Predicting functional effect of human missense mutations

Our aim is to prioritize human missense mutations by their probability of being disease causing. Such a computational method could be used to obtain a reduced set of mutations with a relatively large fraction of disease related mutations, thereby aiding in the search for this type of mutation within a large mutation set.

Whereas a range of methods is available for this purpose, only few employ the availability of the 1000G data to obtain a set of neutral mutations. The novelty of our approach is the use of separate classifiers that were trained on a subset of mutations from one amino acid to any other amino acid. The combined performance of these classifiers show an improved performance compared to the often used prediction method PolyPhen2.