The identification of optimal drug candidates is very important in drug discovery. Researchers in biology and computational sciences have sought to use machine learning (ML) to efficiently predict drug–target interactions (DTIs). In recent years, according to the emerging usefulness of pretrained models in natural language process (NLPs), pretrained models are being developed for chemical compounds and target proteins. This study sought to improve DTI predictive models using a Bidirectional Encoder Representations from the Transformers (BERT)-pretrained model, ChemBERTa, for chemical compounds. Pretraining features the use of a simplified molecular-input line-entry system (SMILES). We also employ the pretrained ProBERT for target proteins (pretraining employed the amino acid sequences). The BIOSNAP, DAVIS, and BindingDB databases (DBs) were used (alone or together) for learning. The final model, taught by both ChemBERTa and ProtBert and the integrated DBs, afforded the best DTI predictive performance to date based on the receiver operating characteristic area under the curve (AUC) and precision-recall-AUC values compared with previous models. The performance of the final model was verified using a specific case study on 13 pairs of subtrates and the metabolic enzyme cytochrome P450 (CYP). The final model afforded excellent DTI prediction. As the real-world interactions between drugs and target proteins are expected to exhibit specific patterns, pretraining with ChemBERTa and ProtBert could teach such patterns. Learning the patterns of such interactions would enhance DTI accuracy if learning employs large, well-balanced datasets that cover all relationships between drugs and target proteins.

Machine learning techniques are extensively employed in drug discovery, with a significant focus on developing QSAR models that interpret the structural information of potential drugs. In this study, the pre-trained natural language processing (NLP) model, ChemBERTa, was utilized in the drug discovery process. We proposed and evaluated four core model architectures as follows: deep neural network (DNN), encoder, concatenation (concat), and pipe. The DNN model processes physicochemical properties as input, while the encoder model leverages the simplified molecular input line entry system (SMILES) along with NLP techniques. The latter two models, concat and pipe, incorporate both SMILES and physicochemical properties, operating in parallel and with sequential manners, respectively. We collected 5238 entries from DrugBank, including their physicochemical properties and absorption, distribution, metabolism, excretion, and toxicity (ADMET) features. The models’ performance was assessed by the area under the receiver operating characteristic curve (AUROC), with the DNN, encoder, concat, and pipe models achieved 62.4%, 76.0%, 74.9%, and 68.2%, respectively. In a separate test with 84 experimental microsomal stability datasets, the AUROC scores for external data were 78% for DNN, 44% for the encoder, and 50% for concat, indicating that the DNN model had superior predictive capabilities for new data. This suggests that models based on structural information may require further optimization or alternative tokenization strategies. The application of natural language processing techniques to pharmaceutical challenges has demonstrated promising results, highlighting the need for more extensive data to enhance model generalization.

Background: Alzheimer’s disease (AD) is a progressive neurodegenerative disease. Traditional models for estimating AD onset cannot capture nonlinear interactions (epistasis) among the numerous genetic variables that contribute to AD risk. Methods: We developed a feedforward neural network (FFN)–Weibull survival model to predict AD onset using large-scale single-nucleotide polymorphism (SNP) data. We integrated an XAI technique, Shapley additive explanations (SHAP), to address the black-box nature of deep learning, interpret model predictions, and quantify the contribution of each genetic factor to AD. Results: The FFN model achieved a mean concordance index of 0.647, demonstrating an approximately 3.6% improvement over the traditional linear baseline (0.625). The FFN-SHAP model validated established findings, identifying APOE E4 as a primary AD risk factor. APOE E2 strongly protected against AD. Metabolic-disorder-related SNPs had conflicting effects, suggesting gene–environment interactions influence AD onset. Conclusions: By effectively bypassing the combinatorial explosion of interaction terms, the predictive power of an FFN combined with XAI provides a robust methodological tool for identifying the genetic basis of complex diseases, even in cohorts with limited sample sizes. Our model generated novel testable hypotheses regarding the intricate roles of gene–gene and gene–environment interactions in AD pathogenesis.

Decision-making in drug development spans heterogeneous stages from molecular design to clinical optimization, yet computer-aided workflows across stages remain fragmented, limiting traceability from evidence-derived clinical questions to simulation scenarios and decision endpoints. We present a model-driven hybrid AI framework for end-to-end decision support that treats PICO (Participants, Intervention, Comparison, Outcome) as a machine-actionable specification to define scenarios and align endpoints across modules. Given a PICO-defined clinical question and a drug SMILES, the framework infers PBPK-ready parameters from structure, generates exposure via mechanistic PBPK simulation, and links exposure outputs to clinic-facing therapeutic drug monitoring (TDM) decision support using an open-source NLME-based Clinical Pharmacokinetic Consultant Service (CPCS). We additionally implement an LLM-based PICO extraction layer to structure clinical abstracts into P/I/C/O elements for systematic evidence ingestion. Results include backbone verification for DTI-guided intrinsic clearance prediction under fold-error bounds and PBPK trajectory plausibility, and external validation of the hybrid AI–PBPK pipeline showing substantial dispersion for several endpoints. An input-controlled comparison between PhysioSim and PK-Sim on N = 9 drugs yields similarly degraded accuracy under identical AI-predicted inputs, suggesting upstream parameter quality as a dominant constraint. For evidence ingestion, prompting-only PICO extraction achieves F1 of 0.687 on EBM-NLP and 0.552 on TB-PICO. For clinical utility, CPCS improves TDM performance over a PKS baseline on phenobarbital and vancomycin, reducing MAPE by approximately 26–60% depending on configuration. Overall, the framework provides a modular, traceable blueprint linking evidence-defined questions to exposure simulation and TDM decision support while preserving mechanistic interpretability. Future work will integrate outcome-aligned time-to-event modeling and federated learning to close the PICO-to-outcome loop and improve parameter inference under multi-site governance.