Abstract
SAERMA: Stacked Autoencoder Rule Mining Algorithm for the Interpretation of Epistatic Interactions in GWAS for Extreme Obesity
Highlights
Understanding the genetic architecture of common diseases remains a significant challenge
RESULTS the results are presented using the proposed methodology outlined above. This is reported in four experiments conducted after QC and association analysis (Statistical filtering): 1) Baseline classification with generalised linear model (GLM) using Single Nucleotide Polymorphisms (SNPs) with P-value < 1 × 10−2; 2) multi-layer perceptron neural network (MLPNN) classification using SNPs with P-value < 1 × 10−2; 3) stacked autoencoders (SAEs)-based classification using non-linear SNP-SNP interactions with P-values < 1 × 10−2; and 4) our proposed approach, SAERMA
The minute non-linear transformations of the input space that occur in the autoencoders, makes it is very difficult to trace the amount of variance they contribute from case-control data
Summary
Understanding the genetic architecture of common diseases remains a significant challenge. Single nucleotide polymorphisms (SNPs) [2] are the most common type of genetic variation among humans. GWAS implements single-loci analysis where SNPs are independently tested for association with phenotypes of interest, without consideration of the interactions that occur between loci. This is a major limitation in GWAS, when studying complex disorders caused by SNP-SNP, gene-gene and gene-environment interactions. To better understand the missing heritability inherent in GWAS it is necessary to examine epistasis interactions [4] This approach assumes that genes do not work independently but create ‘‘gene networks’’ that have major effects on tested phenotypes.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
