Deep learning network for evolutionary conservation

Abstract

The technology disclosed relates to a deep learning network system for evolutionary conservation prediction. In one implementation, the system includes a first model for processing a first multiple sequence alignment that aligns a query sequence with a masked base at a target position to N non-query sequences and predicting a first identity of the masked base at the target position. The system also includes a second model for processing a second multiple sequence alignment that aligns the query sequence to M non-query sequences, where M>N, and predicting a second identity of the masked base at the target position. The system further includes an evolutionary conservation determination logic configured to measure an evolutionary conservation of the masked base at the target position based on the first and second identities of the masked base.

What we claim is: 
     
         1 . A system, comprising:
 a first model configured to
 process a first multiple sequence alignment that aligns a query sequence to N non-query sequences, wherein the query sequence includes a masked base at a target position that is flanked by right and left bases, and 
 predict a first identity of the masked base at the target position; 
   a second model configured to
 process a second multiple sequence alignment that aligns the query sequence to M non-query sequences, where M>N, 
 predict a second identity of the masked base at the target position; and 
   an evolutionary conservation determination logic configured to measure an evolutionary conservation of the masked base at the target position based on the first and second identities of the masked base.   
     
     
         2 . The system of  claim 1 , wherein the N non-query sequences are N closest homologues to the query sequence. 
     
     
         3 . The system of  claim 1 , wherein the M non-query sequences are M closest homologues to the query sequence. 
     
     
         4 . The system of  claim 1 , wherein the M non-query sequences include the N non-query sequences. 
     
     
         5 . The system of  claim 1 , wherein the query sequence belongs to a first species. 
     
     
         6 . The system of  claim 5 , wherein the first species is human. 
     
     
         7 . The system of  claim 5 , wherein the N non-query sequences are non-human primates. 
     
     
         8 . The system of  claim 7 , wherein the N non-query sequences belong to a first group of species that shares a family with the first species. 
     
     
         9 . The system of  claim 8 , wherein the family is hominids. 
     
     
         10 . The system of  claim 8 , wherein the first group of species shares an order with the first species. 
     
     
         11 . The system of  claim 10 , wherein the order is primates. 
     
     
         12 . The system of  claim 5 , wherein the M non-query sequences belong to a second group of species that shares a class with the first species. 
     
     
         13 . The system of  claim 12 , wherein the class is mammals. 
     
     
         14 . The system of  claim 12 , wherein the second group of species shares a phylum with the first species. 
     
     
         15 . The system of  claim 14 , wherein the phylum is chordates. 
     
     
         16 . The system of  claim 12 , wherein the second group of species shares a kingdom with the first species. 
     
     
         17 . The system of  claim 16 , wherein the kingdom is animals. 
     
     
         18 . The system of  claim 1 , wherein a first average evolutionary distance determined from evolutionary distances between species in a first group of species to which the N non-query sequences belong is less than a second average evolutionary distance determined from evolutionary distances between species in a second group of species to which the M non-query sequences belong. 
     
     
         19 . The system of  claim 18 , wherein non-overlapping species between the first and second groups of species are more evolutionarily distant from the first species than overlapping species between the first and second groups of species. 
     
     
         20 . The system of  claim 1 , wherein the first model is further configured to predict the first identity as a first probability distribution that specifies base-wise likelihoods of the masked base at the target position being adenine (A), cytosine (C), thymine (T), and guanine (G). 
     
     
         21 . The system of  claim 20 , wherein the first probability distribution further specifies base-wise likelihoods of the masked base at the target position being A, C, T, G, an unknown base (X), and a missing base (-). 
     
     
         22 . The system of  claim 1 , wherein the second model is further configured to predict the second identity as a second probability distribution that specifies base-wise likelihoods of the masked base at the target position being A, C, T, G. 
     
     
         23 . The system of  claim 22 , wherein the second probability distribution further specifies base-wise likelihoods of the masked base at the target position being A, C, T, G, -, and X. 
     
     
         24 . The system of  claim 23 , wherein the evolutionary conservation determination logic is further configured to measure the evolutionary conservation of the masked base at the target position based on the first and second probability distributions. 
     
     
         25 . The system of  claim 24 , wherein the evolutionary conservation determination logic is further configured to use t-test statistics to measure the evolutionary conservation of the masked base at the target position. 
     
     
         26 . The system of  claim 1 , wherein the first model is further configured to encode a background mutation probability estimation in the first identity of the masked base at the target position. 
     
     
         27 . The system of  claim 1 , wherein the second model is further configured to encode a group-specific mutation probability estimation in the second identity of the masked base at the target position.