Neural network architectures for linking biological sequence variants based on molecular phenotype, and systems and methods therefor
Inventors
Frey, Brendan • Delong, Andrew
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11183271-B2
Publication Date
2021-11-23
Expiration Date
Abstract
We describe a system and a method that ascertains the strengths of links between pairs of biological sequence variants, by determining numerical link distances that measure the similarity of the molecular phenotypes of the variants. The link distances may be used to associate knowledge about labeled variants to other variants and to prioritize the other variants for subsequent analysis or interpretation. The molecular phenotypes are determined using a neural network, called a molecular phenotype neural network, and may include numerical or descriptive attributes, such as those describing protein-DNA interactions, protein-RNA interactions, protein-protein interactions, splicing patterns, polyadenylation patterns, and microRNA-RNA interactions. Linked genetic variants may be used to ascertain pathogenicity in genetic testing, to identify drug targets, to identify patients that respond similarly to a drug, to ascertain health risks, or to connect patients that have similar molecular phenotypes.
Core Innovation
The invention provides a system for determining numerical link distances between two or more biologically related variants. The system includes one or more trained molecular phenotype neural networks (MPNNs) that generate a molecular phenotype for each variant as one or more output values, and each molecular phenotype comprises numerical elements that quantify biological molecules of cells.
Biologically related variants are represented as input values derived from a biological sequence through substitutions, insertions, or deletions. The input layer of the MPNN obtains digitally represented variant information from DNA, RNA, or protein sequences, and one or more feature detectors obtain input values either from the input layer or from outputs of other feature detectors.
The comparator determines numerical link distances for pairs of variants based at least in part on differences between the numerical elements of the molecular phenotypes. The numerical link distances connect variants by comparing predicted molecular-phenotype numerical elements between variant pairs, and a trained link neural network may optionally be used for computing the numerical link distance while also incorporating additional similarity information such as genomic proximity, expression/splicing QTLs, and linkage disequilibrium.
Claims Coverage
The document includes two independent claims, one system claim and one computer-implemented method claim. Across these claims, the inventive coverage centers on producing numerical molecular phenotypes from sequence-derived variant inputs using trained MPNNs, and determining numerical link distances for variant pairs using a comparator based on differences between molecular-phenotype numerical elements.
Sequence-derived variant inputs to a trained MPNN
The input layer obtains one or more input values digitally representing a variant, where the variant is derived from a biological sequence through substitutions, insertions, or deletions, and the biological sequence is a DNA sequence, an RNA sequence, or a protein sequence.
Molecular phenotype output with numerical elements quantifying cell biological molecules
Feature detectors obtain input values from the input layer or from output values of another feature detector, and the output layer outputs one or more output values representing a molecular phenotype for the variant, where the molecular phenotype comprises numerical elements that quantify biological molecules of cells.
Comparator determining numerical link distance from differences between molecular-phenotype numerical elements
A comparator obtains the output values of the output layer of each trained MPNN and determines a numerical link distance for pairs of variants based at least in part on a difference between the numerical elements of the molecular phenotypes for the pairs of variants.
Determine numerical link distances via trained MPNN processing and comparator difference
A computer-implemented method obtains input values digitally representing sequence-derived variants, processes the input values with a trained MPNN to generate molecular phenotypes comprising numerical elements quantifying biological molecules of cells, and determines numerical link distances for variant pairs with a comparator based at least in part on differences between the numerical elements of the molecular phenotypes.
Overall, the independent claim coverage is directed to using trained MPNNs to generate numerical molecular-phenotype elements from DNA, RNA, or protein sequence-derived variant inputs, and to using a comparator to compute numerical link distances for variant pairs based on differences between those numerical molecular-phenotype elements. Dependent claims further refine the comparator computation and add label propagation and optional trained link neural network processing.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Variant prioritization for genetic testing.
Pathogenicity assessment.
Drug target identification.
Patient stratification.
Health-risk assessment.
Connecting patients with similar molecular phenotypes.
Interested in licensing this patent?