Consistency of tools that predict the impact of SNPs on gene functionality. Application to BRCA1 gene

GARCIA LABARI IGNACIO; GUSTAVO VAZQUEZ; MURILLO JAVIER; BULACIO, PILAR; TAPIA ELIZABETH; SPETALE FLAVIO EZEQUIEL; OLIVE CAILLOUX

Montevideo

Conferencia; X International Conference on Bioinformatics 10th Anniversary of SoIBio; 2019

Many Bioinformatics tools have been developed to predict the effect of single nucleotide polymorphisms (SNPs) on gene functionality in an effort to reduce the need for in-vivo assays. Their use in scientific research and even in personalized medicine is frequent and the need of understanding their output essential. However, the large number of tools available and the heterogeneity of their output make their selection, understanding and comparison a non trivial task. Most of the works in the literature that compare the tools simplify the problem through a conversion of their output to a binary scale, which reduces the information they provide. In this work two issues are studied: i) Dothe prediction tools provide similar results (inner consistency)?, and ii) Are the results they provide consistent with the ones published in the literature (outer consistency)?In order to answer the questions, the consistency of six tools werestudied. Their selection was based on the diversity the learning method they are based, the possibility to be run on-line and their popularity in bibliographic reports. Two indices were proposed to evaluate the inner and outer consistency of the tools. Moreover conversion of their output to a binary scale was considered to compare the outer consistency against the proposed indices. Proposed indices quantify the systematic disagreement between each pair of SNPs, i.e., count pairs of predictions ordered differently in each tool scale, without performing any scalenormalization. The tools were tested with 2730 SNPs from breast cancer type one susceptibility protein encoded by the BRCA1 gene.The comparison on BRCA1 shows that the predictionsprovided by the tools vary widely depending on the tool. The learning method diversity is clearly characterized by the indices. The consistency of the tools with bibliographic results varies from 6% to 92% in the worst case if the binary output conversion approach is followed. The inner consistency of the tools is reflected in the analysis of their output with the results of the bibliography. Different outputs are not necessarily a problem, since they enable outputs to be integrated in order to achieve a more accurate prediction of SNP effects.