Microbial succession and machine learning for postmortem interval estimation in bromadiolone-poisoned cadavers

Accurate estimation of the postmortem interval (PMI) remains a major challenge in forensic investigations. Although microbial succession has emerged as a promising biological clock, the impact of xenobiotic exposure on internal microbial communities during decomposition remain poorly understood. Here, we established a bromadiolone-poisoned rat model and systematically sampled gut and anus swabs at 6 h intervals over 96 h postmortem period. Bromadiolone exposure produced dose-dependent effects on decomposition, with high-doses delaying the progression, potentially by altering tissue integrity or microbial activity. 16 S rRNA sequencing revealed that Firmicutes and Proteobacteria were the dominant phyla during decomposition, followed by Bacteroidota, Desulfobacterota, and Actinobacteriota. Notably, microbial community dynamics were driven more strongly by anatomical location than by toxicant exposure, highlighting the critical role of sampling site. Furthermore, five machine-learning algorithms were evaluated for PMI prediction. GB regression outperformed RF, RR, SV, and LR, achieving an R² of 0.954 and an RMSE of 5.6 h. Feature analysis identified Proteus mirabilis and Morganella morganii as dominant positive biomarkers, while gut-associated taxa (e.g., Lactobacillus and Ligilactobacillus) declined over time. Together, these results demonstrate that bromadiolone exposure induces site-specific perturbations in microbial community, the underlying successional dynamics remain sufficiently robust to support accurate PMI estimation. The GB model provides both predictive accuracy and biological interpretability, highlighting the importance of incorporating anatomical location and toxicological stress into microbiome-based PMI estimation.