Genomic GC bias correction improves species abundance estimation from metagenomic data

Metagenomic sequencing measures the species composition of microbial communities and has revealed the crucial role of microbiomes in the etiology of a range of diseases such as colorectal cancer. Quantitative comparisons of microbial communities are, however, affected by GC-content-dependent biases. Here, we present GuaCAMOLE, a computational method to detect and remove GC bias from metagenomic sequencing data. The algorithm relies on comparisons between individual species in a single sample to estimate the sequencing efficiency at levels of GC content, and outputs unbiased species abundances. GuaCAMOLE thus works regardless of the specific amount or direction of GC-bias present in the data and does not rely on calibration experiments or multiple samples. Applying our algorithm to 3435 gut microbiomes of colorectal cancer patients from 33 individual studies reveals that the type and severity of GC bias vary considerably between studies. In many studies, we observe a clear bias against GC-poor species in the abundances reported by existing methods. GuaCAMOLE successfully removes this bias and corrects the abundance of clinically relevant GC-poor species such as F. nucleatum (28% GC) by up to a factor of two. GuaCAMOLE thus contributes to a better quantitative understanding of microbial communities by improving the accuracy and comparability of species abundances across experimental setups.