Microbial signatures define the ecosystem functions of the pelagic microbiome in a basin-scale, Southwest Atlantic Ocean.

The pelagic environment represents a mosaic of biogeographical domains shaped by regional oceanographic processes. Here, a coastal-to-open ocean microbiome investigation was conducted from 64 water samples of the Santos Basin (SB), located in the subtropical South Atlantic Ocean. We combined shotgun metagenomics with a hybrid machine learning workflow to investigate the taxonomic diversity, community structure, and ecosystem functions of pelagic microbiomes. The workflow integrated self-organizing maps (unsupervised) for pattern discovery and Random Forest (supervised) for predictive modeling. Unsupervised machine learning revealed a clear spatial and vertical (light-driven) distribution, with indicator taxa reflecting biogeochemical patterns consistent with global surveys. Supervised learning identified phosphate, salinity, and nitrate, influenced by local upwelling and La Plata River plume, as the primary environmental drivers of microbial community structure. In terms of functionality, the SB microbiome displayed depth- and region-specific patterns: photoautotrophs and nitrogen fixers dominated photic waters (with differences between coastal and oceanic stations), whereas chemolithoautotrophs and mixotrophs prevailed in the aphotic zone. Notably, nitrification signatures were more frequent in northern mesopelagic communities, while sulfur-oxidation pathways were enriched toward the south. Genes for CO bio-oxidation and dimethylsulfoniopropionate (DMSP) degradation were present across all depths. Furthermore, potential non-cyanobacterial diazotrophs were detected in the deep waters, underscoring previous underappreciated to nitrogen cycling. Our findings indicated that the Santos Basin hosts a functionally diverse microbiome including putative novel lineages. The taxonomic and functional patterns observed in the SB might provide insights into potential ecological responses to shifts in nutrient dynamics and physical processes. This investigation provides an ecogenomic baseline for understanding the microbial ecosystem services in subtropical oceans and reveals the potential of machine learning to uncover ecological patterns in underexplored marine regions.