Advancement in statistical modelling of chemical impurity profiles in forensic science

At crime scenes, investigators may encounter harmful and illegal chemical substances. Forensic technicians can usually conclude the identity of the chemical substance using proper analytical techniques. However, to tie the chemical substance to a manufacturer is more challenging. Yet, substances usually contain clues of its origin in terms of impurities present in the sample. Impurity profiles of chemicals has been extensively studied and the multivariate modelling method OPLS-DA have been relied on to differentiate classes of samples based on their impurity profile. Despite the good performance of OPLS-DA, its performance becomes worse with increasing number of classes, which is often the case in forensic science when there exist several possible manufactures and synthesis routes available for a specific chemical substance of interest. Recently, the multivariate classification method OPLS-HDA showed to possess the differentiating and predicting power of OPLS-DA and at the same time circumvents the problem with multiple classes. In this study, we applied OPLS-HDA to predict synthesis routes of fentanyl analogues based on impurity profiles. We analyzed the impurity profiles of fentanyl analogues using NMR and IR. OPLS-HDA showed improved performance and a higher accuracy compared to the traditional OPLS-DA, in predicting the synthesis route used for fentanyl analogue production. These findings show the potential to apply OPLS-HDA in forensic science and in criminal investigations, to aid in solving crimes.