[P003] Poster

12:10 PM - 1:00 PM JST(3:10 AM - 4:00 AM UTC)

[P02-221] Reverse engineering of microbiome functions: uncovering complex community functions through subcommunity analysis

○Hidehiro Ishizawa^1,2, Miku Kito¹, Sunao Noguchi¹, Kodai Kimura¹, Masahiro Takeo¹ (1. University of Hyogo (Japan), 2. The University of Osaka (Japan))

Keywords:

Microbiome,Interspecies interactions,Machine learning,Bioremediation

[Purpose] Microbiomes play fundamental roles in industrial processes and ecosystem stability. Yet dissecting the mechanistic basis of microbiome functions remains a formidable challenge, as these functions arise from intricate interspecies interactions across taxonomically diverse members. Here, we establish an experimental–computational workflow for subcommunity-based analysis (SubCom analysis) aimed at decomposing microbiome functions into taxon-level contributions and their interaction contexts.[Method] Using an aniline-degrading bacterial community as a model system, we generated paired composition–function datasets from 558 randomly assembled, low-complexity subcommunities built via a dilution-and-dispense approach. Decision-tree–based machine learning models (XGBoost and RuleFit) were subsequently trained to predict community function from taxonomic composition. Learned decision rules were then interpreted to pinpoint taxa and compositional contexts associated with the promotion or suppression of community function.[Results] The trained models demonstrated strong predictive accuracy (r = 0.77–0.89). Interpretation of the learned rules highlighted taxa with reproducible effects: particular Pseudomonas and Acinetobacter taxa were linked to enhanced community-level aniline utilization, while an Achromobacter taxon exhibited a negative association despite its anticipated role in downstream metabolic steps. The models additionally pointed to potential functional interactions, notably a Corynebacterium taxon that attenuated the positive contributions of Pseudomonas and Acinetobacter. A targeted augmentation assay employing representative isolates corroborated the inferred direction of several effects.[Conclusion] Our findings offer a practical implementation of SubCom analysis within a complex, non-synthetic microbiome, and present a systematic route for identifying candidate taxa and interaction contexts that shape community-level function.