Estimation of phenotypes using DNA, pedigree, and historical data
Disclosed are techniques for predicting a trait of an individual and identifying a set of enriched record collections of a genetic community. To predict a trait of an individual, DNA features and non-DNA features of the individual are accessed to generate a feature vector that is inputted into a machine learning model. The machine learning model generates a prediction of the trait. The prediction may be based on an inheritance prediction and/or a community prediction. To identify a set of enriched record collections, individuals belonging to a genetic community are identified and a set of candidate record collections are accessed. A community count and a background count is determined for each candidate record collection. The set of enriched record collections are identified based on a comparison of the community count and the background count. The genetic community may be annotated using the set of enriched record collections.
1. A computer-implemented method for improving a machine learning model used for trait detection, the computer-implemented method comprising:
accessing a target set of DNA features of a target individual;
identifying, based on the DNA features of the target individual or a family tree which includes the target individual, one or more related individuals who are related to the target individual;
accessing one or more non-DNA features associated with the related individuals;
generating a target feature vector that combines the target set of DNA features of the target individual and the one or more non-DNA features associated with the related individuals, the target feature vector including a target set of numerical values with one or more of the numerical values representing one or more of the DNA features and with one or more of the numerical values representing the one or more of the non-DNA features; and
inputting the target feature vector to a machine learning model to generate a prediction of a trait of the target individual, the prediction being a classification of the trait or a probability that the target individual has the trait, wherein training of the machine learning model comprises:
inputting training samples with training labels and training feature vectors that combine DNA features and non-DNA features of the training samples;
determining, in forward propagation, predicted results of the machine learning model; and
adjusting coefficients of the machine learning model in backpropagation by comparing the predicted results to the training labels.
2. The computer-implemented method of claim 1 , further comprising:
identifying community members who belong to a genetic community of the target individual;
determining a community prediction based on a prevalence of the trait among the community members; and
revising or confirming the prediction of the trait based on the community prediction.
3. The computer-implemented method of claim 1 , wherein identifying, based on the DNA features of the target individual, one or more related individuals who are related to the target individual comprises:
representing the target individual and a plurality of candidate individuals as nodes in an identity-by-descent (IBD) network;
clustering the nodes into a plurality of clusters in the IBD network, each cluster being a genetic community;
identifying as the related individuals one or more of candidate individuals in a cluster to which the target individual belongs.
4. The computer-implemented method of claim 3 , further comprising:
identifying a subset of DNA features that are disproportionately associated with the related individuals in the genetic community to which the target individual belongs; and
identifying a subset of non-DNA features that are disproportionately associated with the related individuals in the genetic community to which the target individual belongs, wherein the target feature vector comprises the subset of non-DNA features and the subset of DNA features.
5. The computer-implemented method of claim 1 , further comprising:
normalizing or standardizing the target set of DNA features and/or the one or more non-DNA features.
6. The computer-implemented method of claim 1 , wherein the trait is an appearance trait, a wellness trait, a health trait, a disease, a preference, behavior, or a language of the target individual.
7. The computer-implemented method of claim 1 , wherein the machine learning model is a regression model, a random forest classifier, a support vector machine, a neural network, or a model trained by an unsupervised approach.
8. The computer-implemented method of claim 1 , wherein at least one feature of the DNA features is determined based on a length of identity-by-descent (IBD) segments shared between the target individual and one of the related individuals.
9. The computer-implemented method of claim 1 , wherein at least one of the related individuals is connected to the target individual through a family tree.
10. The computer-implemented method of claim 1 , wherein the one or more non-DNA features include one or more of the following: customer data, birth year, sex, information about a sequencing array, ethnicity compositions, residency information, socioeconomic information, family tree details, lifestyle, health, covariates, historical records, medical records, survey responses, or digital photographs.
11. The computer-implemented method of claim 1 , wherein training the machine learning model further comprises:
initializing weights of the machine learning model with an initial set of values;
accessing the DNA features of sample individuals of the training samples, each of the sample individuals associated with a label indicating the sample individual trait;
accessing the one or more non-DNA features of the sample individuals;
generating the training feature vectors that include the DNA features and the one or more non-DNA features;
inputting the training feature vectors to the machine learning model to generate predictions of the sample individuals' traits; and
updating weights of the machine learning model based on the predictions and the label associated with each of the sample individuals.
12. A computer-implemented method for improving a machine learning model used for trait detection, the computer-implemented method comprising:
accessing one or more non-DNA features associated with a target individual;
generating a target feature vector that combines the one or more non-DNA features associated with the target individual the target feature vector including a target set of numerical values with one or more of the numerical values representing the one or more non-DNA features; and
inputting the target feature vector to a machine learning model to generate a prediction of a trait of the target individual, the prediction being a classification of the trait or a probability that the target individual has the trait, wherein training of the machine learning model comprises:
inputting training samples with training labels and training feature vectors that non-DNA features of the training samples;
determining, in forward propagation, predicted results of the machine learning model; and
adjusting coefficients of the machine learning model in backpropagation by comparing the predicted results to the training labels.
13. The computer-implemented method of claim 12 , further comprising:
identifying one or more related individuals who are related to the target individual; and
generating the prediction of the trait based additionally on the trait of each related individual and the non-DNA features of the target individual.
14. The computer-implemented method of claim 12 , further comprising:
identifying one or more related individuals who are related to the target individual; and
generating the prediction of the trait based additionally on the one or more non-DNA features of the related individuals.
15. The computer-implemented method of claim 12 , wherein the one or more non-DNA features include one or more of the following of the target individual or their relatives: customer data, birth year, sex, polygenic risk scores, information about a sequencing array, ethnicity compositions, residency information, socioeconomic information, family tree details, lifestyle, health, covariates, historical records, medical records, survey responses, or digital photographs.
16. The computer-implemented method of claim 12 , wherein the trait is an appearance trait, a wellness trait, a health trait, a disease, or a preference, behavior, or a language of the target individual.
17. The computer-implemented method of claim 12 , wherein the probability is binned into percentile ranges.
18. The computer-implemented method of claim 12 , wherein the one or more non-DNA features comprises a plurality of non-DNA features, and the computer-implemented method further comprises normalizing or standardizing the plurality of non-DNA features.
19. The computer-implemented method of claim 12 , wherein training the machine learning model further comprises:
initializing weights of the machine learning model with an initial set of values;
accessing the one or more non-DNA features of sample individuals of the training samples, each of the sample individuals associated with a label indicating the sample individual trait;
generating the training feature vectors that include the one or more non-DNA features of the sample individuals;
inputting the training feature vectors to the machine learning model to generate predictions of each sample individuals' traits; and
updating weights of the machine learning model based on the prediction and the label associated with each of the sample individuals.
20. The computer-implemented method of claim 12 , further comprising:
identifying additional non-DNA features that are disproportionately associated with related individuals in a genetic community to which the target individual belongs, wherein the target feature vector comprises the additional non-DNA features.