To reconstruct a crime scene, forensic experts are challenged with uncovering the meaning of diverse evidence. The correct classification of the body source of human biological traces found at crime scenes provides crucial information for this purpose. Misclassifications have been particularly noted in samples containing epithelial cells, such as skin, saliva, and vaginal secretion traces. Additionally, many forensic investigations, where the presence of blood has been indicated from presumptive testing, would greatly benefit from more detailed knowledge of the particular body site a bloodstain originated. The human microbiome can serve as a useful forensic tool for the classification of the body source of human traces, as distinct microbial communities inhabit different body sites.

For that, we introduced a novel taxonomy-independent deep learning (DL) microbiome approach based on a large set of reference data. The DL is based on two algorithms: one for epithelial (skin, saliva, vaginal secretion) and another for blood sources (venous, menstrual, nasal, finger prick). Reference 16S rRNA gene sequencing data generated de novo and from the Human Microbiome Project (HMP) from over 2,400 samples were used to train 50 DL networks. Validation testing in over 200 test samples achieved high classification accuracies with AUC values over 0.97 for all categories. Additional forensic validation testing in over 80 mock casework samples aged up to 21 years revealed similar classification results. The work presented in this webinar allows for the determination of body site of origin classification of forensically relevant human traces, with high potential for future applications in casework, provided forensic developmental validation is completed successfully.