INTER-MODEL PREDICTION SCORE RECALIBRATION
The technology disclosed relates to inter-model prediction score recalibration. In one implementation, the technology disclosed relates to a system including a first model that generates, based on evolutionary conservation summary statistics of amino acids in a target protein sequence, a first pathogenicity score-to-rank mapping for a set of variants in the target protein sequence; and a second model that generates, based on epistasis expressed by amino acid patterns spanning the target protein sequence and a plurality of non-target protein sequences aligned in multiple sequence alignment, a second pathogenicity score-to-rank mapping for the set of variants. The system also includes a reassignment logic that reassigns pathogenicity scores from the first set of pathogenicity scores to the set of variants based on the first and second score-to-rank mappings, and an output logic to generate a ranking of the set of variants based on the reassigned scores.
1 . A system, comprising:
a first model configured to generate, based in part on evolutionary conservation summary statistics of amino acids in a target protein sequence,
a first score-to-rank mapping that maps a first set of pathogenicity scores for a set of variants observed in the target protein sequence to a first set of score rankings;
a second model configured to generate, based in part on epistasis expressed by amino acid patterns spanning the target protein sequence and a plurality of non-target protein sequences aligned with the target protein sequence in a multiple sequence alignment,
a second score-to-rank mapping that maps a second set of pathogenicity scores for the set of variants to a second set of score rankings;
a reassignment logic configured to reassign pathogenicity scores from the first set of pathogenicity scores to the set of variants based on the first and second score-to-rank mappings; and
an output logic configured to generate a ranking of the set of variants based on the reassigned pathogenicity scores.
2 . The system of claim 1 , wherein the first score-to-rank mapping assigns a given variant in the set of variants a given pathogenicity score from the first set of pathogenicity scores.
3 . The system of claim 2 , wherein the second score-to-rank mapping assigns the given variant a given score ranking from the second set of score rankings.
4 . The system of claim 3 , wherein the first score-to-rank mapping assigns the given score ranking an another pathogenicity score from the first set of pathogenicity scores, wherein the another pathogenicity score is different from the given pathogenicity score.
5 . The system of claim 4 , wherein the reassignment logic is further configured to reassign the given variant the another pathogenicity score.
6 . The system of claim 5 , further configured to comprise a combination logic configured to combine the another pathogenicity score and the given pathogenicity score to generate a combined pathogenicity score.
7 . The system of claim 6 , wherein the reassignment logic is further configured to reassign the given variant the combined pathogenicity score.
8 . The system of claim 6 , wherein the combined pathogenicity score is an average of the another pathogenicity score and the given pathogenicity score.
9 . The system of claim 8 , wherein the combined pathogenicity score is a weighted average of the another pathogenicity score and the given pathogenicity score.
10 . The system of claim 9 , wherein weights used for the weighted average are preset and respectively specified for the first model and the second model.
11 . The system of claim 10 , wherein the weights correspond to respective rankings of the another pathogenicity score and the given pathogenicity score.
12 . The system of claim 6 , wherein the combined pathogenicity score is a sum of the another pathogenicity score and the given pathogenicity score.
13 . The system of claim 12 , wherein the combined pathogenicity score is a weighted sum of the another pathogenicity score and the given pathogenicity score.
14 . The system of claim 13 , wherein weights used for the weighted sum are preset and respectively specified for the first model and the second model.
15 . The system of claim 14 , wherein the weights correspond to respective rankings of the another pathogenicity score and the given pathogenicity score.
16 . The system of claim 1 , wherein the first model uses a first scale of pathogenicity scores to differentiate pathogenic variants from benign variants.
17 . The system of claim 16 , wherein the first scale ranges from 0 to 1.
18 . The system of claim 1 , wherein the second model uses a second scale of pathogenicity scores to differentiate pathogenic variants from benign variants.
19 . The system of claim 18 , wherein the second scale ranges from wherein the second scale ranges from a maximum real number represented digitally to a minimum real number represented digitally.
20 . The system of claim 1 , wherein the first model is further configured to generate the first set of pathogenicity scores based in part on three-dimensional (3D) structural information about the amino acids in the target protein sequence.
21 . The system of claim 1 , wherein the first model uses voxelized features as input.
22 . The system of claim 1 , wherein the first model is a site-independent model that factorizes single-position variations in a plurality of aligned sequences.
23 . The system of claim 22 , wherein the first model is a pairwise-interaction model that factorizes two-position variations in the plurality of aligned sequences.
24 . The system of claim 23 , wherein the second model is a non-linear latent variable model that posts hidden variables to jointly detect global patterns and local patterns of sequence variations across windows spanning multiple positions and multiple sequences in the plurality of aligned sequences.