Determining Informative Microbial Single Nucleotide Polymorphisms for Human Identification.

Area of Science:

• Forensic Genetics
• Microbiology
• Bioinformatics

Background:

• The skin microbiome is a rich and stable source of human DNA, suitable for human identification (HID).
• Existing methods for HID can be limited by DNA sample quality and quantity.
• The hidSkinPlex panel has been used for HID, but improvements in marker robustness and information clarity are needed.

Purpose of the Study:

• To develop an improved panel of single nucleotide polymorphisms (SNPs) from the skin microbiome for enhanced human identification.
• To utilize a machine learning framework for selecting informative SNPs and predicting the human host from microbiome DNA.
• To create a smaller, more robust SNP panel (hidSkinPlex+) that is less susceptible to marker drop-out.

Main Methods:

• Selection of SNPs with high fixation index (FST) and abundance from the hidSkinPlex panel.
• Application of a machine learning framework (LASSO) to generate a SNP panel for predicting human hosts from skin microbiome samples.
• Testing the predictive accuracy of the developed SNP panel on independent datasets from hand, manubrium, and foot samples.

Main Results:

• The machine learning framework achieved 96% accuracy (MCC=0.954) in classifying unknown samples in the test dataset.
• The final hidSkinPlex+ panel, comprising 365 SNPs, demonstrated 95% prediction accuracy (MCC=0.949) for the correct human host.
• The hidSkinPlex+ panel is significantly smaller and more robust to marker drop-out compared to previous iterations.

Conclusions:

• The hidSkinPlex+ panel represents a significant advancement in using skin microbiome DNA for human identification.
• The developed machine learning approach effectively identifies informative SNPs for robust and accurate host prediction.
• This improved method offers a more reliable alternative for forensic genetics, especially when traditional DNA sources are compromised.