Hierarchically tracking aquaculture effluent in waters by microbial fingerprints-driven machine learning ensemble

Aquaculture effluent threatens aquatic ecosystem health, yet source tracking remains challenging due to the lack of specific microbial fingerprints and result priority. Here, we developed a microbial fingerprints-based machine learning approach for hierarchically tracking aquaculture effluent sources. Through high-throughput sequencing of 386 source samples (aquaculture effluent, domestic sewage effluent, cropland and orchard runoff), we screened four microbial taxa (g_ML602J-51, g_Silicimonas, g_Lewinella and f_Balneolaceae) as fingerprints for aquaculture effluent, exhibiting high sensitivity (0.512–0.649) and specificity (0.804–0.974). Artificial Neural Network and Support Vector Machine-radial were optimal models using fingerprint relative abundance and presence data, with accuracies of 0.8335 ± 0.0090 and 0.8221 ± 0.0047, respectively. The models’ ensemble improved accuracy to 0.8706 ± 0.0175, outperforming individual classifiers by 4.44%–5.90% and fingerprint matching by 17.29%. The predicted uncertainty was stratified into five-credibility tier for tracking primary aquaculture effluent sources. Application across three coastal regions of China demonstrated the generalizability of the approach. Effective tracing of aquaculture effluent becomes feasible using microbial fingerprints and a combined machine learning model, with high throughput sequencing of 386 samples enabling hierarchical identification of targeted source.