Can We Use Standard Tools to Predict Functional Effects of Missense Gene Variations Outside Conserved Domains? TET2 Example

Abstract

The most common genetic variations in humans are Single Nucleotide Polymorphisms (SNPs), so predicting their associations with cancers is a significant issue. Here, we were particularly interested in SNPs occurring outside protein Conserved Domains (CDs) of TET2, a recently discovered epigenetic regulator involved in leukemogenesis. Functional effects of TET2 gene variations were assessed with four publicly available tools: PhD-SNP, MutPred, PolyPhen-2 and SIFT. The methods were tested on the dataset of 166 SNPs and somatic TET2 mutations, and separately on the subset of 69 variations outside TET2 CDs. Abilities of tested tools to separate neutral SNPs from pathogenic mutations were similar to previously reported on complete TET2 dataset. However, we observed significantly lower accuracy of predictions outside CDs, ranging from 0.54 to 0.62. Also, areas under the ROC curves were low, 0.51-0.55. Correlations between predictions and positions of variations inside/outside CDs were significant and high, 0.46-0.78. Low efficiency of commonly used tools in predicting functional effects of variations outside CDs emphasize the need for new or modified algorithms.