Related Experiment Video
Updated: Jun 9, 2025
07:56
Scanning Skeletal Remains for Bone Mineral Density in Forensic Contexts
Published on: January 29, 2018
17.5K
Enhancing late postmortem interval prediction: a pilot study integrating proteomics and machine learning to
Camila Garcés-Parra1,2, Pablo Saldivia3, Mauricio Hernández3
1Gene Expression and Regulation Laboratory (GEaRLab), Department of Biochemistry and Molecular Biology, Faculty of Biological Sciences, University of Concepción, Concepción, Chile.
Biological Research
|October 24, 2024
Summary
Forensic scientists can now estimate late postmortem intervals (PMI) with 100% accuracy using three protein biomarkers identified through proteomics and machine learning. This breakthrough enhances PMI prediction for skeletal remains, especially beyond 15 years.
Area of Science:
- Forensic Science
- Biochemistry
- Computational Biology
Background:
- Accurate postmortem interval (PMI) determination is challenging, particularly for intervals exceeding 5 years.
- Traditional methods struggle with degraded soft tissues, necessitating bone analysis.
- Current PMI estimation accuracy decreases significantly over time, especially between 1-5 years.
Purpose of the Study:
- To identify novel protein biomarkers for accurate late PMI estimation using proteomics.
- To leverage machine learning to enhance PMI prediction accuracy for intervals >15 years.
- To address limitations in current forensic science methods for determining time since death.
Main Methods:
- Proteomic analysis (LC-MS/MS) was performed on tibia and rib skeletal remains.
- Protein identification utilized both tryptic and semitryptic searches, with semitryptic proving advantageous for degraded samples.
- The Random Forest algorithm modeled protein abundance for PMI prediction, with biomarker selection via importance scores and SHAP values.
Main Results:
- A core set of three biomarkers (K1C13, PGS1, CO3A1) significantly improved prediction accuracy for PMIs between 15-20 years.
- A machine learning model using semitryptic peptides from tibia samples achieved 100% accuracy over 100 iterations.
- Semitryptic peptides demonstrated superior performance over tryptic peptides in tibia proteomes for late PMI prediction.
Conclusions:
- Combining proteomics and machine learning offers a feasible approach for accurate late PMI prediction.
- The identified biomarkers show promise for refining forensic proteomic methodologies.
- Future research should expand sample size and PMI ranges to further validate and standardize these methods.
