Improving Machine Learning Methods for Predicting the Postmortem Interval Using Microbiome Data

The time since death, or postmortem interval (PMI), is critical to establish for forensic investigation as it can help investigators identify the victim and validate alibis and witness statements. Most available tools for measuring PMI (temperature, lividity) are no longer useful after approximately 48 hours, so there is a need for tools to estimate PMI over longer periods of time.

Recent research has indicated that microbial communities associated with decomposing cadavers change systematically after death. These changes can be harnessed to create a microbial clock that begins at death and changes predictably over time, which can be used as a tool for PMI estimation. In this study, we will test the accuracy of different machine learning techniques and model parameters for microbially-based PMI predictions. We will utilize microbiome data sets generated from human cadavers placed in three different geographic locations and four different seasons. We will apply several different machine learning techniques, including Random Forest Regression, K-neighbors Regression, and regularized linear regression techniques, to determine the method for estimating PMI from microbiome data. Then, we will test parameters (e.g. microbial taxonomic level) and variables (e.g. temperature) included in the model to generate the most robust model. Finally, we will validate error rates by sequencing DNA from samples collected from cadavers with known PMIs that were not included in the data set used to create the model. We will apply the final model to these data to determine if a model built on approximately 40 human cadavers is accurate when applied to test samples. Through this study, we expect to create a single generalizable model that will predict the PMI of a human cadaver based on the microbiome. This will provide several contributions to the field of forensic science. First, it will provide a complimentary tool that can be used during forensic investigation to estimate PMI and thus provide scientific evidence that can be utilized by prosecutors during trial. Furthermore, knowledge of the most accurate parameters in the model will lead to improved sampling methodologies. For instance, if it is determined that the gravesoil is the most accurate predictor of PMI, soil swabs for microbiome analysis may become part of forensic investigations.

We will publish peer-reviewed scientific articles, lead conference presentations, and develop a workshop for the forensic science community at the American Association of Forensic Sciences conference to disseminate the study results.

"Note: This project contains a research and/or development component, as defined in applicable law," and complies with Part 200 Uniform Requirements - 2 CFR 200.210(a)(14).

ca/ncf