Robust Aitchison distance calculation.

We applied DEICODE (https://github.com/biocore/DEICODE) to calculate the robust Aitchison distance (rAD) to benchmark the performance of metagenomics profilers. DEICODE represents a form of Aitchison Distance that is robust to high levels of sparsity. It preprocesses the compositional data using the centered log-ratio (CLR) transform only on the non-zero values of the data (hence no pseudo counts are used). Then it performs dimensionality reduction through robust PCA based on the non-zero values of the data. The Euclidean distance of the robust CLR-transformed abundance profiles (i.e., rAD) was finally employed to evaluate the performance of metagenomic profilers. To avoid the impact of false positives on the benchmarking results, we further filtered out false positives in all output taxonomic profiles and compared the performance of different profilers using rAD calculated from the true positives only. This is termed as the modified rAD in Fig.3. For other evaluation measures, the same procedure was performed and presented in Fig.S1.