Expression quantitative locus mapping for identification of hotspots using an empirical Bayes mixture model

Abstract

Identification of genomic regions that regulate gene expression can help our understanding of the mechanisms underlying genetic contributions to phenotypic variations. Hence, we consider a mixture model to locate candidate genomic regions that are more frequently associated with gene expression traits. A modified two-sample t-statistic was used, and single-nucleotide polymorphisms (SNPs) with P-values <10-5 were considered for a subsequent two-component negative binomial mixture model. An expectation-maximisation algorithm was adopted to identify the parameters involved in the model. The SNPs were then ranked based on their false discovery rate (FDR) values. Any SNP with a FDR value <1% was considered as a potential hotspot. Three independent datasets were used to replicate the findings. A number of common hotspots were identified, and many hotspots have annotated function as the binding site of transcription factors or histone proteins.