# Postmortem submersion interval prediction model based on the rat muscle microbiome

## TL;DR

This study explores how the microbiome in rat skeletal muscle changes predictably after submersion in water, offering a new method to estimate post-mortem submersion intervals.

## Contribution

The study introduces a high-precision two-stage machine learning model for PMSI estimation using skeletal muscle microbiome data, independent of cause of death.

## Key findings

- The rat skeletal muscle microbiome shows a predictable successional pattern after submersion.
- A two-stage model achieved 90.9% accuracy in classifying early and late phases of submersion.
- The model's mean absolute error was 0.303 days in the early phase and 1.293 days in the late phase.

## Abstract

Accurate estimation of post-mortem submersion interval (PMSI) is a critical challenge in forensic science. The current research has largely focused on the microbial communities in the skin and gut, which are susceptible to environmental contamination, while the potential of internal tissues remains underexplored. This study aimed to investigate whether the microbiome of skeletal muscle, a relatively closed ecosystem, undergoes a predictable succession following submersion in water PMSI and to evaluate its potential for building a high-precision PMSI prediction model, which is independent of the cause of death (drowning vs. post-mortem submersion).

Using 72 male Sprague–Dawley rats, we established drowning (D group) and post-mortem submersion (PS group) models. After submersion in natural aquatic environment for 14 days, skeletal muscle samples were collected at six time. The microbial communities were profiled by high-throughput sequencing of the V3–V4 region of the 16S rRNA gene, followed by analyses of alpha and beta diversity. Based on the observed successional patterns, a two-stage prediction model combining classification and regression algorithms (e.g., random forest, RF) was developed.

The skeletal muscle microbiome exhibited a significant and predictable successional pattern, clearly partitioning into an early-phase (0–3 days) and a late-phase (5–14 days) (PERMANOVA, p < 0.001). This succession was characterized by a shift from the dominant community of Proteobacteria to the dominant community of Firmicutes. Importantly, the cause of death did not significantly impact either the alpha or beta diversity of the microbial communities (PERMANOVA, p = 0.251). The resulting two-stage prediction model demonstrated excellent performance: the classification model distinguished the early and late phases with an accuracy of 90.9% (AUC = 0.9504), and the mean absolute errors (MAE) of regression models was 0.303 days in the early phase and 1.293 days in the late phase.

The rat skeletal muscle microbiome undergoes a regular and predictable post-mortem succession unrelated to the cause of death. The stable “microbial clock” within the internal tissue allows the construction of a high-precision two-stage machine learning model for PMSI estimation. Our results establish skeletal muscle as a highly promising new target for forensic microbiology, offering a robust theoretical basis and technical approach to resolving challenges in long-term PMSI estimation.