Artificial intelligence can solve theoretical math problems, identify a mysterious enzyme system in viruses, potentially render humans extinct by the end of the decade, and, according to new research, estimate the length of time you’ve been dead.
In a study published Monday in Nature Communications, researchers developed an AI system that uses microbes on cadavers to estimate their postmortem intervals (PMI), or time since death. The system, called mHolmes, was trained on public data from 34 human cadavers. While further work is needed, it could one day greatly improve forensic investigations.
“Estimating time since death is very difficult in forensic science,” Kang Ning, co-author of the study and a bioinformatics expert at Huazhong University of Science and Technology, told Gizmodo. “Previous methods using microbiome samples only sample a few time points (usually 3 to 5) and focus on just one body part, causing errors often larger than ±3 days, especially for dismembered or partial remains.”
34 helpful bodies #
As such, Ning and his colleagues built mHolmes and trained it with data from skin microbiome samples collected daily for 21 days from the face and hip of 34 human cadavers. This taught the system how microbes on dead bodies develop over time. The researchers also enabled mHolmes to use training gained from one body part to inform the PMI estimation for a different body part, which is beneficial in the case of dismembered or partial remains.
mHolmes is essentially a “‘weather forecaster for microbes’—it learns how bacteria grow and decline on a dead body over a long time,” Ning explained. With a few days of microbial data, the system can estimate what the microbes were up to during any potential gap days or predict future microbial compositions. mHolmes uses those predictions to estimate the body’s PMI. The system can also work backwards and figure out the microbial development that took place in the first week since death, which is exceptionally valuable when bodies are discovered late.
“mHolmes can predict microbial changes almost day by day, achieving an average error of less than two days when forecasting across different body sites,” Ning explained. Previous approaches usually have an average error of plus or minus three days. mHolmes’s ability to reconstruct past microbial development also remained stable when the team removed more than half of their input data, demonstrating its ability to perform even with missing information.
We know what it’s doing #
Interestingly, mHolmes pinpointed seven groups of bacteria that are each associated with known decomposition stages. “For example, Gammaproteobacteria bloom early, Clostridia dominate during active decay, and Deinococci appear in the dry stage. This means the AI’s predictions are not a ‘black box’ but are grounded in real biology (explainable),” Ning highlighted.
Overall, mHolmes has the capacity to improve forensic investigations, particularly when it comes to situations such as fragmented remains or a very decomposed body. But additional work is necessary before it can completely replace traditional approaches—for now it should be regarded as a powerful assistant. Ning pointed out that the dataset in their study was small, and mHolmes needs to be further trialed in real-world scenarios with various environments and additional body parts. Researchers will also have to create sampling protocols before mHolmes’s results can be used in court.
But maybe in a few years, the name mHolmes will pop up in your favorite true crime podcast.