Antibiotic treatment can lead to a loss of diversity of gut microbiota and may adversely affect gut microbiota composition and host health. Previous studies have indicated that the recovery of gut microbes from antibiotic-induced disruption may be guided by specific microbial species. We expect to predict recovery or non-recovery using these crucial species or other indices after antibiotic treatment only when the gut microbiota changes. This study focused on this prediction problem using a novel ensemble learning framework to identify a set of common and reasonably predictive recovery-associated bacterial species (p-RABs), enabling us to predict the host microbiome recovery status under broad-spectrum antibiotic treatment. Our findings also propose other predictive indicators, suggesting that higher taxonomic and functional diversity may correlate with an increased likelihood of successful recovery. Furthermore, to explore the validity of p-RABs, we performed a metabolic support analysis and identified Akkermansia muciniphila and Bacteroides uniformis as potential key supporting species for reconstruction interventions. Experimental results from a C57BL/6J male mouse model demonstrated the effectiveness of p-RABs in facilitating intestinal microbial reconstitution. Thus, we proved the reliability of the new p-RABs and validated a practical intervention scheme for gut microbiota reconstruction under antibiotic disturbance.
Keywords: Akkermansia muciniphila; antibiotic disturbance; cross-cohort analysis; ensemble learning framework; gut microbiome; predictive recovery-associated bacterial species (p-RABs).