Decoding temporal miRNA signatures of semen under in vitro exposure for forensic time since deposition estimation using machine learning‐driven modeling

Abstract In forensic cases, the accurate prediction of time since deposition (TsD) for body fluids plays a critical role in evaluating the relevance of biological evidence to criminal cases and reconstructing the timelines of criminal events. While transcriptomics offers avenues for TsD analysis, the environmental sensitivity of mRNA limits its practical utility. In contrast, miRNAs demonstrate superior potential as biomarkers due to their short sequences, high stability, and environmental resistance; however, their forensic application for TsD estimation remains underexplored. This study applied small RNA sequencing to analyze miRNA expression in semen samples from 10 donors across seven TsD intervals (0–48 h). Time‐dependent miRNA expression modules were identified through Mfuzz clustering and weighted gene co‐expression network analysis. We implemented a multi‐stage feature selection pipeline, commencing with least absolute shrinkage and selection operator regression and random forest (RF) that selected 261 candidate miRNAs for model development, followed by recursive feature elimination with ElasticNet to refine the set to 12 miRNAs, and concluding with XGBoost‐based multicollinearity reduction and exhaustive optimization to yield a minimal set of 7 miRNAs. The selected miRNA candidates were subsequently validated using reverse transcription‐quantitative polymerase chain reaction on an independent sample set. Machine learning models constructed with the initial 261 miRNAs demonstrated that RF achieved optimal performance in the binary classification of early (0–12 h) versus late (24–48 h) TsD, with an accuracy of 0.76, F1‐score of 0.75, and area under the curve of 0.82. In regression analysis, an ensemble model integrating partial least squares, ElasticNet, support vector machine, and Ridge attained a test mean absolute error of 6.76 h and an R 2 of 0.72. This research establishes a novel miRNA‐based prediction framework for TsD estimation of semen, integrating dynamic expression patterns with machine learning for the advancement of forensic body fluid analysis.