| Abstract: | Drug-induced liver injury (DILI) is associated with an imbalance in the homeostasis of bile salts (BAs). However, a clear connection between BAs and different types of DILI remains to be established. In the present study, random forest (RF) machine learning prediction systems were deployed with 17 individual BAs for categorizing DILI. BAs were analyzed via LC-MS/MS in the serum using the model of seven known hepatotoxins (isoniazid, acetaminophen, bendazac, 17α-ethinylestradiol, 1-naphthylisothiocyanate, tetracycline, and ticlopidine), which caused cholestasis, steatosis, and necrosis in rats. The RF model was validated via leave-one-out cross-validation. The importance of each individual BA with respect to prediction ability was determined. The RF model achieved the best prediction performance, producing accuracy values of 0.98, 0.97, and 1.00 for leave-one-out cross-validation, the training set, and the external test set, respectively. The order of descriptor’s importance was obtained, which was TUDCA > GUDCA > TCA > THDCA. The specificity values for necrosis, cholestasis, and steatosis were 0.94, 1.00, and 1.00, respectively. The results indicated the potential value of individual BA level in serum for categorizing DILI. The RF model in the present work was an inexpensive and readily available tool for categorizing DILI. |
