Illumina Corporation

Illumina Cambridge Limited

Kai-How Farh

Hong Gao

Jeremy Francis McRae

Donavan Cheng

Sofia Kyriazopoulou Panagiotopoulou

Laksshman Sundaram

Kishore Jaganathan

Milan Karangutkar

Jorg Hakenberg

Tobias Hamp

Inderjit Singh

Andrew Warren

Ian D Patrick

Daniel Roche

John Shon

Eugene Bolotin

James Casey Geaney

Pam Cheng

Anthony James Cox

Ole Schulz-Trieglaff

Samskruthi Reddy Padigepati

Amirali Kia

Dorna Kashefhaghighi

Wenwu Cui

Jeremy Francis Mcrae

Ole Benjamin Schulz-Trieglaff

A system includes at least a voxelizer, an alternative allele encoder, an evolutionary conservation encoder, and a convolutional neural network. The voxelizer accesses a three-dimensional structure of a reference amino acid sequence of a protein and fits a three-dimensional grid of voxels on atoms in the three-dimensional structure on an amino acid-basis to generate amino acid-wise distance channels. The alternative allele encoder encodes an alternative allele sequence to each voxel in the three-dimensional grid of voxels. The evolutionary conservation encoder encodes an evolutionary conservation sequence to each voxel in the three-dimensional grid of voxels. The convolutional neural network applies three-dimensional convolutions to a tensor that includes the amino acid-wise distance channels encoded with the alternative allele sequence and respective evolutionary conservation sequences and determines a pathogenicity of a variant nucleotide based at least in part on the tensor.

Multi-channel protein voxelization to predict variant pathogenicity using deep convolutional neural networks

The technology disclosed relates to determining pathogenicity of variants. In particular, the technology disclosed relates to generating amino acid-wise distance channels for a plurality of amino acids in a protein. Each of the amino acid-wise distance channels has voxel-wise distance values for voxels in a plurality of voxels. A tensor includes the amino acid-wise distance channels and at least an alternative allele of the protein expressed by a variant. A deep convolutional neural network determines a pathogenicity of the variant based at least in part on processing the tensor. The technology disclosed further augments the tensor with supplemental information like a reference allele of the protein, evolutionary conservation data about the protein, annotation data about the protein, and structure confidence data about the protein.

Artificial intelligence-based analysis of protein three-dimensional (3D) structures

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.

Deep learning-based variant classifier

The technology disclosed relates to splice site prediction and aberrant splicing detection. In particular, it relates to a splice site predictor that includes a convolutional neural network trained on training examples of donor splice sites, acceptor splice sites, and non-splicing sites. An input stage of the convolutional neural network feeds an input sequence of nucleotides for evaluation of target nucleotides in the input sequence. An output stage of the convolutional neural network translates analysis by the convolutional neural network into classification scores for likelihoods that each of the target nucleotides is a donor splice site, an acceptor splice site, and a non-splicing site.

Splicing site classification using neural networks

The technology disclosed presents a deep learning-based framework, which identifies sequence patterns that cause sequence-specific errors (SSEs). Systems and methods train a variant filter on large-scale variant data to learn causal dependencies between sequence patterns and false variant calls. The variant filter has a hierarchical structure built on deep neural networks such as convolutional neural networks and fully-connected neural networks. Systems and methods implement a simulation that uses the variant filter to test known sequence patterns for their effect on variant filtering. The premise of the simulation is as follows: when a pair of a repeat pattern under test and a called variant is fed to the variant filter as part of a simulated input sequence and the variant filter classifies the called variant as a false variant call, then the repeat pattern is considered to have caused the false variant call and identified as SSE-causing.

Deep learning-based framework for identifying sequence patterns that cause sequence-specific errors (SSEs)

Systems, computer-implemented methods, and non-transitory computer readable media are provided for sharing medical data. The disclosed systems may be configured to create a first workgroup having a first knowledgebase. This first knowledgebase may be federated with a common knowledgebase, and with a second knowledgebase of a second workgroup. At least one of the first knowledgebase, common knowledgebase, and second knowledgebase may be configured to store data items comprising associations, signs, and evidence. The signs may comprise measurements and contexts, and the associations may describe the relationships between the measurements and contexts. The evidence may support these associations. The disclosed systems may be configured to receive a request from a user in the first workgroup, retrieve matching data items, and optionally then output to the user at least some of the retrieved matching data items. The request may comprise at least one of a first association and a first measurement.

Federated systems and methods for medical data sharing

We disclose computational models that alleviate the effects of human ascertainment biases in curated pathogenic non-coding variant databases by generating pathogenicity scores for variants occurring in the promoter regions (referred to herein as promoter single nucleotide variants (pSNVs)). We train deep learning networks (referred to herein as pathogenicity classifiers) using a semi-supervised approach to discriminate between a set of labeled benign variants and an unlabeled set of variants that were matched to remove biases.

Deep learning-based pathogenicity classifier for promoter single nucleotide variants (pSNVs)

The technology disclosed relates to constructing a convolutional neural network-based classifier for variant classification. In particular, it relates to training a convolutional neural network-based classifier on training data using a backpropagation-based gradient update technique that progressively match outputs of the convolutional neural network-based classifier with corresponding ground truth labels. The convolutional neural network-based classifier comprises groups of residual blocks, each group of residual blocks is parameterized by a number of convolution filters in the residual blocks, a convolution window size of the residual blocks, and an atrous convolution rate of the residual blocks, the size of convolution window varies between groups of residual blocks, the atrous convolution rate varies between groups of residual blocks. The training data includes benign training examples and pathogenic training examples of translated sequence pairs generated from benign variants and pathogenic variants.

Deep learning-based aberrant splicing detection

Semi-supervised learning for training an ensemble of deep convolutional neural networks

The technology disclosed relates to determining pathogenicity of nucleotide variants. In particular, the technology disclosed relates to specifying a particular amino acid at a particular position in a protein as a gap amino acid, and specifying remaining amino acids at remaining positions in the protein as non-gap amino acids. The technology disclosed further relates to generating a gapped spatial representation of the protein that includes spatial configurations of the non-gap amino acids, and excludes a spatial configuration of the gap amino acid, and determining a pathogenicity of a nucleotide variant based at least in part on the gapped spatial representation, and a representation of an alternate amino acid created by the nucleotide variant at the particular position.

Protein structure-based protein language models

Deep convolutional neural networks to predict variant pathogenicity using three-dimensional (3D) protein structures

Deep learning-based splice site classification

The technology disclosed relates to constructing a convolutional neural network-based classifier for variant classification. In particular, it relates to training a convolutional neural network-based classifier on training data using a backpropagation-based gradient update technique that progressively match outputs of the convolutional network network-based classifier with corresponding ground truth labels. The convolutional neural network-based classifier comprises groups of residual blocks, each group of residual blocks is parameterized by a number of convolution filters in the residual blocks, a convolution window size of the residual blocks, and an atrous convolution rate of the residual blocks, the size of convolution window varies between groups of residual blocks, the atrous convolution rate varies between groups of residual blocks. The training data includes benign training examples and pathogenic training examples of translated sequence pairs generated from benign variants and pathogenic variants.

Aberrant splicing detection using convolutional neural networks (CNNs)

Recurrent neural network-based variant pathogenicity classifier

Growing community of inventors

San Mateo, CA, United States of America

Kai-How Farh