Neural network architectures for linking biological sequence variants based on molecular phenotype, and systems and methods therefor

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.

1 .- 36 . (canceled) 
     
     
         37 . A non-transitory computer-readable medium comprising executable instructions stored thereon that, when executed by a processor, are operable to implement a method for determining numerical link distances between two or more biologically related variants, the method comprising:
 a. using an encoder to generate a set of input values digitally representing a variant of the two or more biologically related variants, wherein each of the two or more biologically related variants is derived from a biological sequence through a combination of substitutions, insertions, or deletions to the biological sequence;   b. obtaining at an input layer of a trained molecular phenotype neural network (MPNN), the set of input values generated by the encoder;   c. processing the set of input values by the trained MPNN to generate a set of numerical output values representing a molecular phenotype for the variant, wherein the molecular phenotype comprises numerical elements which quantify biological molecules of cells; and   d. determining, by a comparator, a numerical link distance for pairs of variants of the two or more biologically related variants at least in part by comparing the numerical elements of the molecular phenotypes for the pairs of variants.   
     
     
         38 . The non-transitory computer-readable medium of  claim 37 , wherein the biological sequence is a deoxyribonucleic acid (DNA) sequence, a ribonucleic acid (RNA) sequence, or a protein sequence. 
     
     
         39 . The non-transitory computer-readable medium of  claim 37 , wherein the set of input values corresponds to an encoded representation of a set of contexts. 
     
     
         40 . The non-transitory computer-readable medium of  claim 37 , wherein the method further comprises using the input layer to obtain an additional a set of values digitally representing a set of contexts, wherein the molecular phenotype further comprises numerical elements for at least one of the set of contexts. 
     
     
         41 . The non-transitory computer-readable medium of  claim 37 , wherein the method further comprises using the comparator to determine the numerical link distance for a pair of variants at least in part by applying a function to a difference between the numerical elements of the molecular phenotypes for the pair of variants. 
     
     
         42 . The non-transitory computer-readable medium of  claim 41 , wherein the function is selected from the group consisting of an identity function, a square function, and an absolute value function. 
     
     
         43 . The non-transitory computer-readable medium of  claim 37 , wherein at least one of the two or more biologically related variants comprises:
 a. a DNA sequence, an RNA sequence, or a protein sequence from an individual;   b. a DNA sequence, an RNA sequence, or a protein sequence which is modified by applying a DNA editing system, an RNA editing system, or a protein modification system;   c. a DNA sequence, an RNA sequence, or a protein sequence which is modified by setting one or more nucleotides which are targeted by a therapy to fixed nucleotide values;   d. a DNA sequence, an RNA sequence, or a protein sequence which is modified by setting one or more nucleotides which are targeted by a therapy to values different from existing nucleotide values; or   e. a DNA sequence, an RNA sequence, or a protein sequence which is modified by deleting one or more nucleotides which overlap with nucleotides that are targeted by a therapy.   
     
     
         44 . The non-transitory computer-readable medium of  claim 37 , wherein the molecular phenotype comprises a numerical element selected from the group consisting of: a percentage of transcripts that include an exon; a percentage of transcripts that use an alternative splice site; a percentage of transcripts that use an alternative polyadenylation site; an affinity of an RNA-protein interaction; an affinity of a DNA-protein interaction; a specificity of a microRNA-RNA interaction; and a level of protein phosphorylation. 
     
     
         45 . The non-transitory computer-readable medium of  claim 37 , wherein one or more variants of the two or more biologically related variants are labeled variants, wherein the labeled variants have associated labels, and wherein the method further comprises using a labeling unit to obtain the numerical link distances for the pairs of variants of the two or more biologically related variants from the comparator, and associate labels with unlabeled variants of the two or more biologically related variants based at least in part on the numerical link distances. 
     
     
         46 . The non-transitory computer-readable medium of  claim 45 , further comprising associating each of the unlabeled variants with the associated label of the labeled variant of the labeled variants having a smallest numerical link distance to the unlabeled variant. 
     
     
         47 . The non-transitory computer-readable medium of  claim 46 , wherein the unlabeled variants comprise at least two unlabeled variants, wherein the labels comprise numerical values, and wherein the method further comprises at least partially sorting the unlabeled variants using at least one of the numerical values of the labels. 
     
     
         48 . The non-transitory computer-readable medium of  claim 45 , wherein the method further comprises, for each of the unlabeled variants and for each of the labeled variants, determining a numerical weight for the unlabeled variant and the labeled variant by applying a weighting module to the numerical link distance between the unlabeled variant and the labeled variant; and determining an associated label for the unlabeled variant by summing terms corresponding to the labeled variants, wherein each of the terms is obtained by multiplying the numerical weight for the unlabeled variant and the corresponding labeled variant into the associated label for the corresponding labeled variant. 
     
     
         49 . The non-transitory computer-readable medium of  claim 48 , wherein the method further comprises, for each of the unlabeled variants and for each of the labeled variants, dividing the numerical weight for the unlabeled variant and the labeled variant by a sum of the weights for the unlabeled variant and the labeled variant. 
     
     
         50 . The non-transitory computer-readable medium of  claim 37 , wherein the method further comprises using the comparator to determine, for each of one or more pairs of variants in the two or more biologically related variants, a measure of proximity of the pair of variants within the biological sequence, wherein the numerical link distance is determined at least in part by processing the measure of proximity of the pair of variants with the numerical elements of the molecular phenotypes for the pair of variants. 
     
     
         51 . The non-transitory computer-readable medium of  claim 37 , wherein the weighting unit determines weights differently for different values of the labels. 
     
     
         52 . The non-transitory computer-readable medium of  claim 37 , wherein the method further comprises using the comparator to determine the numerical link distance at least in part by:
 using a trained link neural network to process the numerical elements of the molecular phenotypes for a pair of variants to determine the numerical link distance for the pair of variants.   
     
     
         53 . The non-transitory computer-readable medium of  claim 52 , wherein the method further comprises using the trained link neural network to process additional information pertaining to a similarity of function of the pair of variants. 
     
     
         54 . The non-transitory computer-readable medium of  claim 52 , wherein a set of parameters of the trained link neural network are determined at least in part by applying a training procedure to a dataset of examples, wherein each of the examples comprises a pair of variants and a target value for a link distance of the pair of variants.