Large Language Models in Computational Pharmacology: A Scoping Review of Knowledge Extraction, Mechanistic Reasoning, Drug Repurposing, and Decision Support

Download PDF

Large language models (LLMs) are emerging as generative AI systems with potential relevance to computational pharmacology, including biomedical text mining, drug–target evidence organization, adverse-event extraction, mechanistic hypothesis generation, drug repurposing support, and pharmacology-related decision support. This scoping review maps how LLMs, biomedical language models, domain-specific language models, and retrieval-augmented or tool-augmented systems are being used or proposed across computational pharmacology, while distinguishing evidence organization and hypothesis generation from validated pharmacological or clinical decision-making. PubMed, Scopus, Web of Science, IEEE Xplore, ScienceDirect, SpringerLink, Wiley Online Library, Oxford Academic journals, ACM Digital Library, Nature Portfolio journals, Frontiers journals, MDPI journals, and publisher records were searched for peer-reviewed journal articles published from 1 January 2017 to 9 July 2026; records were screened using predefined eligibility criteria, duplicates were removed, titles and abstracts were screened, full texts were assessed, and 34 studies were included. The mapped literature clustered around biomedical language-model foundations, knowledge extraction, drug–drug interaction extraction, drug–target relation extraction, adverse-event extraction, pharmacovigilance support, mechanistic reasoning assistance, drug repurposing hypothesis generation, medication-information support, and bounded decision-support interfaces. Major gaps involved source grounding, hallucination control, prompt sensitivity, external validation, expert evaluation, reproducibility, privacy, and decision-boundary governance. LLMs may support computational pharmacology by organizing evidence, extracting relations, assisting literature review, and generating hypotheses, but reliable use requires source grounding, task-specific validation, expert oversight, uncertainty communication, and carefully defined boundaries between computational assistance and pharmacological, clinical, or regulatory claims.