The cNED framework: modeling core microbial communities and metabolic functions with SHAP-interpretable environmental...

Area of Science:

• River ecology
• Microbial ecology
• Computational biology

Background:

• River microbial communities are crucial for biogeochemical cycling and pollutant bioremediation.
• Predicting microbial community dynamics under environmental change is complex due to intricate interactions and feedbacks.

Purpose of the Study:

• To develop and validate a novel predictive framework for microbial community and functional responses to environmental changes in river ecosystems.
• To identify key environmental drivers influencing microbial community structure and function.

Main Methods:

• A compositional neural encoder-decoder (cNED) framework was developed, integrating environmental variables with microbial profiles from 473 upper Yangtze River samples.
• Core bacterial operational taxonomic units (OTUs) were identified using the occupancy-frequency method.
• Shapley additive explanations (SHAP) and Generalized Additive Models were employed for driver identification and tipping point analysis.

Main Results:

• The cNED framework significantly outperformed conventional models in taxonomic and functional prediction (e.g., R² = 0.85 for carbon cycling).
• Spatial gradients and temperature were identified as key environmental drivers.
• Specific microbial taxa (e.g., Proteobacteria) and functions (e.g., methylotrophy, nitrogen fixation) exhibited nonlinear responses and tipping points related to temperature and total nitrogen (TN).

Conclusions:

• The cNED framework provides an interpretable and accurate method for forecasting microbial community and functional shifts in response to environmental perturbations.
• Findings offer valuable insights for evidence-based river ecosystem management and climate adaptation strategies.