Population Informative Markers Selected Using Wright's Fixation Index and Machine Learning Improves Human...

Area of Science:

• Forensic Science
• Microbiology
• Genetics

Background:

• Microbial DNA from human skin can individualize donors of forensic evidence.
• Previous single-locus microbial DNA marker approaches for human identification (HID) lack robustness due to microbiome fluctuations.
• Multimarker strategies and genetic differentiation show promise for more reliable HID.

Purpose of the Study:

• To evaluate the efficacy of a multiplex panel (hidSkinPlex) for skin microbiome profiling.
• To assess the utility of single-nucleotide polymorphisms (SNPs) with high genetic differentiation (F_ST) for human identification.
• To determine the accuracy of a support vector machine (SVM) learning model in predicting donor identity using selected SNPs.

Main Methods:

• Collected triplicate skin swabs from the nondominant hands of 51 individuals (n=153).
• Analyzed samples using the hidSkinPlex panel to identify candidate SNPs.
• Selected SNPs based on F_ST estimates using three criteria (overall, per marker, selected) and trained an SVM model for donor identification.

Main Results:

• The SVM model achieved high accuracy in predicting donor identity using microbiome SNPs.
• Accuracy rates were 92.00% (overall), 94.77% (per marker), and 88.00% (selected) based on SNP selection criteria.
• Estimates of F_ST combined with SVM significantly improved forensic HID through skin microbiome profiling.

Conclusions:

• Skin microbiome profiling using high F_ST SNPs and SVM offers a powerful supplementary method for forensic human identification.
• This approach provides additional genetic information to identify or exclude individuals from biological evidence.
• Enhanced utility of forensic DNA profiling can be achieved, supporting criminal investigations.