From microbial data to forensic insights: systematic review of machine learning models for PMI estimation

Area of Science:

• Forensic Science
• Microbiology
• Bioinformatics

Background:

• Traditional post-mortem interval (PMI) estimation methods face limitations due to environmental variability and human error.
• Emerging molecular and microbial techniques offer enhanced accuracy for PMI determination.
• Machine learning (ML) integration with microbial data shows potential for improving PMI estimation reliability.

Purpose of the Study:

• To systematically review and compare microbiome-based PMI prediction methods.
• To analyze the performance of various machine learning techniques across different organs and environments.
• To identify the most effective ML models and microbial data types for PMI estimation.

Main Methods:

• Comprehensive literature search across major scientific databases (PubMed, Scopus, Web of Science, IEEE, Cochrane Library) up to September 2024.
• Systematic data extraction by two independent reviewers, focusing on study details, sample types, PMI ranges, ML algorithms, and performance metrics.
• Ranking and analysis of ML models based on error metrics (e.g., Mean Absolute Error) and explained variance.

Main Results:

• Random Forests (RF) models demonstrated high accuracy in PMI estimation, with reported Mean Absolute Errors (MAE) as low as 6.93 hours (Wang, 2024).
• Studies utilizing soil samples and 16S rRNA data with RF models achieved MAEs around 1.5 days (Yang, 2023; Belk, 2018).
• Neural networks also showed effectiveness, with one study reporting an MAE of 14.483 hours (Liu, 2020).

Conclusions:

• Machine learning, particularly RF models combined with 16S rRNA and soil microbial data, shows significant promise for accurate PMI estimation.
• Further research is needed to standardize parameters and validate these models across diverse forensic contexts.
• Microbiome-based approaches represent a significant advancement over traditional PMI estimation techniques.