output devices 624 can be provided in addition to or instead of a display. Examples include indicator lights, speakers, tactile “display” devices, printers, and so on.

[0112] Some embodiments include electronic components, such as microprocessors, storage and memory that store computer program instructions in a computer-readable storage medium. Many of the features described in this specification can be implemented as processes that are specified as a set of program instructions encoded on a computer-readable storage medium. When these program instructions are executed by one or more processing units, they cause the processing unit(s) to perform various operation indicated in the program instructions. Examples of program instructions or computer code include machine code, such as is produced by a compiler, and files including higher-level code that are executed by a computer, an electronic component, or a microprocessor using an interpreter. Through suitable programming, processing unit(s) 604 and 616 can provide various functionality for server system 600 and client computing system 614, including any of the functionality described herein as being performed by a server or client, or other functionality.[0113| It will be appreciated that server system 600 and client computing system 614 are illustrative and that variations and modifications are possible. Computer systems used in connection with embodiments of the present disclosure can have other capabilities not specifically described here. Further, while server system 600 and client computing system 614 are described with reference to particular blocks, it is to be understood that these blocks are defined for convenience of description and are not intended to imply a particular physical arrangement of component parts. For instance, different blocks can be but need not be located in the same facility, in the same server rack, or on the same motherboard. Further, the blocks need not correspond to physically distinct components. Blocks can be configured to perform various operations, e.g., by programming a processor or providing appropriate control circuitry, and various blocks might or might not be reconfigurable depending on how the initial configuration is obtained. Embodiments of the present disclosure can be realized in a variety of apparatus including electronic devices implemented using any combination of circuitry and software.

[0114] While the disclosure has been described with respect to specific embodiments, one skilled in the art will recognize that numerous modifications are possible. Embodiments of the disclosure can be realized using a variety of computer systems and communication technologies including but not limited to the specific examples described herein. Embodiments of the present disclosure can be realized using any combination of dedicated components and / or programmable processors and / or other programmable devices. The various processes described herein can be implemented on the same processor or different processors in any combination. Where components are described as being configured to perform certain operations, such configuration can be accomplished, e.g., by designing electronic circuits to perform the operation, by programming programmable electronic circuits (such as microprocessors) to perform the operation, or any combination thereof. Further, while the embodiments described above may make reference to specific hardware and software components, those skilled in the art will appreciate that different combinations of hardware and / or software components may also be used and that particular operations described as being implemented in hardware might also be implemented in software or vice versa.[0115| Computer programs incorporating various features of the present disclosure may be encoded and stored on various computer-readable storage media; suitable media include magnetic disk or tape, optical storage media such as compact disk (CD) or DVD (digital versatile disk), flash memory, and other non-transitory media. Computer-readable media encoded with the program code may be packaged with a compatible electronic device, or the program code may be provided separately from electronic devices (e.g., via Internet download or as a separately packaged computer-readable storage medium).

[0116] Thus, although the disclosure has been described with respect to specific embodiments, it will be appreciated that the disclosure is intended to cover all modifications and equivalents within the scope of the following claims.

Claims

WHAT IS CLAIMED IS:

1. A method of generating frames across time from sequence data using machine learning (ML) models, comprising: retrieving, by one or more processors, a sequence dataset comprising a plurality of identifiers generated using a gene segment from a subject at a first time, each identifier of the plurality of identifiers having an alphanumeric character indicating a base type of a plurality of base types at a respective position of a plurality of positions in the gene segment; converting, by the one or more processors, the plurality of identifiers of the sequence dataset to generate a first frame comprising a first plurality of coordinates for the gene segment at the first time; providing, by the one or more processors, the first frame as an input to a ML model, wherein the ML model is established using a plurality of examples, each of the plurality of examples comprising a respective plurality of example frames for a respective gene segment across a corresponding plurality of times; generating, by the one or more processors, based on providing the first frame dataset as the input to the ML model, a second frame comprising a second plurality of coordinates for the gene segment at a second time; and storing, by the one or more processors, using one or more data structures, an association between the subject and the second frame for the gene segment at the second time.

2. The method of claim 1, wherein the subject is at risk of or diagnosed with cancer, and further comprising: determining, by the one or more processors, a score indicating a degree of progression of mutation associated with the cancer in the subject based on the first frame at the first time and the second frame at the second time; and generating, by the one or more processors, a classification for the subject in accordance a comparison between the score and a threshold.

3. The method of claim 2, wherein generating the classification further comprise generating the classification identifying the subject as a candidate for administration of therapy for the cancer, responsive to the score satisfying the threshold, and further comprising: providing, by the one or more processors, an output indicating the subject as the candidate for administration of the therapy for the cancer.

4. The method of claim 3, wherein the subject is administered with the therapy at a therapeutically effective dosage for the cancer subsequent to provision of the output, wherein the therapy comprises at least one of immunotherapy, chemotherapy, radiotherapy, or a targeted therapy.

5. The method of claim 2, wherein generating the classification further comprise generating the classification identifying the subject as a non-candidate for administration of therapy for the cancer, responsive to the score not satisfying the threshold, and further comprising: providing, by the one or more processors, an output indicating the subject as the noncandidate for administration of the therapy for the cancer, wherein the subject is administered with at least one of no therapy or a previous therapy for the cancer subsequent to provision of the output.

6. The method of claim 1, further comprising identifying, by the one or more processors, for each identifier of the plurality of identifiers of the sequence dataset, one of a presence or an absence of a mutation at the respective position of the plurality of positions in the gene segment, wherein the plurality of mutations comprises at least one of a copy number alteration, a deletion, a duplication, an inversion, an insertion, or a translocation; wherein converting the sequence dataset further comprises generating the first frame comprising the first plurality of coordinates, each coordinate of the first plurality of coordinates having a respective first set of channel values identifying the base type and one of the presence or the absence of the mutation at the respective position of the plurality of positions in the gene segment at the first time, andwherein generating the second frame further comprises generating the second frame comprising a second plurality of coordinates for the gene segment at the second time, each coordinate of the second plurality of coordinates having a respective second set of channel values identifying the base type and one of the presence or the absence of the mutation at the respective position of the plurality of positions in the gene segment at the second time.

7. The method of claim 1, further comprising: determining, by the one or more processors, a plurality of variant allele frequencies (VAFs) corresponding to the plurality of identifiers of the sequence dataset, each of the plurality VAFs identifying a frequency of a variant allele at the respective position of the plurality of positions in the gene segment; and generating, by the one or more processors, using the plurality of VAFs, a third frame comprising a third plurality of coordinates for the gene segment at a third time, wherein providing the first frame to the ML model further comprises providing the first frame and third frame as input to the input ML model to generate the second frame.

8. The method of claim 1, wherein converting the sequence dataset further comprises generating the first frame further comprises generating a frame dataset comprising a plurality of frames corresponding to a plurality of times, the plurality of frames including (i) the first frame corresponding to the first time of the plurality of times and (ii) a null frame as a placeholder for the second frame corresponding to the second time of the plurality of times, wherein a time span between the first time and the second time ranges between 1 day and 5 years.

9. The method of claim 1, wherein the ML model is established by: identifying, from the respective plurality of frames of an example of the plurality of examples, a first example frame comprising a first example plurality of coordinates, each of the first example plurality of coordinates having a corresponding first example set of channel values at a corresponding position of a plurality of positions in the respective gene segment;modifying the first example frame by masking one or more of the respective plurality of coordinates to generate a second example frame comprising a second example plurality of coordinates, one or more of the second example plurality of coordinates having null values; providing the second example frame as input to the ML model to generate a third example frame comprising a third example plurality of coordinates; determining a loss metric based on a comparison between the first example plurality of coordinates of the first frame and the third example plurality of coordinates of the third frame; and updating one or more weights of the ML model in accordance with the loss metric.

10. The method of claim 1, wherein the subject is at risk of or diagnosed with cancer comprising at least one of skin cancer, lung cancer, brain cancer, bone cancer, leukemia, lymphoma, head and neck cancer, colon cancer, rectal cancer, uterine cancer, endometrial cancer, pancreatic cancer, thyroid cancer, stomach cancer, ovarian cancer, cervical cancer, bladder cancer, or breast cancer, and wherein the gene segment is obtained via a sample from the subject, wherein the sample comprises at least one of a skin, lung, brain, bone, bone marrow, lymph node, head, neck, colon, rectum, uterus, endometrium, pancreas, thyroid, stomach, ovary, cervix, bladder, breast, urine, plasma, or blood.

11. A system for generating frames across time from sequence datasets using machine learning (ML) models, comprising: one or more processors coupled with memory, configured to: retrieve a sequence dataset comprising a plurality of identifiers generated using a gene segment from a subject at a first time, each identifier of the plurality of identifiers having an alphanumeric character indicating a base type of a plurality of base types at a respective position of a plurality of positions in the gene segment; convert the plurality of the identifiers of the sequence dataset to generate a first frame comprising a first plurality of coordinates for the gene segment at the first time;provide the first frame as an input to a ML model, wherein the ML model is established using a plurality of examples, each of the plurality of examples comprising a respective plurality of example frames for a respective gene segment across a corresponding plurality of times; generate, based on providing the first frame dataset as the input to the ML model, a second frame comprising a second plurality of coordinates for the gene segment at a second time; and store, using one or more data structures, an association between the subject and the second frame for the gene segment at the second time.

12. The system of claim 11, wherein the subject is at risk of or diagnosed with cancer, and wherein the one or more processors are configured to: determine a score indicating a degree of progression of mutation associated with the cancer in the subject based on the first frame at the first time and the second frame at the second time; and generate a classification for the subject in accordance a comparison between the score and a threshold.

13. The system of claim 12, wherein the one or more processors are configured to: generate the classification identifying the subject as a candidate for administration of therapy for the cancer, responsive to the score satisfying the threshold, and provide an output indicating the subject as the candidate for administration of the therapy for the cancer.

14. The system of claim 13, wherein the subject is administered with the therapy for the cancer subsequent to provision of the output, wherein the therapy comprises at least one of immunotherapy, chemotherapy, radiotherapy, or a targeted therapy.

15. The system of claim 12, wherein the one or more processors are configured to:generate the classification identifying the subject as a non-candidate of administration of therapy for the cancer, responsive to the score not satisfying the threshold; provide an output indicating the subject as the non-candidate for administration of the therapy for the cancer, wherein the subject is administered with at least one of no therapy or a previous therapy for the cancer subsequent to provision of the output.

16. The system of claim 12, wherein the one or more processors are further configured to: identify, for each identifier of the plurality of identifiers of the sequence dataset, one of a presence or an absence of a mutation at the respective position of the plurality of positions in the gene segment, wherein the plurality of mutations comprises at least one of a copy number alteration, a deletion, a duplication, an inversion, an insertion, or a translocation; convert the sequence dataset to generate the first frame comprising the first plurality of coordinates, each coordinate of the first plurality of coordinates having a respective first set of channel values identifying the base type and one of the presence or the absence of the mutation at the respective position of the plurality of positions in the gene segment at the first time, and generate, based on providing the input to the ML model, the second frame comprising a second plurality of coordinates for the gene segment at the second time, each coordinate of the second plurality of coordinates having a respective second set of channel values identifying the base type and one of the presence or the absence of the mutation at the respective position of the plurality of positions in the gene segment at the second time.

17. The system of claim 11, wherein the one or more processors are further configured to: determine a plurality of variant allele frequencies (VAFs) corresponding to the plurality of identifiers of the sequence dataset, each of the plurality of VAFs identifying a frequency of a variant allele at the respective position of the plurality of positions in the gene segment; generate, using the plurality of VAFs, a third frame comprising a third plurality of coordinates for the gene segment at a third time; and provide the first frame and third frame as input to the ML model to generate the second frame.

18. The system of claim 11, wherein the one or more processors are further configured to generate a frame dataset comprising a plurality of frames corresponding to a plurality of times, the plurality of frames including (i) the first frame corresponding to the first time of the plurality of times and (ii) a null frame as a placeholder for the second frame corresponding to the second time of the plurality of times, wherein a time span between the first time and the second time ranges between 1 day and 5 years.

19. The system of claim 11, wherein the ML model is established by: identifying, from the respective plurality of frames of an example of the plurality of examples, a first example frame comprising a first example plurality of coordinates, each of the first example plurality of coordinates having a corresponding first example set of channel values at a corresponding position of a plurality of positions in the respective gene segment; modifying the first example frame by masking one or more of the respective plurality of coordinates to generate a second example frame comprising a second example plurality of coordinates, one or more of the second example plurality of coordinates having null values; providing the second example frame as input to the ML model to generate a third example frame comprising a third example plurality of coordinates; determining a loss metric based on a comparison between the first example plurality of coordinates of the first frame and the third example plurality of coordinates of the third frame; and updating one or more weights of the ML model in accordance with the loss metric.

20. The system of claim 11, wherein the subject is at risk of or diagnosed with cancer comprising at least one of skin cancer, lung cancer, brain cancer, bone cancer, leukemia, lymphoma, head and neck cancer, colon cancer, rectal cancer, uterine cancer, endometrial cancer, pancreatic cancer, thyroid cancer, stomach cancer, ovarian cancer, cervical cancer, bladder cancer, or breast cancer, andwherein the gene segment is obtained via a sample from the subject, wherein the sample comprises at least one of a skin, lung, brain, bone, bone marrow, lymph node, head, neck, colon, rectum, uterus, endometrium, pancreas, thyroid, stomach, ovary, cervix, bladder, breast, urine, plasma, or blood.

Images