Probabilistic insertion, deletion and substitution error correction using Markov inference in next generation sequencing reads

Abstract

Error correction of noisy reads obtained from high-throughput DNA sequencers is an important problem since read quality significantly affects downstream analyses such as detection of genetic variation and the complexity and success of sequence assembly. Most of the current error correction algorithms are only capable of recovering substitution errors. In this work, Pindel, an algorithm that simultaneously corrects insertion, deletion and substitution errors in reads from next generation DNA sequencing platforms is presented. Pindel corrects insertion, deletion and substitution errors by modelling the sequencer output as emissions of an appropriately defined Hidden Markov Model (HMM). Reads are corrected to the corresponding maximum likelihood paths using an appropriately modified Viterbi algorithm. When compared with Karect and Fiona, the top two current algorithms capable of correcting insertion, deletion and substitution errors, Pindel exhibits superior accuracy across a range of datasets.