Predicting phenotypes using a probabilistic predictor
View Patent ↗Aspects of the subject matter described herein relate to predicting phenotypes. In aspects, a probabilistic predictor is used to summarize a relationship between a set of biological predictors and a phenotype. The probabilistic predictor may use a function that is selected based on the type of the phenotype (e.g., binary, multi-state, or continuous). The probabilistic predictor may use genetic and/or epigenetic information. The probabilistic predictor may be trained on a portion of the data in conjunction with predicting phenotypes in another portion of the data. The probabilistic predictor may be used for various analyses including genome-wide association analysis and gene-set enrichment analysis.
1. A method for determining a relationship between biological predictors and a phenotype, the method implemented at least in part by a computer, the method comprising:
obtaining data regarding a set of biological predictors, the set of biological predictors including a first portion comprising labeled training wherein biological predictors in the first portion have labels to indicate corresponding phenotypes, the set of biological predictors including a second portion;
obtaining an indication of the phenotype;
training a probabilistic predictor using the labeled training data to predict the phenotype based on the set of biological predictors, the probabilistic predictor comprising a statistical model of the set of biological predictors that, given input biological predictors and input phenotypes, computes predictive probabilities of the input biological predictors in relation to the input phenotypes;
applying the trained probabilistic predictor to the second portion of the data to compute predictive probabilities of the phenotype for the second portion of the data, the predictive probabilities comprising probabilities that the biological predictors in the second portion are predictive of the phenotype; and
using statistical significance between an aspect of the predictive probabilities and actual phenotype observations to summarize the relationship between the set of biological predictors and the phenotype.
2. The method of claim 1 , wherein applying the probabilistic predictor comprises using the probabilistic predictor in a genome-wide association analysis.
3. The method of claim 1 , wherein applying the probabilistic predictor comprises using the probabilistic predictor in a gene-set enrichment analysis.
4. The method of claim 1 , wherein the biological predictors comprise genetic predictors.
5. The method of claim 4 , wherein the genetic predictors comprise single-nucleotide polymorphisms.
6. The method of claim 1 , wherein the biological predictors comprise epigenetic predictors.
7. The method of claim 1 , wherein applying the probabilistic predictor comprises using an L1-regularized function.
8. The method of claim 1 , wherein applying the probabilistic predictor comprises using a statistical test known as the Mann-Whitney test.
9. The method of claim 1 , wherein applying the probabilistic predictor comprises using a statistical test according to Spearman correlation.
10. The method of claim 1 , wherein the aspect comprises a mean of a distribution or a probability of a discrete value.
11. The method of claim 1 , wherein obtaining data regarding a set of biological predictors comprises obtaining data regarding a metabolic pathway associated with the set of biological predictors.
12. The method of claim 1 , further comprising displaying a value indicative of the relationship.
13. In a computing environment, an apparatus for determining a relationship between biological predictors and a target phenotype, the apparatus comprising:
a store that includes data regarding a set of biological predictors, the set of biological predictors including a first portion comprising labeled training wherein biological predictors in the first portion have labels to indicate corresponding phenotypes, the set of biological predictors including a second portion;
a phenotype receiver operable to obtain the target phenotype;
a probabilistic predictor operable to summarize the relationship between the second portion of the set of biological predictors and the target phenotype; and
a trainer operable to train the probabilistic predictor on a portion of the data using the labeled training data on the set of biological predictors, the probabilistic predictor comprising a statistical model of the set of biological predictors that, given input biological predictors and input phenotypes, computes predictive probabilities of the input biological predictors in relation to the input phenotypes.
14. The apparatus of claim 13 , wherein the probabilistic predictor is operable to use an L1-regularized logistic regression function to assist in summarizing the relationship.
15. The apparatus of claim 13 , wherein the probabilistic predictor is operable to use a function that implements the Mann-Whitney test to assist in summarizing the relationship.
16. The apparatus of claim 13 , wherein the probabilistic predictor uses a statistical test according to Spearman correlation to assist in summarizing the relationship.
17. The apparatus of claim 13 , further comprising a user interface operable to receive indications of biological predictors and phenotypes.
18. The apparatus of claim 13 , wherein the biological predictors comprise genetic predictors.
19. The apparatus of claim 13 , wherein the biological predictors comprise epigenetic predictors.
20. A computer storage medium having computer-executable instructions for determining a relationship between biological predictors and a phenotype, the computer-executable instructions when executed performing actions, comprising:
obtaining data regarding a set of biological predictors that are genetic and/or epigenetic in nature, the biological predictors related to a metabolic pathway, the set of biological predictors including a first portion comprising labeled training wherein biological predictors in the first portion have labels to indicate corresponding phenotypes, the set of biological predictors including a second portion;
obtaining an indication of the phenotype, the phenotype comprising a characteristic of an organism, the phenotype having a nature of binary, multi-state, or continuous;
selecting a function to use for a probabilistic predictor based on the nature of the phenotype, the function corresponding to one of a L1-regularized function, Mann-Whitney test, and Spearman correlation;
training the probabilistic predictor based on a labeled portion of the data, the labeled portion comprising biological predictors having labels indicating phenotypes associated therewith; and
using the probabilistic predictor to summarize the relationship between the set of biological predictors and the phenotype.