Machine Learning Model Estimates Time Since Death Using Routine Toxicology Data

Researchers have developed a machine learning model that estimates the post-mortem interval (PMI) using chemical information from routine forensic toxicology tests. The approach could help investigators estimate how long a person has been dead when conventional methods become less reliable.

Published in Nature Communications on 11 February 2026, the study analysed femoral blood samples from 4,876 forensic cases with known post-mortem intervals ranging from 1 to 67 days. The researchers used metabolomic data—measurements of small molecules in biological samples—originally generated during toxicological screening.

A neural network model achieved a mean absolute error of 1.45 days and a median absolute error of 1.03 days on unseen test cases. The team also tested the model on an independent dataset of 512 cases collected in a different year and analysed using a separate mass-spectrometry platform. Its mean absolute error on this dataset was 1.78 days.

The model identified chemical patterns associated with post-mortem changes, including lipid degradation, mitochondrial dysfunction and protein breakdown. These changes may provide measurable indicators of the time elapsed after death.

Traditional approaches, including body-temperature measurements and vitreous humour potassium analysis, have limitations as the post-mortem interval increases. By repurposing data already produced during toxicology examinations, the proposed method could add information without requiring an entirely separate analytical workflow.

The findings are promising, but the model is not a universal clock for determining time of death. Environmental conditions, individual biological variation, sample handling and differences between laboratories may affect its accuracy. Further validation across forensic settings is needed before it can be relied upon in routine casework.