Human Splicing Finder: an online bioinformatics tool to predict splicing signals.

Thousands of mutations are identified yearly. Although many directly affect protein expression, an increasing proportion of mutations is now believed to influence mRNA splicing. They mostly affect existing splice sites, but synonymous, non-synonymous or nonsense mutations can also create or disrupt splice sites or auxiliary cis-splicing sequences. To facilitate the analysis of the different mutations, we designed Human Splicing Finder (HSF), a tool to predict the effects of mutations on splicing signals or to identify splicing motifs in any human sequence. It contains all available matrices for auxiliary sequence prediction as well as new ones for binding sites of the 9G8 and Tra2-beta Serine-Arginine proteins and the hnRNP A1 ribonucleoprotein. We also developed new Position Weight Matrices to assess the strength of 5' and 3' splice sites and branch points. We evaluated HSF efficiency using a set of 83 intronic and 35 exonic mutations known to result in splicing defects. We showed that the mutation effect was correctly predicted in almost all cases. HSF could thus represent a valuable resource for research, diagnostic and therapeutic (e.g. therapeutic exon skipping) purposes as well as for global studies, such as the GEN2PHEN European Project or the Human Variome Project.

UMD-predictor, a new prediction tool for nucleotide substitution pathogenicity -- application to four genes: FBN1, FBN2, TGFBR1, and TGFBR2.

Approximately half of gene lesions responsible for human inherited diseases are due to an amino acid substitution, showing that this mutational mechanism plays a large role in diseases. Distinguishing neutral sequence variations from those responsible for the phenotype is of major interest in human genetics. Because in vitro validation of mutations is not always possible in diagnostic settings, indirect arguments must be accumulated to define whether a missense variation is causative. To further differentiate neutral variants from pathogenic nucleotide substitutions, we developed a new tool, UMD-Predictor. This tool provides a combinatorial approach that associates the following data: localization within the protein, conservation, biochemical properties of the mutant and wild-type residues, and the potential impact of the variation on mRNA. To evaluate this new tool, we compared it to the SIFT, PolyPhen, and SNAP software, the BLOSUM62 and Yu's Biochemical Matrices. All tools were evaluated using variations from well-validated datasets extracted from four UMD-LSDB databases (UMD-FBN1, UMD-FBN2, UMD-TGFBR1, and UMD-TGFBR2) that contain all published mutations of the corresponding genes, that is, 1,945 mutations, among which 796 different substitutions corresponding to missense mutations. Our results show that the UMD-Predictor algorithm is the most efficient tool to predict pathogenic mutations in this context with a positive predictive value of 99.4%, a sensitivity of 95.4%, and a specificity of 92.2%. It can thus enhance the interpretation of variations in these genes, and could easily be applied to any other disease gene through the freely available UMD generic software (http://www.umd.be).