A new review in Advanced Cancer Research explains how artificial intelligence is being used across cancer drug discovery, from finding targets to designing and testing new drug candidates. The authors show where AI is already useful, and what still needs to improve before more AI-designed therapies can reach patients.
Developing cancer drugs is slow, expensive, and uncertain. Tumors differ widely from patient to patient, and many promising drug candidates fail when they move from computer models or lab tests into real biological systems. This has made cancer drug discovery one of the hardest areas of biomedical research.
In a recent review published in Advanced Cancer Research, the authors describe how artificial intelligence (AI) is helping researchers work through this complexity. AI can bring together genomic data, single-cell data, pathology images, protein structures, chemical libraries, and clinical information. Used well, these tools can help scientists spot cancer vulnerabilities, rank possible drug candidates, and design new molecules for difficult targets.
Key highlights include:
Finding better targets: AI can combine many types of cancer data to uncover driver genes, synthetic lethal pairs, immune targets, and tumor-specific weaknesses.
Screening compounds faster: Graph neural networks and structure-aware models can help search large chemical libraries and predict how molecules may bind to cancer-related proteins.
Designing new therapies: Generative AI can propose small molecules, protein and peptide binders, antibodies, nucleic acid drugs, PROTACs, and molecular glues.
Testing candidates earlier: AI models can help predict ADMET properties, drug response, toxicity, and phenotypic effects before costly late-stage experiments.
The review also makes clear that AI is not a shortcut around biology. Many datasets are noisy or incomplete, and some models remain hard to explain. A prediction still has to survive chemistry, biology, and experimental validation.
Looking ahead, the authors point to cleaner shared datasets, more interpretable models, physics-informed AI, organoid-based testing, and automated design-make-test-analyze loops as key steps. The goal is not to replace experiments, but to make each experiment better chosen and more informative.
Citation: Tang J, Xu J, Zhang W, Gong D, Huang Q, et al. Artificial intelligence in oncology drug discovery: from target identification to therapeutic molecule generation. Adv. Cancer Res. 2026(2):0005, https://doi.org/10.55092/acr20260005.
Advanced Cancer Research
Literature review
Not applicable
Artificial intelligence in oncology drug discovery: from target identification to therapeutic molecule generation
11-May-2026