Towards a computational approach for the identification of functional regulatory genetic polymorphism.

摘要

Distinguishing functional, non-coding, regulatory single-nucleotide polymorphisms (SNPs) from among the millions of known non-functional variants is challenging. Efforts to develop functional sequence annotation such as the ENCODE project may enable development of more accurate prediction algorithms, yet it remains unclear which forms of sequence annotation will be most useful for identifying functional variants. Using results from three genome-wide expression trait mapping studies and 12 genome sequence annotation tracks, we assessed whether or not functional DNA elements are enriched for cis-acting SNPs. We found significant enrichment for most functional tracks tested, with greatest enrichment in CpG islands, RNA Pol II sites, DNase hypersensitivity sites, histone modification sites, and microRNA target sites. In contrast, enrichment was inconsistent when relying only on sequence conservation or transcription-factor binding-site conservation prediction tracks. These data suggest that functional sequence annotation may facilitate more efficient functional SNP selection for genetic association studies. We took advantage of these findings and built a Random Forests (RF) model to predict functional regulatory SNPs (rSNPs) with 38 SNP attributes including 22 histone profiling tracks, distances to the transcription start and stop sites of the closest gene, genic region relative to the closest gene, and sequence features such as minor allele frequency (MAF) and transition/transversion mutation types. The RF model far outperforms individual attributes in identifying cis-acting SNPs in terms of sensitivity and specificity, and the significant SNPs predicted to be functional are about 4 times as likely to replicate across independent studies as those significant SNPs predicted to be non-functional. The data suggest that identification of functional variation using sequence data is feasible and can be a useful means of identifying candidate variants for disease association studies.