Researchers have built an AI system that can estimate time since death by reading the microbes on a body. The tool, called mHolmes, appeared in a new Nature Communications study. It could help forensic teams when remains are fragmented or badly decomposed.
Why It Is So Hard to Estimate Time Since Death
Forensic scientists call this timeframe the postmortem interval, or PMI. Getting it right matters in any death investigation, yet the task remains difficult.
Kang Ning, a co-author of the study, works in bioinformatics at Huazhong University of Science and Technology. He explains that earlier microbiome methods collect samples at only three to five time points. They also focus on a single body part. As a result, errors often exceed three days in either direction, especially with dismembered or partial remains.
How mHolmes Learns From Cadaver Microbes
The team trained mHolmes on public data from 34 human cadavers. Researchers had collected skin microbiome samples from the face and hip every day for 21 days. From this data, the system learned how microbial communities change as a body decomposes.
The researchers also gave mHolmes a useful skill. It can apply what it learned from one body part to a different one. Therefore, investigators could still work with partial remains.
A “Weather Forecaster” for Microbes
Ning compares the system to a weather forecaster for microbes. It studies how bacteria rise and fall on a body over time. With just a few days of microbial data, mHolmes can fill in the gaps between samples. It can also forecast how the microbial mix will look later.
The system then uses those predictions to estimate the PMI. It can even work backward to reconstruct the microbial changes of the first week after death. This ability is especially valuable when someone discovers a body late.
How Accurate Is the AI?
According to Ning, mHolmes predicts microbial changes almost day by day. Across different body sites, it reached an average error of less than two days. Previous approaches typically miss by about three days in either direction.
The system also held up under pressure. Its reconstruction of past microbial development stayed stable even after the team removed more than half of the input data. In other words, mHolmes can still perform when information is missing.
Why mHolmes Is Not a Black Box
Many AI tools give answers without showing their reasoning. This one works differently. The system identified seven groups of bacteria, and each links to a known stage of decomposition.
For example, Gammaproteobacteria bloom early. Clostridia dominate during active decay, while Deinococci appear in the dry stage. Because of this, Ning says the predictions rest on real biology rather than hidden patterns. Experts can explain and check the results.
What Must Happen Before Courts Can Use It
Despite the promising results, mHolmes cannot replace traditional methods yet. For now, forensic experts should treat it as a powerful assistant.
Several steps remain. The study used a small dataset, so the team needs to test the system on more body parts. Real-world trials in varied environments are also necessary. Finally, researchers must create sampling protocols before a court can accept mHolmes results.
