Development of drug-induced gastrointestinal injury models based on ANN and SVM algorithms and their applications in the field of natural products†

Abstract

The broad use of natural products and the accompanied incidences of gastrointestinal injury have attracted considerable interest in investigating the responsible toxic ingredients. Computer models are efficient tools to predict toxicity, but research on drug-induced gastrointestinal injury (DIGI) related to the use of natural products remains lacking. In the present study, a total of 1295 compounds were retrieved from SIDER and AdisInsight databases to investigate whether they cause diseases such as colitis, intestinal perforation, intestinal obstruction, irritable bowel syndrome, intestinal bleeding, inflammatory bowel disease, colon cancer, colorectal cancer and duodenal ulcer as datasets. The ANN and SVM algorithms were evaluated to construct a series of classification prediction models, and finally, a computer model was built based on an ANN algorithm to rapidly screen DIGI induced by natural products. A dataset containing 201 toxic components was established, and the ANN model was used to screen 104 potential DIGI ingredients. Finally, based on molecular docking and CCK-8 methods, the intestinal injury effects of veratramine, emodin and euphobiasteroid were verified. The results of the molecular docking showed that these three components could bind well with the intestinal injury targets PIK3CA, SLC9A3, ACTG2 and HSP90AA1. According to NCM-460 cell experiments, the IC50 values of veratramine, emodin and euphobiasteroid were 75.13, 340.9 and 339.6 μmol L⁻¹, respectively. The study findings further proved the accuracy of the ANN model in screening DIGI components caused by natural products.