Machine learning-based linking of bacterial genomes to optimal growth pH: a foundation for rational microbial engineering.

Abstract

Background: Bacterial optimal growth pH is pivotal for enzymatic activity, niche adaptation, and synthetic biology applications (e.g., probiotic design, silage fermentation). Traditional experiments are inefficient, resource-intensive, and miss most unculturable taxa, while direct genome-based prediction of this trait remains unavailable-creating a critical genomic-phenotypic gap that hinders microbial engineering.

Results: We developed BactoGenopH ( http://silagedb.com/BactoGenopH/ ), a web platform for predicting the optimal growth pH of bacteria. We curated a high-quality dataset of 3,476 samples, integrating directly measured pH values from the BacDive database and peer-reviewed literature with corresponding representative genomes from GTDB. Genomic features were extracted via Prodigal for gene prediction and HMMER for Pfam-based functional annotation, with high-importance genes retained and encoded as a binary presence/absence matrix. The XGBoost regression model exhibited robust performance: test set MAE = 0.477, RMSE = 0.666, and 88.82% accuracy (1-pH-unit tolerance); the independent validation set yielded MAE = 0.492, RMSE = 0.694, and 89.37% accuracy. SHAP analysis identified key pH-adaptation genes (e.g., Na_Ala_symp, MgtE) with well-documented roles in ion transport and pH homeostasis. The freely accessible platform supports real-time predictions via FASTA sequence input or file upload, complemented by data visualization and curated dataset browsing.

Conclusion: BactoGenopH fills the unmet need for direct, phenotype-grounded bacterial optimal growth pH prediction, bridging genomic-phenotypic gaps with robust performance. This free resource accelerates trait-driven microbial research and supports rational microbial engineering.