Abstract
The gut microbiome is an essential metabolic organ influencing host health through metabolite production. While metabolite production levels can be directly measured by gas or liquid chromatography, they are commonly inferred from the abundance of metagenomic functional pathways. To evaluate the accuracy of these inferences, we established a single, manually curated metabolite-pathway/enzyme mapping list as a standardized biological reference. We then compared liquid chromatography–mass spectrometry (LC–MS)-based faecal metabolites with functional pathways/enzymes inferred from four approaches: 16S rRNA gene amplicons, reference-based shotgun,
de novo
assembly-based contigs and
de novo
assembly-based metagenome-assembled genomes (MAGs). Our results demonstrate that predictive accuracy is strongly metabolite-specific and method-dependent rather than a uniform characteristic of metagenomic data. While 16S (9.9%), ref-shotgun (14.2%) and
de novo
-contigs (5.5%) yielded only a small fraction of well-predicted metabolites, the
de novo
-MAGs approach significantly outperformed other approaches, achieving a 36.9% well-predicted rate for mapped metabolites. Notably, this approach provided the most robust functional-metabolite associations for indicators of gut health. In conclusion, while microbial functional potential does not always mirror metabolic reality, high-quality genomic binning via MAGs offers a significantly more robust framework for selective metabolite prediction.