# US 12,217,832 B2 · Deep learning-based variant classifier

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,217,832 B2
- Document ID: US12217832B2
- Status: Granted
- Kind: B2
- Application number: 18/314,638
- Filed: 2023-05-09
- Published: 2023-11-30
- Granted: 2025-02-04
- Earliest claimed: 2018-01-15
- Assignees: Illumina, Inc.; Illumina Cambridge Limited
- Applicants: Illumina, Inc.; Illumina Cambridge Limited
- Inventors: Ole Schulz-Trieglaff; Anthony James Cox; Kai-How Farh
- Web page: https://patentgrep.com/patent/US12217832B2

## CPC

- G16B 40/20: Supervised data analysis
- G06F 9/3877: {using a secondary processor, e.g. coprocessor (peripheral processor G06F13/12)}
- G06F 18/2148: {characterised by the process organisation or structure, e.g. boosting cascade}
- G06F 18/2431: Multiple classes
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/0499: Feedforward networks
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/09: Supervised learning
- G16B 20/00: ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
- G16B 20/20: Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection
- G16B 40/00: ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
- G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

The technology disclosed directly operates on sequencing data and derives its own feature filters. It processes a plurality of aligned reads that span a target base position. It combines elegant encoding of the reads with a lightweight analysis to produce good recall and precision using lightweight hardware. For instance, one million training examples of target base variant sites with 50 to 100 reads each can be trained on a single GPU card in less than 10 hours with good recall and precision. A single GPU card is desirable because it a computer with a single GPU is inexpensive, almost universally within reach for users looking at genetic data. It is readily available on could-based platforms.

## Claims

### Claim 1

A system comprising:
at least one processor; and
a non-transitory computer readable medium storing a convolutional neural network and instructions that, when executed by the at least one processor, cause the system to:
identify a group of reads aligned with a reference genome and spanning a candidate variant at a target base position;
provide, to the convolutional neural network, an array of input features generated from a text file comprising sequencing data output by a sequencer instrument, the array of input features encoding:
bases from the group of reads in the text file at the target base position,
bases flanking each side of the target base position in the text file, and
corresponding base features for bases within the group of reads; and
generate, based on an analysis of the array of input features by the convolutional neural network, classification scores indicating likelihoods that the candidate variant at the target base position is a variant.

### Claim 2 (depends on 1)

The system of claim 1, further comprising instructions that, when executed by the at least one processor, cause the system to generate the classification scores by generating a first probability that the candidate variant is a true variant and a second probability that the candidate variant is a false variant.

### Claim 3 (depends on 1)

The system of claim 1, further comprising instructions that, when executed by the at least one processor, cause the system to generate the classification scores by generating one or more of a first score that the candidate variant is a homozygous variant, a second score that the candidate variant is a heterozygous variant, a third score that the candidate variant is a non-variant, or a fourth score that the candidate variant is a complex-variant.

### Claim 4 (depends on 1)

The system of claim 1, further comprising instructions that, when executed by the at least one processor, cause the system to:
process the array of input features through one or more layers of the convolutional neural network to generate convolutional output features;
concatenate the convolutional output features with per-variant characterization data corresponding to the candidate variant at the target base position to generate concatenated features; and
generate the classification scores based on an analysis of the concatenated features by one or more layers of the convolutional neural network.

### Claim 5 (depends on 4)

The system of claim 4, wherein the per-variant characterization data corresponding to the candidate variant comprise empirical variant score (EVS) features for the candidate variant.

### Claim 6 (depends on 4)

The system of claim 4, further comprising instructions that, when executed by the at least one processor, cause the system to generate the classification scores based on the analysis of the concatenated features by one or more fully connected layers and a classification layer of the convolutional neural network.

### Claim 7 (depends on 4)

The system of claim 4, further comprising instructions that, when executed by the at least one processor, cause the system to process the array of input features through the one or more layers of the convolutional neural network by processing the array of input features through a one or more of an input layer, a convolution layer, a batch normalization layer, a max pooling layer, or a flattening layer.

### Claim 8 (depends on 1)

The system of claim 1, further comprising instructions that, when executed by the at least one processor, cause the system to provide the array of input features in which each feature corresponds to a base in a read from the group of reads and each feature comprises a plurality of dimensions.

### Claim 9 (depends on 1)

The system of claim 1, wherein the corresponding base features for bases within the group of reads comprise one or more of:
a corresponding reference base in a reference read corresponding to the reference genome;
a base call accuracy score of calling a base in a read of the group of reads;
a strandedness of calling the base;
an insertion count of changes adjoining a position of the base;
a deletion flag at the position of the base; or
a mapping quality score of aligning a corresponding read that contains the base to the reference read.

### Claim 10 (depends on 1)

The system of claim 1, wherein the text file comprises a binary alignment map (BAM) file.

### Claim 11

A non-transitory computer readable storage medium storing instructions that, when executed by at least one processor, cause a computing system to:
identify a group of reads aligned with a reference genome and spanning a candidate variant at a target base position;
provide, to a convolutional neural network, an array of input features generated from a text file comprising sequencing data output by a sequencer instrument, the array of input features encoding:
bases from the group of reads in the text file at the target base position,
bases flanking each side of the target base position in the text file, and
corresponding base features for bases within the group of reads; and
generate, based on an analysis of the array of input features by the convolutional neural network, classification scores indicating likelihoods that the candidate variant at the target base position is a variant.

### Claim 12 (depends on 11)

The non-transitory computer readable storage medium of claim 11, further storing instructions that, when executed by the at least one processor, cause the computing system to generate the classification scores by generating a first probability that the candidate variant is a true variant and a second probability that the candidate variant is a false variant.

### Claim 13 (depends on 11)

The non-transitory computer readable storage medium of claim 11, further storing instructions that, when executed by the at least one processor, cause the computing system to generate the classification scores by generating one or more of a first score that the candidate variant is a homozygous variant, a second score that the candidate variant is a heterozygous variant, a third score that the candidate variant is a non-variant, or a fourth score that the candidate variant is a complex-variant.

### Claim 14 (depends on 11)

The non-transitory computer readable storage medium of claim 11, further storing instructions that, when executed by the at least one processor, cause the computing system to:
process the array of input features through one or more layers of the convolutional neural network to generate convolutional output features;
concatenate the convolutional output features with per-variant characterization data corresponding to the candidate variant at the target base position to generate concatenated features; and
generate the classification scores based on an analysis of the concatenated features by one or more layers of the convolutional neural network.

### Claim 15 (depends on 14)

The non-transitory computer readable storage medium of claim 14, wherein the per-variant characterization data corresponding to the candidate variant comprise empirical variant score (EVS) features for the candidate variant.

### Claim 16 (depends on 14)

The non-transitory computer readable storage medium of claim 14, further storing instructions that, when executed by the at least one processor, cause the computing system to generate the classification scores based on the analysis of the concatenated features by one or more fully connected layers and a classification layer of the convolutional neural network.

### Claim 17

A computer-implemented method comprising:
identifying a group of reads aligned with a reference genome and spanning a candidate variant at a target base position;
providing, to a convolutional neural network, an array of input features generated from a text file comprising sequencing data output by a sequencer instrument, the array of input features encoding:
bases from the group of reads in the text file at the target base position,
bases flanking each side of the target base position in the text file, and
corresponding base features for bases within the group of reads; and
generating, based on an analysis of the array of input features by the convolutional neural network, classification scores indicating likelihoods that the candidate variant at the target base position is a variant.

### Claim 18 (depends on 17)

The computer-implemented method of claim 17, wherein generating the classification scores comprises generating a first probability that the candidate variant is a true variant and a second probability that the candidate variant is a false variant.

### Claim 19 (depends on 17)

The computer-implemented method of claim 17, wherein generating the classification scores comprises generating one or more of a first score that the candidate variant is a homozygous variant, a second score that the candidate variant is a heterozygous variant, a third score that the candidate variant is a non-variant, or a fourth score that the candidate variant is a complex-variant.

### Claim 20 (depends on 17)

The computer-implemented method of claim 17, wherein the corresponding base features for bases within the group of reads comprise one or more of:
a corresponding reference base in a reference read corresponding to the reference genome;
a base call accuracy score of calling a base in a read of the group of reads;
a strandedness of calling the base;
an insertion count of changes adjoining a position of the base;
a deletion flag at the position of the base; or
a mapping quality score of aligning a corresponding read that contains the base to the reference read.