Consistent and correctable bias in metagenomic sequencing measurements

Measurements of biological communities by marker-gene and metagenomic sequencing are biased: The measured relative abundances of taxa or their genes are systematically distorted from their true values, as each step in the experimental workflow preferentially detects certain taxa over others. Bias can lead to qualitatively incorrect conclusions and makes measurements from different protocols incomparable. A quantitative understanding of bias is therefore essential. Here we propose, test, and apply a simple mathematical model of how bias distorts marker-gene and metagenomics measurements: Bias multiplies the true relative abundances within each sample by taxon- and protocol-specific factors that describe the different efficiencies with which taxa are detected by the workflow. Critically, these factors are consistent across samples with different compositions, allowing bias to be estimated and corrected. We validate this model in 16S rRNA gene and shotgun metagenomics data from bacterial communities with defined compositions. We also use it to reason about the effects of bias on downstream statistical analyses, finding that analyses based on taxon ratios are less sensitive to bias than analyses based on taxon proportions. Finally, we demonstrate how this model can be used to quantify bias from samples of defined composition, partition bias into steps such as DNA extraction and PCR amplification, and to correct biased measurements. Our model improves upon previous models by providing a better fit to experimental data and by providing a composition-independent approach to analyzing, measuring, and correcting bias.