As I process the data sequences from our in-house IonTorrent Next Generation Sequencer, I can't stop noticing quite many sequencing errors in my reads. There are a number of gaps, insertions and mismatches in the index region (and primer region as well). I would expect many of these errors in the primer region, but not in the index region. Not so many anyway. It may be interesting to run our libraries in Illumina to compare the error rate in both sequencing technologies.

A recent paper by Golan & Medvedev (2013)1 has made discouraging statements about the error rate of sequences produced by the IonTorrent:

Despite its advantages, Ion Torrent read accuracy remains a significant concern

According to Golan & Medvedev (2013)1, the base-calling process of the IonTorrent is very simple. It involves rounding the measurements of changes in electricity as nucleotides are incorporated during the sequencing cycles. Which is very prone to errors as distinguishing electricity changes becomes difficult due to higher noise levels towards the end of DNA fragments.

They describe a piece of software that can be used to reanalyze the raw data of the sequencing process (from SFF files). They proposed to combine the information on the bases that are incorporated during the sequencing cycles (flows) to better infer the right nucleotide that gets incorporated in the regions close to the 3' end. They report improvements in the base-calling results from 4 to 21%.

Now I have to go back to the IonTorrent server test whether this software improves my reads.


1. Golan, D., & Medvedev, P. (2013). Using state machines to model the Ion Torrent sequencing process and to improve read error rates. Bioinformatics, 29(13), i344–i351. doi:10.1093/bioinformatics/btt212

Tags: Ion-Torrent Next-Generation-Sequencing bioinformatics