Abstract
BACKGROUND: With the recent growth of information on sequence variations in the human genome, predictions regarding the functional effects and relevance to disease phenotypes of coding sequence variations are becoming increasingly important. The aims of this study were to catalog protein-coding sequence variations (CVs) occurring in genetic variation databases and to use bioinformatic programs to analyze CVs. In addition, we aim to provide insight into the functionality of the reference databases. METHODOLOGY AND FINDINGS: To catalog CVs on a genome-wide scale with regard to protein function and disease, we investigated three representative databases; the Human Gene Mutation Database (HGMD), the Single Nucleotide Polymorphisms database (dbSNP), and the Haplotype Map (HapMap). Using these three databases, we analyzed CVs at the protein function level with bioinformatic programs. We proposed a combinatorial approach using the Support Vector Machine (SVM) to increase the performance of the prediction programs. By cataloging the coding sequence variations using these databases, we found that 4.36% of CVs from HGMD are concurrently registered in dbSNP (8.11% of CVs from dbSNP are concurrent in HGMD). The pattern of substitutions and functional consequences predicted by three bioinformatic programs was significantly different among concurrent CVs, and CVs occurring solely in HGMD or in dbSNP. The experimental results showed that the proposed SVM combination noticeably outperformed the individual prediction programs. CONCLUSIONS: This is the first study to compare human sequence variations in HGMD, dbSNP and HapMap at the genome-wide level. We found that a significant proportion of CVs in HGMD and dbSNP overlap, and we emphasize the need to use caution when interpreting the phenotypic relevance of these concurrent CVs. Combining bioinformatic programs can be helpful in predicting the functional consequences of CVs because it improved the performance of functional predictions.