Deep Learning Identification of the Gene-Gene Interactions without the Need of thorough Investigation of Each Genomic Combination

• 발행기관: 한국자료분석학회
• 저자명: Jaeyong Yee Mira Park
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.19 No.5, 2277~2285쪽, 전체 9쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2017.10.31

영문 초록

Genomic association has been measured by various statistics including chi-square, balanced accuracy, and mutual information. When investigating the gene-gene interactions, however, estimating the association measure for all of the possible genomic combination becomes computationally intensive rapidly as the order of interaction increases. We show first that the deep learning neural network can be used to measure the strength of the genomic association with case-control phenotype. Then we demonstrate that the selection of the interacting genotypes is possible without going through all of the interacting combinations. Instead of examining each SNP or SNP combination, as is the conventional way, we start from using all of the available SNPs as the input to the neural network. In that way, the association strength by the whole SNPs’ interaction may be obtained. Then each SNP is excluded in turn from the input and the change in accuracy is measured. A set of SNPs that could cause statistically significant change may be identified. By estimating the accuracy changes when excluding the combinations made of this set, we could successfully identify the causal combination from the used genomic data set.

목차

1. Introduction

2. Methods

3. Analysis

4. Conclusion

키워드

deep learning gene-gene interactions accuracy categorical phenotype