A novel algorithm for prediction of drug resistance using cross-accuracy computation

Abstract

Genome-wide association studies (GWAS) are vital in determining disease etiology. Since most of the existing computational models for GWAS require many features, they are often too computationally expensive for clinical use. An algorithm for prediction of Mycobacterium tuberculosis drug resistance is proposed by using cross-accuracy computations. We do exhaustive investigation of SNP pairs followed by a greedy algorithm to improve prediction accuracy of SNP combinations. A dataset of 3178 Mtb genomes was tested to predict resistance to 8 drugs. The results were compared with those from TB-profiler and Mykrobe web systems. They indicate that our algorithm closely approximates the performance of the existing methods for first-line drugs, while is outperforming them in accuracy for the second-line drugs tested