Artificial Intelligence in Oncologic Drug Discovery and Development: A Systematic Review of Transformative Pathways, Current Applications, and Future Prospects

Background: Cancer remains a leading cause of mortality worldwide, with traditional oncologic drug development characterized by lengthy timelines exceeding 15 years, costs surpassing $2.6 billion per approved drug, and success rates below 5%. The complexity of cancer biology, high attrition rates in clinical trials, and urgent need for personalized therapeutic approaches necessitate innovative technologies to accelerate discovery timelines and improve success rates. Artificial Intelligence (AI) has emerged as a transformative technology with potential to revolutionize every aspect of oncologic drug discovery and development, from target identification and virtual screening to clinical trial optimization and drug repurposing. Despite promising individual studies demonstrating AI efficacy across diverse applications, the evidence base remains fragmented across multiple disciplines and publication venues, lacking systematic evaluation of methodological quality, performance outcomes, and clinical translation potential. Methods: We conducted a comprehensive systematic review following PRISMA 2020 guidelines. Five major databases (PubMed, Scopus, Web of Science, Embase, IEEE Xplore) were systematically searched from January 2012 to December 2025 using predefined search strategies combining AI methodologies with oncologic drug discovery concepts. Inclusion criteria encompassed peer-reviewed original studies and reviews focusing on AI applications in cancer drug discovery and development with quantitative outcomes and sufficient methodological detail. Quality assessment employed a standardized six-domain framework evaluating study design, AI methodology, validation approaches, statistical analysis, clinical relevance, and reproducibility potential. Results: From 812 initially identified studies, 11 high-quality studies met inclusion criteria after systematic screening and full-text assessment. Studies demonstrated exceptional methodological quality (mean score: 0.90±0.07, range: 0.75-1.00) with 90.9% achieving high quality ratings. Deep learning emerged as the predominant methodology (45.5%), demonstrating consistently high performance across diverse applications with mean AUROC of 0.882±0.073 (range: 0.780-0.950) and accuracy of 84.0±6.5% (range: 78.3%-91.0%). Applications spanned the entire drug development pipeline from preclinical discovery (36.4%) to clinical validation (18.2%), with pan-cancer approaches most prevalent (45.5%). Global research contributions included USA (36.4%), China (27.3%), and other countries (36.3%), reflecting international recognition of AI's transformative potential.