bioRxiv2018
BackgroundAnalysis of 16S rRNA marker-gene surveys, used to characterize prokaryotic microbial communities, may be performed by numerous bioinformatic pipelines and downstream analysis methods. However, there is limited guidance on how to decide between methods, appropriate data sets and statistics for assessing these methods are needed. We developed a mixture dataset with real data complexity and an expected value for assessing 16S rRNA bioinformatic pipelines and downstream analysis methods. We generate an assessment dataset using a two-sample titration mixture design. The sequencing data were processed using multiple bioinformatic pipelines, i) DADA2 a sequence inference method, ii) Mothur a de novo clustering method, and iii) QIIME with open-reference clustering. The mixture dataset was used to qualitatively and quantitatively assess count tables generated using the pipelines.\n\nResultsThe qualitative assessment was used to evalute features only present in unmixed samples and titrations. The abundance of Mothur and QIIME features specific to unmixed samples and titrations were explained by sampling alone. However, for DADA2 over a third of the unmixed sample and titration specific feature abundance could not be explained by sampling alone. The quantitative assessment evaluated pipeline performance by comparing observed to expected relative and differential abundance values. Overall the observed relative abundance and differential abundance values were consistent with the expected values. Though outlier features were observed across all pipelines.\n\nConclusionsUsing a novel mixture dataset and assessment methods we quantitatively and qualitatively evaluated count tables generated using three bioinformatic pipelines. The dataset and methods developed for this study will serve as a valuable community resource for assessing 16S rRNA marker-gene survey bioinformatic methods.