The Detection of rare variants responsible for human complex diseases has been receiving more and more attentions. However, most existing computational methods for this purpose require the selection of functional variants before statistical analysis. Based on the assumption that nonsynonymous single nucleotide polymor-phisms (nsSNPs) associated with specific diseases should be similar in their properties, we propose a method that utilize conservation scores of nsSNPs and the guilt-by-association principle to prioritize the candi-date nsSNPs for specific diseases. Systematic validation demonstrates that our approach is effective in recovering the relationship between nsSNPs and diseases, with the Manhattan distance measure achieving the most pre-cise prediction results.

Although remarkable success has been achieved by genome-wide association (GWA) studies over the past few years, genetic variants discovered in GWA studies can typically account for only a small fraction of heritability of most common diseases. As such, the identification of multiple rare variants that are associated with complex diseases has been receiving more and more attentions. However, most of the recently developed statistical approaches for detecting association of rare variants with diseases require the selection of functional variants before the successive analysis, making an effective bioinformatics method for filtering out non-relevant rare variants indispensible. In this paper, we focus on a specific type of genetic variants called single amino acid polymorphisms (SAAPs). We propose to prioritize candidate SAAPs for a specific disease according to their association scores that are calculated using a guilt-by-association model with a set of features derived from protein sequences. We validate the proposed approach in a systematic way and demonstrate that the proposed model is powerful in distinguishing disease-associated SAAPs for the specific disease of interest.