Scholarlink

Background <p>Estimating post-mortem interval (PMI) is crucial for forensic timelines, yet traditional methods are prone to errors from witness testimony and biological markers sensitive to environmental factors. New molecular and microbial techniques, such as DNA degradation patterns and bacterial community analysis, have shown promise by improving PMI estimation accuracy and reliability over traditional methods. Machine learning further enhances PMI estimation by leveraging complex microbial data. This review addresses the gap by systematically analyzing how microbiome-based PMI predictions compare across organs, environments, and machine learning techniques.</p> Methods <p>We retrieved relevant articles up to September 2024 from PubMed, Scopus, Web of Science, IEEE, and Cochrane Library. Data were extracted from eligible studies by two independent reviewers. This included the number and species of subjects, tissue sample used, PMI range in the study, machine learning algorithms, and model performance.</p> Results <p>We gathered 1252 records from five databases after excluding 750 duplicates. After screening titles and abstracts, 43 records were assessed for eligibility, resulting in 28 included articles. Our ranking of machine learning models for PMI estimation identified the top five based on error metrics and explained variance. Wang (2024) achieved a mean absolute error (MAE) of 6.93 h with a random forests (RF) model. Liu (2020) followed with an MAE of 14.483 h using a neural network. Cui (2022) used soil samples for PMI predictions up to 36 days, reaching an MAE of 1.27 days. Yang (2023) employed an RF model using soil samples, achieving an MAE of 1.567 days in summer and an MAE of 2.001 days in winter. Belk (2018) an RF model on spring soil samples with 16S rRNA data, attaining an MAE of 48 accumulated day degrees (ADD) (~ 3–5 days) across a PMI range of 142 days.</p> Conclusion <p>Machine learning models, particularly RF, have demonstrated effectiveness in PMI estimation when combined with 16S rRNA and soil samples. However, improving model performance requires standardized parameters and validation across diverse forensic environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

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

  • 期刊论文

  • 发表日期：

    2025年4月21日

查看全文
  • Abdulkreem Abdullah Al-Juhani,
  • Arwa Mohammad Gaber,
  • Rodan Mahmoud Desoky,
  • Abdulaziz A. Binshalhoub,
  • Mohammed Jamaan Alzahrani,
  • Mofareh Shubban Alraythi,
  • Saleh Showail,
  • Amjad Aoussi Aseeri

摘要

Background

Estimating post-mortem interval (PMI) is crucial for forensic timelines, yet traditional methods are prone to errors from witness testimony and biological markers sensitive to environmental factors. New molecular and microbial techniques, such as DNA degradation patterns and bacterial community analysis, have shown promise by improving PMI estimation accuracy and reliability over traditional methods. Machine learning further enhances PMI estimation by leveraging complex microbial data. This review addresses the gap by systematically analyzing how microbiome-based PMI predictions compare across organs, environments, and machine learning techniques.

Methods

We retrieved relevant articles up to September 2024 from PubMed, Scopus, Web of Science, IEEE, and Cochrane Library. Data were extracted from eligible studies by two independent reviewers. This included the number and species of subjects, tissue sample used, PMI range in the study, machine learning algorithms, and model performance.

Results

We gathered 1252 records from five databases after excluding 750 duplicates. After screening titles and abstracts, 43 records were assessed for eligibility, resulting in 28 included articles. Our ranking of machine learning models for PMI estimation identified the top five based on error metrics and explained variance. Wang (2024) achieved a mean absolute error (MAE) of 6.93 h with a random forests (RF) model. Liu (2020) followed with an MAE of 14.483 h using a neural network. Cui (2022) used soil samples for PMI predictions up to 36 days, reaching an MAE of 1.27 days. Yang (2023) employed an RF model using soil samples, achieving an MAE of 1.567 days in summer and an MAE of 2.001 days in winter. Belk (2018) an RF model on spring soil samples with 16S rRNA data, attaining an MAE of 48 accumulated day degrees (ADD) (~ 3–5 days) across a PMI range of 142 days.

Conclusion

Machine learning models, particularly RF, have demonstrated effectiveness in PMI estimation when combined with 16S rRNA and soil samples. However, improving model performance requires standardized parameters and validation across diverse forensic environments.