HashSeq: a Simple, Scalable, and Conservative De Novo Variant Caller for 16S rRNA Gene Data Sets

16S rRNA gene sequencing is a common and cost-effective technique for characterization of microbial communities. Recent bioinformatics methods enable high-resolution detection of sequence variants of only one nucleotide difference. In this study, we utilized a very fast HashMap-based approach to detect sequence variants in six publicly available 16S rRNA gene data sets. We then use the normal distribution combined with locally estimated scatterplot smoothing (LOESS) regression to estimate background error rates as a function of sequencing depth for individual clusters of sequences. This method is computationally efficient and produces inference that yields sets of variants that are conservative and well supported by reference databases. We argue that this approach to inference is fast, simple, and scalable to large data sets and provides a high-resolution set of sequence variants which are less likely to be the result of sequencing error. IMPORTANCE Recent bioinformatics development has enabled the detection of sequence variants with a high resolution of only one single-nucleotide difference in 16S rRNA gene sequence data. Despite this progress, there are several limitations that can be associated with variant calling pipelines, such as producing a large number of low-abundance sequence variants which need to be filtered out with arbitrary thresholds in downstream analyses or having a slow runtime. In this report, we introduce a fast and scalable algorithm which infers sequence variants based on the estimation of a normally distributed background error as a function of sequencing depth. Our pipeline has attractive performance characteristics, can be used independently or in parallel with other variant callers, and provides explicit P values for each variant evaluating the hypothesis that a variant is caused by sequencing error.