Drug toxicity prediction plays a crucial role in the drug research and development process, ensuring clinical drug safety. However, traditional methods are hampered by high cost, low throughput, and uncertainty of cross-species extrapolation, which has become a key bottleneck restricting the efficiency of new drug research and development. The breakthrough development of Artificial Intelligence (AI) technology, especially the application of deep learning and multimodal data fusion strategy, is reshaping the scientific paradigm of drug toxicology assessment. In this review, we focus on the application of AI in the field of drug toxicity prediction and systematically summarize the relevant literature and development status globally in the past years. The application of various toxicity databases in the prediction was elaborated in detail, and the research results and methods for the prediction of different toxicity endpoints were analyzed in depth, including acute toxicity, carcinogenicity, organ-specific toxicity, etc. Furthermore, this paper discusses the application progress of AI technologies (e.g., machine learning and deep learning model) in drug toxicity prediction, analyzes their advantages and challenges, and outlines the future development direction. It aims to provide a comprehensive and in-depth theoretical framework and actionable technical strategies for toxicity prediction in drug development.

This study was designed to determine the significance of DNA methyltransferases (DNMTs) in DNA hypermethylation in esophageal squamous cell carcinoma (ESCC) and to identify DNA methylation markers in serum for the early diagnosis of ESCC. A promoter methylation profile of 12 tumor-related genes was assessed using methylation-specific PCR in ESCC and paired non-tumor tissue samples from 47 patients. Expression levels of DNMTs were examined by real-time reverse transcription-PCR and immunohistochemistry. Using MethyLight, the methylation status of 5 genes was analyzed in serum samples from 45 patients and 15 healthy individuals. A total of 46 (97.9%) of 47 ESCC samples showed methylation in at least one of the examined genes, and methylation was most frequent for RAR-β (46.8%), DAPK (46.8%), p16 (44.7%), and CDH1 (42.6%). Methylation of RASSF1A was significantly correlated with the poorly differentiated tumors and the early pathologic tumor classification (P=0.035 and P=0.046, respectively). Tumoral DNMT3b mRNA up-regulation was significantly correlated with hypermethylation of multiple tumor-related genes (P=0.021). In addition, hypermethylation of cell-free serum DNA was common in ESCC patients, and diagnostic accuracy was increased when methylation of multiple genes (RAR-β, DAPK, CDH1, p16 and RASSF1A) were analyzed in combination (ROC AUC 0.911, 82.2% sensitivity and 100% specificity). The present study suggests that hypermethylation of multiple tumor-related genes may be involved in the pathogenesis of ESCC and mediated by the increase of DNMT3b expression. A cluster of multiple methylated genes in serum DNA has the potential as a novel biomarker for ESCC diagnosis.

A microfluidic multi-organ chip emulates the tissue culture microenvironment, enables interconnection of organ equivalents and overcomes interspecies differences, making this technology a promising and powerful tool for preclinical drug screening. In this study, we established a microfluidic chip-based model that enabled non-contact cocultivation of liver spheroids and renal proximal tubule barriers in a connecting media circuit over 16 days. Meanwhile, a 14-day repeated-dose systemic administration of cyclosporine A (CsA) alone or in combination with rifampicin was performed. Toxicity profiles of the two different doses of CsA on different target organs could be discriminated and that concomitant treatment with rifampicin from day6 onwards decreased the CsA concentration and attenuated the toxicity compared with that after treatment with CsA for 14 consecutive days. The latter is manifested with the changes in cytotoxicity, cell viability and apoptosis, gene expression of metabolic enzymes and transporters, and noninvasive toxicity biomarkers. The on chip coculture of the liver and the proximal tubulus equivalents showed its potential as an effective and translational tool for repeated dose multi-drug toxicity screening in the preclinical stage of drug development.

Drug-induced liver injury (DILI) is one of the major concerns during drug development. Wide acceptance of the 3R principles and the innovation of in-vitro techniques have introduced various novel model options, among which the three-dimensional (3D) cell spheroid cultures have shown a promising prospect in DILI prediction. The present study developed a 3D quadruple cell co-culture liver spheroid model for DILI prediction via self-assembly. Induction by phorbol 12-myristate 13-acetate at the concentration of 15.42ng/mL for 48hours with a following 24-hour rest period was used for THP-1 cell differentiation, resulting in credible macrophagic phenotypes. HepG2 cells, PUMC-HUVEC-T1 cells, THP-1-originated macrophages, and human hepatic stellate cells were selected as the components, which exhibited adaptability in the designated spheroid culture conditions. Following establishment, the characterization demonstrated the competence of the model in long-term stability reflected by the maintenance of morphology, viability, cellular integration, and cell-cell junctions for at least six days, as well as the reliable liver-specific functions including superior albumin and urea secretion, improved drug metabolic enzyme expression and CYP3A4 activity, and the expression of MRP2, BSEP, and P-GP accompanied by the bile acid efflux transport function. In the comparative testing using 22 DILI-positive and 5 DILI-negative compounds among the novel 3D co-culture model, 3D HepG2 spheroids, and 2D HepG2 monolayers, the 3D culture method significantly enhanced the model sensitivity to compound cytotoxicity compared to the 2D form. The novel co-culture liver spheroid model exhibited higher overall predictive power with margin of safety as the classifying tool. In addition, the non-parenchymal cell components could amplify the toxicity of isoniazid in the 3D model, suggesting their potential mediating role in immune-mediated toxicity. The proof-of-concept experiments demonstrated the capability of the model in replicating drug-induced lipid dysregulation, bile acid efflux inhibition, and α-SMA upregulation, which are the key features of liver steatosis and phospholipidosis, cholestasis, and fibrosis, respectively. Overall, the novel 3D quadruple cell co-culture spheroid model is a reliable and readily available option for DILI prediction.