<b>Inappropriate application of mapping algorithms results in length-dependent gene abundances in metagenomic analysis</b>

Multiple biases still exist in metagenomic analysis workflows compromising a precise quantification of microbial species and function. In this study, we demonstrated that inappropriate reads mapping could result in length-dependent gene abundances and in turn influence downstream analyses. Specifically, mapping reads directly to predicted genes using Bowtie 2 with the default global-alignment mode resulted in a gradual decrease of abundance values with the diminution of gene length. A local-alignment mode could reduce a large part, but not all, of such technical variability. Particularly, the alignment-free Salmon, which was originally designed and tested for the mapping of RNA-seq reads, were not suitable for metagenomes due to a rapid increase of the TPM (transcripts per million) measure with the diminution of gene length. Different mapping methodologies identified different sets of KEGG Orthologs with significant differences in abundance. For short-read sequencing datasets, we proposed the 'contig mapping' method, i.e., calculating gene abundances using coverage depths of contigs and gene coordinates, which minimized the technical variability.