DyeSPY-LINK: The first likelihood-based inference of near-source kinship for dyed hair evidence comparisons

Hair and fiber evidence is among the most frequently encountered and environmentally persistent forms of trace material in forensic casework. Within this category, forensic practitioners consistently recognize that treated hair, such as bleached or dyed, provides substantially greater probative value than untreated hair. At the same time, modern forensic practice is moving toward likelihood ratio (LR)-based interpretation to ensure transparent, quantitative assessments of evidential strength. Despite this shift, no LR-based framework has yet been established for dyed hair. Here we introduce DyeSPY-LINK, a score-based LR (SLR) system for the probabilistic evaluation of near-source kinship between dyed hairs. Specifically, DyeSPY-LINK is designed to determine the strength of evidence for whether two dyed hair samples originate from the same dye. Performance was assessed on 17 nonoxidative and 43 oxidative commercial dyes previously characterized in DyeSPY. Using surface-enhanced Raman (SER) spectra, we construct known-match (KM) and known-non-match (KNM) comparisons between dyed-hair sources and transform cosine similarity scores into SLRs via kernel density estimation (KDE). For nonoxidative dyes, KM and KNM score distributions were well separated (AUC = 0.9990) coupled with a very low log-likelihood ratio cost (CLLR, 0.0325) indicating excellent discrimination and near-perfect calibration. Misleading-evidence rates were below ∼1% across a wide range of thresholds, and each additional shared colorant increased the odds of obtaining strong support for a common source (LR ≥ 100) by approximately 2.2-fold. For oxidative dyes, KM and KNM distributions were also well separated (AUC = 0.9616) with higher, but still appropriate CLLR (0.3298). Misleading evidence rates remained low, and each additional shared colorant increased the odds of strong support for a common source by 2.6-fold. DyeSPY-LINK thus lays the groundwork for a fully probabilistic, empirically validated system for interpreting dyed-hair evidence in forensic casework.