Machine Learning-Guided Synthetic Microbial Communities Enable Functional and Sustainable Degradation of Persistent...

Persistent environmental pollutants require diverse microbial metabolic capabilities for effective degradation. While naturally occurring consortia or single strains often fall short in efficiency, synthetic microbial communities (SynComs) hold greater promise for enhanced degradation. To address this challenge, we developed GENIA (Genomically and Environmentally Networked Intelligent Assemblies), a genome-informed, machine learning-guided framework for the rational design of SynComs capable of degrading multiple pollutants. Using a microfluidic high-throughput cultivation platform, 2,155 bacterial strains were isolated from xenobiotic-enriched cotton detritusphere and screened for pollutant-specific growth. Whole-genome sequencing and functional annotation of 45 prioritized strains revealed metabolic traits associated with the degradation of lignin, atrazine, and PFAS. These genomic profiles were encoded into spline-based graph representations and integrated within the GENIA pipeline, which combines graph neural networks, pathway complementarity modeling, and functional redundancy minimization to predict optimal community assemblies. The resulting nine-member community, comprising Atlantibacter hermannii, Bacillus cabrialesii, Bacillus licheniformis, Bacillus pseudomycoides, Micrococcus luteus, Paenibacillus polymyxa, Pantoea dispersa, Pseudomonas fulva, and Pseudomonas pergaminensis, demonstrated broad catabolic capacity. Kinetic experiments in minimal medium showed simultaneous multipollutant degradation: lignin (91.6% by day 5), atrazine (91.4% by day 3), and PFOS (93.1% within 7 days), representing 2.2-fold improvement over best individual performers. Full-length 16S rRNA metabarcoding confirmed stable community composition with predicted hub strains expanding to 14-15.6% relative abundance. Soil microcosm validation demonstrated >70% degradation at 3 weeks. GENIA establishes a scalable framework that integrates systems genomics, phenotypic screening, and predictive modeling to engineer microbial consortia for complex environmental bioremediation.