System and method for genomic association

Abstract

In variants, a method for genomic association can include: determining observed variable values and observed phenotype values for each organism in a population, removing information from variables of interest, determining a phenotype-variable association model, identifying causal variables associated with a phenotype, and/or any other suitable steps.

We claim: 
     
         1 . A method, comprising:
 a) for each of a set of organisms:
 determining a trait value for a trait; and 
 determining variable values for a set of variables; 
   b) selecting a subset of variables from the set of variables; and   c) determining a first model configured to predict values for variables of interest in the set of variables based on values for the subset of variables;   d) determining test variable values for the variables of interest using the first model;   e) using the test variable values, determining a second model comprising a relationship between the set of variables and the trait; and   f) identifying causal variables from the set of variables based on the second model.   
     
     
         2 . The method of  claim 1 , further comprising training the first model using values for a second set of variables, different from the subset of variables. 
     
     
         3 . The method of  claim 2 , wherein the first model comprises an autoencoder. 
     
     
         4 . The method of  claim 1 , wherein determining the first model comprises a fitting a linear regression based on the variable values for the set of variables. 
     
     
         5 . The method of  claim 1 , further comprising clustering the set of variables based on autocorrelation analysis of the variable values, wherein the subset of variables and variables of interest are from a shared cluster. 
     
     
         6 . The method of  claim 5 , wherein variables in the set of variables comprise k-mers. 
     
     
         7 . The method of  claim 1 , wherein (b)-(c) are iteratively repeated until a model fit metric for the first model rises above a threshold. 
     
     
         8 . The method of  claim 7 , further comprising, in a first iteration, segmenting the subset of variables into high importance variables and low importance variables based on the first model, wherein, in a second iteration, selecting the subset of variables comprises replacing the low importance variables. 
     
     
         9 . The method of  claim 1 , wherein a size of the subset of variables is less than a size of the set of organisms. 
     
     
         10 . The method of  claim 1 , further comprising:
 determining target values for the causal variables based on a target trait; and   based on the target values, breeding organisms in the set of organisms to generate a new organism with the target trait.   
     
     
         11 . The method of  claim 1 , wherein variables in the set of variables comprise variables for at least one of: loci, gene expression, protein expression, methylation, environmental parameters, or protein binding affinity. 
     
     
         12 . A method, comprising:
 for each of a set of organisms:
 determining a trait value for a trait; and 
 determining variable values associated with the trait value for a set of variables; 
   transforming values for a subset of variables to a reduced dimension space;   determining transformed test values for a variable of interest based on the transformed values for the subset of variables;   using the transformed values for the subset of variables and the transformed test values, determining a second model comprising a relationship between transformed variables and the trait;   identifying transformed causal variables based on the second model; and   decoding the transformed causal variables to determine causal variables.   
     
     
         13 . The method of  claim 12 , wherein the reduced dimension space comprises a latent space, wherein transforming values for the subset of variables comprises training an autoencoder to encode variable values to the latent space, wherein the trained autoencoder is used to decode the transformed causal variables. 
     
     
         14 . The method of  claim 13 , wherein the autoencoder is trained using values for a second subset of variables, different from the subset of variables. 
     
     
         15 . The method of  claim 12 , further comprising determining a distribution of the subset of transformed variables, wherein determining transformed test values for the variable of interest comprises selecting transformed test values from the distribution. 
     
     
         16 . The method of  claim 12 , wherein transforming values for a subset of variables comprises:
 training a neural network to predict a trait value based on the respective variable values; and   using a first layer of the trained neural network to transform the values for the subset of variables.   
     
     
         17 . The method of  claim 12 , further comprising selecting organisms from the set of organisms for cross-breeding based on the causal variables and a target trait value. 
     
     
         18 . The method of  claim 12 , wherein the set of variables comprise genomic variables, wherein the subset of variables in the reduced dimension space comprises a set of features, wherein the transformed values comprise feature values, wherein the variable of interest comprises a feature of interest, wherein the transformed test values comprise test feature values for the feature of interest, wherein the second model comprises a relationship between the feature of interest, the set of features, and the trait, wherein the transformed causal variables comprise causal features of interest, and wherein decoding the transformed causal variables comprises decoding the causal features of interest into causal genomic variables.