Systematic exploration of error sources in pyrosequencing flowgram data

Abstract

Motivation: 454 pyrosequencing, by Roche Diagnostics, has emerged as an alternative to Sanger sequencing when it comes to read lengths, performance and cost, but shows higher per-base error rates. Although there are several tools available for noise removal, targeting different application fields, data interpretation would benefit from a better understanding of the different error types.

Results: By exploring 454 raw data, we quantify to what extent different factors account for sequencing errors. In addition to the well-known homopolymer length inaccuracies, we have identified errors likely to originate from other stages of the sequencing process. We use our findings to extend the flowsim pipeline with functionalities to simulate these errors, and thus enable a more realistic simulation of 454 pyrosequencing data with flowsim.

Availability: The flowsim pipeline is freely available under the General Public License from http://biohaskell.org/Applications/FlowSim.

Contact: susanne.balzer@imr.no.

Fig. 1.

Empirical flow values distributions (D.labrax) and derived intervals.

Fig. 2.

Bins for homopolymer lengths 0, 1 and 2, based on different flow value interval sizes from Table 1.

Fig. 3.

Flow value histograms for G.morhua mate-pair reads (forward matches, N=7 016 764). The y-axis is on a log10 scale. The 15 flow cycles correspond to the 42 positions of the linker sequence. The gray areas contain correct base calls. Subpeaks point toward putative PCR errors.

Fig. 4.

Putative PCR and pyrosequencing error rates with respect to flow cycles (for underlying flow value intervals of size 5 and 95%).