naiveBayesCall: An Efficient Model-Based Base-Calling Algorithm for High-Throughput Sequencing

Immense amounts of raw instrument data (i.e., images of fluorescence) are currently being generated using ultra high-throughput sequencing platforms. An important computational challenge associated with this rapid advancement is to develop efficient algorithms that can extract accurate sequence information from raw data. To address this challenge, we recently introduced a novel model-based base-calling algorithm that is fully parametric and has several advantages over previously proposed methods. Our original algorithm, called BayesCall, significantly reduced the error rate, particularly in the later cycles of a sequencing run, and also produced useful base-specific quality scores with a high discrimination ability. Unfortunately, however, BayesCall is too computationally expensive to be of broad practical use. In this paper, we build on our previous model-based approach to devise an efficient base-calling algorithm that is orders of magnitude faster than BayesCall, while still maintaining a comparably high level of accuracy. Our new algorithm is called naiveBayesCall, and it utilizes approximation and optimization methods to achieve scalability. We describe the performance of naiveBayesCall and demonstrate how improved base-calling accuracy may facilitate de novo assembly when the coverage is low to moderate.