Synthetic raw-reads for BioSIFTR - Biome-specific Shallow-shotgun Inference of Functional Traits through Read-mapping

The use of 16S rRNA metabarcoding for functional prediction is constrained by various biases, including limited taxonomic resolution and functional inference capabilities. Shallow shotgun sequencing offers a cost-effective alternative with high taxonomic resolution; however, its low sequencing depth presents challenges for accurate functional predictions. Our tool, BioSIFTR, addresses this limitation by mapping shallow shotgun sequencing reads against the MGnify Genomes catalogues biome-specific databases and extrapolating functional profiles from pre-calculated data. To optimise BioSIFTR's settings, we utilised two host-associated biomes: human-gut and chicken-gut. This ENA study contains synthetic reads generated from genome datasets, designed to benchmark taxonomic prediction and functional inference in shallow-shotgun synthetic microbial communities with species richness of 50 and 500 (chicken-gut) and 100 and 1000 (human-gut). By applying the optimised BioSIFTR tool, we found that taxonomic and functional profiles derived from shallow shotgun sequencing closely matched those obtained from deep sequencing in real data of three different biomes. Furthermore, we successfully replicated differences in the human gut microbiome between high and low trimethylamine N-oxide producers, using less than 2% of the original deep sequencing data. BioSIFTR provides a robust and efficient method for approximating the functional information of deep-sequenced metagenomes using a fraction of the sequencing data.