Supervised classification of microbiota mitigates mislabeling errors.

The exponential growth of DNA sequencing technologies and concomitant advances in bioinformatics methods are revolutionizing our understanding of diverse microbial communities (Riesenfeld et al., 2004; Tyson et al., 2004; Hugenholtz and Tyson, 2008; Tringe and Hugenholtz, 2008; Caporaso et al., 2010). Large-scale microbial metagenomics studies have particularly exciting applications in the arena of human health, laying the foundation for the Human Microbiome Project (HMP). In the context of the HMP and related efforts, care has been taken to understand the impact of amplification biases or sequencing errors. However, far less attention has been paid to the impact of errors in metadata on biological interpretations and the mitigation of such errors. During processing and pooling of hundreds of samples, some mislabeling is likely. Figure 1 illustrates a real world example: several 16S rRNA amplifications of bacterial community DNA samples collected along a time series were accidentally mislabeled (late switched to early) (Koenig et al., 2010). Automated detection of such errors will be important as datasets become increasingly large and complex.