Community assignments in identity by descent networks and genetic variant origination
Disclosed are techniques for characterizing variants of interest and predicting assignments of individuals to communities based on obtained genetic information. To characterize a variant, DNA datasets of reference individuals are accessed and used to generate a cluster with additional individuals. Reference individuals carry a variant at a genetic locus and the additional individuals share IBD with reference individuals. Statistics of genealogical data of the cluster are generated. A result summarizing the characterization of the variant is generated based on the statistics. To determine if an individual belongs to a community, a subset of the individual's haplotypes are inputted into a community-specific model. The model is trained using the training samples that each include haplotypes of reference individuals and a label identifying whether the reference individual belongs to the community. Based on the output of the model, it is determined whether the individual is a member of the community.
1. A method comprising:
obtaining a DNA dataset from an individual;
determining genotypes of the individual based on the DNA dataset;
phasing the genotypes to generate haplotypes of the individual;
retrieving a plurality of community-specific random forest machine learning models, each community-specific random forest machine learning model is specific to a community, each community-specific random forest machine learning model is trained by enriched haplotype features of reference individuals and labels identifying whether the reference individuals belong to the community, a reference individual who belongs to the community having the enriched haplotype features that are representative of the community;
extracting sets of community-specific features from the haplotypes of the individuals, wherein each set comprises corresponding enriched haplotype features of the individual specific to the community;
inputting, for each community-specific random forest machine learning model, the set of community-specific features into the community-specific random forest machine learning model, to generate a community-specific score; and
determining whether the individual is a member of a particular community based on community-specific scores generated by the plurality of community-specific random forest machine learning models.
2. The method of claim 1 , wherein one of the community-specific random forest machine learning models is further trained by at least:
phasing genotypes of reference individuals;
identifying common haplotypes at each window of the genotypes;
performing an enrichment analysis on the common haplotypes to identify a set of enriched haplotypes;
generating a feature vector for each reference individual, the feature vector has a set of binary elements, each associated with an enriched haplotype, a value of each binary element indicating whether the reference individual has the enriched haplotype;
generating a data frame that includes the reference individuals with their feature vector and the label identifying whether the reference individual belongs to the community;
applying the community-specific random forest machine learning model to the data frame, the enriched haplotypes are features of the model; and
adjusting parameters of the community-specific random forest machine learning model based on a performance of the community-specific random forest machine learning model.
3. The method of claim 1 , wherein each community-specific random forest machine learning model is trained by training samples, and a positive training sample of the training samples is generated by at least:
phasing a DNA dataset of one of the reference individuals who belongs to the community to generate haplotypes of the reference individuals;
performing an enrichment analysis on the haplotypes with respect to the community;
identifying one or more groups of haplotypes of the reference individual that are representative of the community;
extracting the one or more groups of haplotypes as the positive training sample; and
associating the positive training sample with a positive label that the reference individual belongs to the community.
4. The method of claim 3 , wherein a negative training sample of the training samples is generated by at least:
retrieving a DNA dataset of a reference individual who is known not belonging to the community;
extracting one or more groups of haplotypes as the negative training sample, the extracted one or more groups of haplotypes being at same genetic loci of the one or more groups of haplotypes of the one of the reference individuals who belongs to the community; and
associating the negative training sample with a negative label that the reference individual does not belong to the community.
5. The method of claim 1 , wherein the enriched haplotype features are identified using an enrichment analysis to determine which haplotypes are more likely to be observed in a community.
6. The method of claim 1 , wherein one of the community-specific random forest machine learning models is a non-linear classifier selected from a group consisting of a support vector machine, one or more decision trees, or a neural network.
7. A computer-readable storage medium having stored thereon computer-executable instructions that, when executed by one or more processors, causes the one or more processors to perform at least the following:
obtain a DNA dataset from an individual;
determine genotypes of the individual based on the DNA dataset;
phase the genotypes to generate haplotypes of the individual;
retrieve a plurality of community-specific random forest machine learning models, each community-specific random forest machine learning model is specific to a community, each community-specific random forest machine learning model is trained by enriched haplotype features of reference individuals and labels identifying whether the reference individuals belong to the community, a reference individual who belongs to the community having the enriched haplotype features that are representative of the community;
extract sets of community-specific features from the haplotypes of the individuals, wherein each set comprises corresponding enriched haplotype features of the individual specific to the community;
input, for each community-specific random forest machine learning model, the set of community-specific features into the community-specific random forest machine learning model, to generate a community-specific score; and
determine whether the individual is a member of a particular community based on community-specific scores generated by the plurality of community-specific random forest machine learning models.
8. The computer-readable storage medium of claim 7 , wherein one of the community-specific random forest machine learning models is further trained by at least:
phasing genotypes of reference individuals;
identifying common haplotypes at each window of the genotypes;
performing an enrichment analysis on the common haplotypes to identify a set of enriched haplotypes;
generating a feature vector for each reference individual, the feature vector has a set of binary elements, each associated with an enriched haplotype, a value of each binary element indicating whether the reference individual has the enriched haplotype;
generating a data frame that includes the reference individuals with their feature vector and the label identifying whether the reference individual belongs to the community;
applying the community-specific random forest machine learning model to the data frame, the enriched haplotypes are features of the model; and
adjusting parameters of the community-specific random forest machine learning model based on a performance of the community-specific random forest machine learning model.
9. The computer-readable storage medium of claim 7 , wherein each community-specific random forest machine learning model is trained by training samples, and a positive training sample of the training samples is generated by at least:
phasing a DNA dataset of one of the reference individuals who belongs to the community to generate haplotypes of the reference individuals;
performing an enrichment analysis on the haplotypes with respect to the community;
identifying one or more groups of haplotypes of the reference individual that are representative of the community;
extracting the one or more groups of haplotypes as the positive training sample; and
associating the positive training sample with a positive label that the reference individual belongs to the community.
10. The computer-readable storage medium of claim 9 , wherein a negative training sample of the training samples is generated by at least:
retrieving a DNA dataset of a reference individual who is known not belonging to the community;
extracting one or more groups of haplotypes as the negative training sample, the extracted one or more groups of haplotypes being at same genetic loci of the one or more groups of haplotypes of the one of the reference individuals who belongs to the community; and
associating the negative training sample with a negative label that the reference individual does not belong to the community.
11. The computer-readable storage medium of claim 7 , wherein
the enriched haplotype features are identified using an enrichment analysis to determine which haplotypes are more likely to be observed in a community.
12. The computer-readable storage medium of claim 7 , wherein one of the community-specific random forest machine learning models is a non-linear classifier selected from a group consisting of a support vector machine, one or more decision trees, or a neural network.
13. A computer system for determining whether an individual is a member of a target genetic community, the computer system comprising:
one or more processors;
a hardware storage device having stored thereon computer-executable instructions that, when executed by the one or more processors, causes the computer system to perform at least the following:
determine genotypes of an individual based on an associated DNA dataset for the individual;
phase the determined genotypes to generate a set of haplotypes of the individual;
retrieve a plurality of community-specific random forest machine learning models, each community-specific random forest machine learning model is specific to a community, each community-specific random forest machine learning model is trained by enriched haplotype features of reference individuals and labels identifying whether the reference individuals belong to the community, a reference individual who belongs to the community having the enriched haplotype features that are representative of the community;
extract sets of community-specific features from the haplotypes of the individuals, wherein each set comprises corresponding enriched haplotype features of the individual specific to the community;
input, for each community-specific random forest machine learning model, the set of community-specific features into the community-specific random forest machine learning model, to generate a community-specific score; and
determine whether the individual is a member of a particular community based on community-specific scores generated by the plurality of community-specific random forest machine learning models.
14. The computer system of claim 13 , wherein the output of the one of the community-specific random forest machine learning models comprises a numerical probability that the individual is a member of the target genetic community.
15. The computer system of claim 14 , wherein the individual is determined to be a member of the particular community if the numerical probability exceeds a threshold numerical value.
16. The computer system of claim 15 , wherein the output of the one of the community-specific random forest machine learning models further comprises a binary classification that the individual is a member of the target genetic community, the binary classification based on whether the numerical probability exceeds the threshold numerical value.
17. The computer system of claim 13 , wherein one of the community-specific random forest machine learning models is a non-linear classifier selected from the group consisting of a support vector machine, one or more decision trees, and a neural network.
18. The computer system of claim 13 , wherein the enriched haplotype features comprise haplotypes having differing window lengths of single nucleotide polymorphisms.
19. The computer system of claim 13 , wherein the enriched haplotype features comprise haplotypes having a window length of at least 64 single nucleotide polymorphisms.
20. The computer system of claim 13 , wherein one of the community-specific random forest machine learning models is further trained by at least:
phasing genotypes of reference individuals;
identifying common haplotypes at each window of the genotypes;
performing an enrichment analysis on the common haplotypes to identify a set of enriched haplotypes;
generating a feature vector for each reference individual, the feature vector has a set of binary elements, each associated with an enriched haplotype, a value of each binary element indicating whether the reference individual has the enriched haplotype;
generating a data frame that includes the reference individuals with their feature vector and the label identifying whether the reference individual belongs to the community;
applying the community-specific random forest machine learning model to the data frame, the enriched haplotypes are features of the model; and
adjusting parameters of the community-specific random forest machine learning model based on a performance of the community-specific random forest machine learning model.