Feature selection for efficient epistasis modeling for phenotype prediction

Patent metadata

Field	Value
Publication number	US-10108775-B2
Application number	US-201314030743-A
Country	US
Kind code	B2
Filing date	Sep 18, 2013
Priority date	Jan 21, 2013
Publication date	Oct 23, 2018
Grant date	Oct 23, 2018

Abstract

Various embodiments select markers for modeling epistasis effects. In one embodiment, a processor receives a set of genetic markers and a phenotype. A relevance score is determined with respect to the phenotype for each of the set of genetic markers. A threshold is set based on the relevance score of a genetic marker with a highest relevancy score. A relevance score is determined for at least one genetic marker in the set of genetic markers for at least one interaction between the at least one genetic marker and at least one other genetic marker in the set of genetic markers. The at least one interaction is added to a top-k feature set based on the relevance score of the at least one interaction satisfying the threshold.

First claim

What is claimed is:

1. An information processing system for reducing search time of features in sample data and reducing computation time when training a model for generating specialized data based on the features, the information processing system comprising:
a memory;
a processor communicatively coupled to the memory; and
a feature selection circuit coupled to the memory and the processor, wherein the feature selection circuit is configured to perform a plurality of operations comprising:
receiving a set of genetic markers and a phenotype from at least one external information processing system;
training an epistasis effect model based on the set of genetic markers and the phenotype;
reducing computation time of the information processing system during the training of the epistasis effect model and further reducing feature selection time of the epistasis effect model, wherein reducing the computation time and the feature selection time comprises:
determining, for each of the set of genetic markers, a relevance score with respect to the phenotype according to I(x j training ;c training ), where I is mutual information between a given genetic marker x j and a phenotype c, where mutual information I between two variables x and y is defined, based on their joint marginal probabilities p(x) and p(y) and probabilistic distribution p(x, y), as:

I
⁡

(

x
,
y

)

=

∑

i
,
j

⁢

p
⁡

(

x
i

,

y
i

)

⁢
log
⁢

p
⁢

(

x
i

,

y
i

)

p
⁡

(

x
i

)

⁢

p
⁡

(

y
i

)

;

setting a threshold based on the relevance score of a genetic marker in the set of genetic markers with a highest relevancy score;
determining, for at least one individual genetic marker in the set of genetic markers having a relevance score satisfying the threshold, a relevance score for at least one interaction between the at least one individual genetic marker and at least one other individual genetic marker in the set of genetic markers;
adding the at least one interaction to a top-k feature set based on the relevance score of the at least one interaction satisfying the threshold, wherein the top-k feature set comprises one or more markers and one or more interactions, and wherein each of the one or more genetic markers and each of the one or more interactions comprises a top-k relevance score;
identifying a subset of the top-k feature set based on the set of genetic markers, wherein each feature in the subset of the top-k feature set maximizes a relevancy with the phenotype and minimizes a redundancy with respect to other selected features; and
training the epistasis effect model utilizing at least the subset of the top-k feature set, wherein the epistasis effect model predicts phenotypes for genetic markers;

storing the trained epistasis effect model in memory;
electronically obtaining, a new set of a set of genetic markers that is associated phenotype data; and
executing the trained epistasis effect model, wherein executing the trained epistasis effect model comprises
inputting the new set of genetic markers into the trained epistasis effect model; and
outputting a phenotype for the new set of genetic markers, wherein the phenotype that was outputted was not made available to the feature selection circuit as part of the new set of genetic markers.

2. The information processing system of claim 1 , wherein the method further comprises:
randomly sampling a subset of genetic markers from the set of genetic markers; and
selecting the at least one additional genetic marker from the subset of genetic markers.

3. The information processing system of claim 2 , wherein determining the relevance score of the at least one interaction comprises:
determining a first set of relevance scores comprising a relevance score with respect to the phenotype for each of a first plurality of interactions between the at least one genetic marker and each of the subset of genetic markers;
determining, based on a normal distribution associated with the first set of relevance scores, a probability of the at least one genetic marker being associated with an interaction comprising a relevance score greater than the threshold;
comparing the probability to a probability threshold; and
determining, based on the probability satisfying the probability threshold, a second set of relevance scores comprising a relevance score for each of a second plurality of interactions between the at least one genetic marker and a remaining set of genetic markers in the set of genetic markers, wherein the second plurality of interactions comprises the at least one interaction, and wherein the remaining set of genetic markers comprises the at least one additional genetic marker.

4. The information processing system of claim 1 , wherein the method further comprises:
generating, based on adding the at least one interaction to the top-k feature set, an updated top-k feature set by removing one of a genetic marker and an interaction associated with a lowest relevance score from the top-k feature set.

5. The information processing system of claim 1 , wherein the method further comprises:
updating the thres

Assignees

• IBM

Inventors

• Haws David
• He Dan
• Parida Laxmi P

Classifications

• G06F19/18

  Physics · mapped topic

• G06F19/12 Primary

  Physics · mapped topic

• G06F19/24

  Physics · mapped topic

• G16B5/20 Primary

  Probabilistic models · CPC title

• G16B40/00

  ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding · CPC title