Comparison of error correction algorithms for Ion Torrent PGM data: application to hepatitis B virus

Song L1, Huang W1, Kang J1, Huang Y2, Ren H1, Ding K3.

Sci Rep. 2017 Aug 14;7(1):8106. doi: 10.1038/s41598-017-08139-y.

PMID:28808243 PMCID:PMC5556038   DOI:10.1038/s41598-017-08139-y

Abstract

Ion Torrent Personal Genome Machine (PGM) technology is a mid-length read, low-cost and high-speed next-generation sequencing platform with a relatively high insertion and deletion (indel) error rate. A full systematic assessment of the effectiveness of variouserror correction algorithms in PGM viral datasets (e.g., hepatitis B virus (HBV)) has not been performed. We examined 19 quality-trimmed PGM datasets for the HBV reverse transcriptase (RT) region and found a total error rate of 0.48% ± 0.12%. Deletion errorswere clearly present at the ends of homopolymer runs. Tests using both real and simulated data showed that the algorithms differed in their abilities to detect and correct errors and that the error rate and sequencing depth significantly affected the performance. Of thealgorithms tested, Pollux showed a better overall performance but tended to over-correct 'genuine' substitution variants, whereas Fiona proved to be better at distinguishing these variants from sequencing errors. We found that the combined use of Pollux and Fiona gave the best results when error-correcting Ion Torrent PGM viral data.