Untangling the Postmortem Metabolome: A Machine Learning Approach for Accurate PMI Estimation

Accurate estimation of the postmortem interval (PMI) is crucial for medico-legal investigations, providing critical timelines for criminal cases. Current PMI methods, however, often lack precision, limiting their forensic utility. In this study, we developed models to estimate PMI with high accuracy across various tissues within the first 4 days after death. Using untargeted UHPLC-qTOF-MS, we analyzed thousands of molecules in rat tissues with different PMIs. We employed machine learning on stable and highly reproducible molecules in each tissue to select candidate biomarkers and then built a second model using only the top 15 molecules. Both Lasso and Random Forest approaches yielded high cross-validation accuracy across all tissues, with the latter showing slightly superior performance. Validation was conducted using an independently collected and analyzed set of rats. The identified metabolites, including amino acids, derivatives, nucleosides, and other markers, are common to humans and mammals, underscoring their potential applicability in human forensic contexts. Our findings highlight the tissue-specific predictive potential and variability in predictive accuracy across different tissues in a rodent model.