Core Microbial Species Screening in Acetate Anaerobic Fermentation Based on Granger Causality and RF-RFE
Core microbial species screening in acetate anaerobic fermentation is essential for understanding complex microbial communities and supporting process regulation. However, existing machine learning-based screening methods still mainly rely on correlations, feature importance, or predictive contribution, making it difficult to accurately identify microbial species causally associated with acetate production. To address this issue, this study proposes a two-step framework based on Granger causality and random forest-recursive feature elimination (RF-RFE). First, candidate ASVs are pre-screened according to their causal effect strengths and cumulative causal contribution rate. Then, RF-RFE is applied to determine the optimal ASV subset for core microbial species analysis. Experimental results based on mixed sludge-syngas anaerobic fermentation data show that the proposed framework reduced 1850 ASVs to 115 candidate ASVs and further identified an optimal subset of 9 ASVs. The regression fitting based on the optimal ASV subset achieved an R2 of 0.7953 and an RMSE of 0.0399. Taxonomic analysis indicated that the screened ASVs corresponded to microbial taxa associated with gaseous substrate conversion, endogenous organic matter fermentation, and acetate formation, suggesting that the proposed framework can support core microbial species screening in acetate anaerobic fermentation.