Manner of death prediction: A machine learning approach to classify suicide and non-suicide using blood metabolomics

Abstract

The classification of the manner of death (MOD) is a critical step in forensic investigations. The process is based on scene investigation, autopsy, histological and toxicological findings. However, in complex suicide cases, these findings may be insufficient to clearly establish the MOD and need potential biomarkers to assist judicial determinations. This study aims to identify specific biomarkers in the blood that could distinguish suicide from the non-suicidal deaths group. Heart blood samples were collected from suicide (n = 45) and non-suicide cases (n = 45) and metabolomic profiles were analyzed using proton nuclear magnetic resonance spectroscopy. Nineteen blood metabolites were significantly different between the groups (p < 0.05); especially, 4-hydroxyproline, sarcosine and heparan sulfate emerged as potential biomarkers for differentiating between the groups. A logistic regression-based predictive model incorporating sarcosine and heparan sulfate achieved sensitivity and specificity values of 73 % and 72 %, respectively. The integration of machine learning with blood metabolomics holds significant potential in forensic science and may apply to the model to adopt in criminal justice.