Synergistic integration of forensic transcriptome and microbiome: A robust multi-marker strategy combined with machine learning for accurate body fluid identification.

In recent years, the development of microbiome and transcriptome analyses has significantly improved the efficiency of forensic body fluid identification. However, challenging or limited biological samples in forensic practice demand highly efficient utilization of biological samples to minimize sample loss. In this study, we developed two independent assays based on a capillary electrophoresis (CE) approach: a 21-mRNA assay and a 10-bacteria system. The co-extracted RNA and DNA were amplified independently to identify five body fluids using mRNA profiling, and to specifically identify saliva (SA) and vaginal secretion (VS) using bacterial markers. Validation experiments of the two detection systems evaluated specificity, sensitivity, and performance on mixtures, aged, and degraded samples. In order to achieve accurate and intelligent identification of body fluid types, four machine learning (ML) models (Random Forest, K-Nearest Neighbors, Support Vector Machine, and Naive Bayes) were constructed and evaluated. Validation experiments demonstrated that both systems exhibited high overall specificity of body fluids, although certain markers showed cross-reactivity in some non-target samples. The two different assays yielded robust profiles from the samples as low as 1 ng of RNA or 0.1 ng of DNA, as well as most low-volume samples down to 1 μL or a 1/16 swab. Furthermore, the 21-mRNA and 10-bacteria systems effectively analyzed most aged or degraded samples, and mixtures. Despite suboptimal profiles from challenging samples (e.g. 1 μL semen, and aged or degraded semen samples), the SVM classifier overall outperformed other ML models, achieving a 100% classification accuracy for both single-source body fluids and pairwise mixtures on independent test sets. Overall, this study combining multi-omics biomarkers with ML models for precise body fluid identification provides strong technical support for practical forensic application.