Integrated renal histopathology, RNA decay, and protein degradation signatures enhance post-mortem interval prediction using machine-learning models in a veterinary forensic rat model.

BACKGROUND AND AIM: Accurate estimation of post-mortem interval (PMI) is essential in veterinary forensics for reconstructing events surrounding death and supporting legal, welfare, and biosecurity investigations. Traditional methods relying on gross and histological changes are often subjective and highly influenced by environmental factors. This study aimed to develop a multimodal molecular approach by integrating renal histopathological autolysis scoring, RNA decay profiles, and protein degradation kinetics, then comparing classical regression with machine-learning models for PMI prediction in a controlled rat model.

MATERIALS AND

METHODS

Kidneys were collected from adult male Wistar rats at 0, 6, 12, 24, and 48 h post-mortem under controlled 25°C conditions. Histopathological autolysis was scored on a 5-point ordinal scale. RNA degradation was assessed by reverse transcription quantitative polymerase chain reaction (RT-qPCR) for heat shock protein 70 (HSP70), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), and 18S rRNA. Protein degradation was evaluated by Western blotting for HSP70, Caspase-3, and LC3. Integrated datasets were used to build and compare multiple linear regression, random forest, and support vector regression models using leave-one-out cross-validation.

RESULTS

Autolysis scores increased progressively from 1.2 ± 0.2 at 0 h to 4.7 ± 0.5 at 48 h. RT-qPCR cycle threshold values increased steadily for all RNA targets. Protein markers exhibited distinct kinetics: HSP70 declined continuously, Caspase-3 peaked at 12 h, and LC3-II plateaued after 24 h. Support vector regression achieved the best performance (mean absolute error of 0.75 h, root mean square error of 1.40 h, R² of 0.99, and 92% of predictions within ±3 h), outperforming random forest and linear regression.

CONCLUSION

Integrating renal histopathology, RNA decay, and protein degradation signatures with machine-learning models, particularly support vector regression, enables accurate PMI prediction under controlled conditions. This multimodal framework offers a promising template for advancing molecular veterinary forensic science, though validation under variable field conditions is required.