Gender profiling based on amino acids in fingermark residues: a study on stability†

Abstract

Chemical residues in fingermarks have been proven to assist in suspect tracing and population profiling. However, the composition and levels of these chemicals are derived from complex metabolic systems and are easily influenced by biological activities, which has hindered judicial institutions worldwide from establishing standardized analytical procedures. To develop a rapid, accurate, and straightforward analytical method, this study employed UPLC-QqQ-MS/MS to quantify amino acid levels in fingermark residues, integrating machine learning techniques and intelligent optimization algorithms for gender prediction. We evaluated whether the relative concentrations of amino acids in fingermark residues—normalized to endogenous serine—could reliably serve as indicators for gender determination, while also examining the effects of donors' physical activity levels, living regions, and fingermark aging periods (0–64 days) on gender classification. The results indicate that significant differences in gender were observed. Under various physical activity frequencies, leucine and valine consistently exhibited statistically significant differences, while across different living regions, valine and phenylalanine remained significant. Moreover, a comprehensive Mann–Whitney significance analysis, followed by Bonferroni correction on all measured fingermarks, revealed that the concentrations of Phe, Ile, Leu, Val, Pro, Asn, Glu, His, and Asp differ significantly between genders. Four classification models were developed based on the relative abundances of amino acids in fingermark residues, and their hyperparameters were optimized using the particle swarm optimization algorithm. Ultimately, the PSO-BP model achieved the highest accuracy of 84.49%. In summary, this study introduces a novel approach utilizing the relative concentrations of amino acids in fingermarks for gender determination. The established method is simple, accurate, and does not require derivatization, making it less susceptible to transfer loss, aging time, or individual factors. The developed models exhibit high classification accuracy and robust generalization ability. The conclusions from this study may provide valuable references for the development of sensitive amino acid reagents and also address a gap in the stability discussion of fingermark residue research.