PredictSNP: Robust and Accurate Consensus Classifier for Prediction of Disease-Related Mutations

PredictSNP: robustní a přesný klasifikátor pro predikci mutací asociovaných se vznikem onemocnění

journal article

english

Bendl Jaroslav, Ing. (DIFS FIT BUT)
Štourač Jan (LL)
Šalanda Ondřej, Ing. (FIT BUT)
Pavelka Antonín, RNDr. (LL)
Wieben Eric, Ph.D. (MAYO)
Zendulka Jaroslav, doc. Ing., CSc. (DIFS FIT BUT)
Brezovský Jan, Mgr., Ph.D. (LL)
Damborský Jiří, prof. Mgr., Dr. (LL)

SNP, single nucleotide polymorphism, SNV, single nucleotide variant, pathogenicity prediction, disease-related mutations

Single nucleotide polymorphisms represent very prevalent form of genetic variation. Mutations in coding regions are frequently associated with the development of various diseases. Computational tools for prediction of effect of mutations are becoming very important for the initial analysis of single nucleotide polymorphisms and their consequent prioritization for experimental characterization due to recent massive increase in the number of known mutations. Many computational tools are already widely employed. Unfortunately, their comparison and further improvement is hindered by large overlaps between their training datasets and potential benchmark datasets, which lead to biased and overly optimistic performances. We constructed the independent benchmark dataset from five large datasets by removing all duplicities or inconsistencies, and subtracting all mutations present at any position used in the training of the evaluated tools or in any of the two external testing datasets. The final independent MetaSNP dataset containing of over 40,000 mutations was then employed in the unbiased evaluation of eight well-established prediction tools - i.e. MAPP, nsSNPAnalyzer, PANTHER, PhD-SNP, PolyPhen-1, PolyPhen-2, SIFT and SNAP. Consequently, the six best performing tools were combined into a consensus classifier MetaSNP. In the evaluation on two other independent external testing datasets, MetaSNP outperformed all integrated prediction tools. This comparison shows that MetaSNP represents a robust alternative to prediction by individual tool. Finally, we developed an easy-to-use web interface to allow an access to all eight prediction tools and consensus classifier MetaSNP. Predictions are supplemented by experimental annotations form Protein mutant and UniProt databases. The interface is available at: http://loschmidt.chemi.muni.cz/metasnp

2014

1-11

PLoS Computational Biology, vol. 10, no. 1, ISSN 1553-7358

Public Library of Science

10.1371/journal.pcbi.1003440

000337948500040

2-s2.0-84896698938