AI is no longer just a support tool in drug development. This paper shows how it could help build faster, smarter, and more personalized cancer therapies, while also exposing the scientific and regulatory hurdles that still stand in the way.
Study: Precision oncology in the age of AI: lessons from AI-driven drug discovery and clinical translation. Image Credit: FOTOGRIN / Shutterstock
In a recent Perspective article published in the journal BJC Reports, a group of authors examined how artificial intelligence (AI) is reshaping drug discovery and its translation into precision oncology.
Background
What if life-saving cancer drugs could be discovered in months instead of decades? Typically, drug development takes over ten years and is expensive, making it difficult for patients to get new medications in time. However, recent advancements in AI could help accelerate the identification of promising drug candidates, improve patient stratification, and provide more predictive information on how well some patients may respond to treatment. Data bias, regulatory uncertainty, and insufficient clinical validation are limiting the use of AI. To maximize its potential, more research is needed to ensure the safe, equitable, and clinically meaningful implementation of AI-driven therapies.
The shift toward AI-driven drug development
Drug discovery has traditionally relied on lengthy experimental processes, often marked by high failure rates and escalating costs. As such, in oncology, several challenges arise from tumor heterogeneity, treatment resistance, clonal evolution, and the physiologically complex disease biology. The use of AI to address these challenges has resulted in new opportunities for researchers through computational modeling, predictive analytics, and automated compound design.
A major milestone in this transition is the development of AI-generated small molecules that have progressed into clinical trials. For example, a tumor necrosis factor receptor-associated factor 2 and NCK-interacting kinase (TNIK) inhibitor designed using generative AI demonstrated safety, tolerability, and pharmacodynamic evidence of target engagement in human studies. Importantly, this trial was conducted in idiopathic pulmonary fibrosis rather than cancer, but it provides an early translational reference point that is methodologically relevant to oncology.
Examples of AI-enabled therapeutics
AI is now part of early clinical therapeutic applications. For example, the generative AI-derived compound INS018_055 is currently in phase II clinical studies focusing on the treatment of fibrotic diseases. One immuno-oncology agent (EXS21546) has been enhanced using AI to counteract immune suppression in the tumor microenvironment. Finally, using a computer algorithm-based analysis, the compound baricitinib was originally developed for the treatment of rheumatoid arthritis but was later repurposed for the treatment of coronavirus disease 2019 (COVID-19).
These cases show major advantages, including accelerating early-stage drug discovery by automating target identification and compound optimization, reducing failure rates by predicting toxicity and off-target effects before laboratory testing, and, lastly, improving clinical trial success by selecting the right patients and using biomarkers.
Technological drivers behind AI success
AI in drug discovery has benefited from several technological advances. For instance, new techniques for predicting protein structures, including those that model complex biomolecular interactions, provide a more detailed understanding of how drugs interact with their targets. This allows scientists to create more effective and precise drugs.
With self-supervised learning (SSL), AI can find useful patterns from large unlabeled datasets. This is particularly valuable in oncology, where vast genomic and multi-omics data are available but often lack annotations. SSL enables the identification of novel drug targets and disease mechanisms.
Federated learning allows institutions to collaborate without sharing sensitive patient data. It can help preserve privacy while improving model generalizability, which is important for developing therapies that are effective across diverse populations.
Reducing experimental burden and ethical concerns
AI has the potential to reduce reliance on traditional laboratory and animal experiments. Computational tools can simulate key aspects of how drug candidates may behave and make predictions about how the body metabolizes medications, evaluate toxicity through absorption, distribution, metabolism, excretion, and toxicity (ADMET) modeling.
The use of “digital twins” to test medications in virtual patients allows for simulating how an individual might respond to a therapy under development. This could help refine personalized treatment strategies and may reduce unnecessary early-stage experiments, though these approaches still require empirical validation.
Challenges in clinical translation
A major issue in AI-driven drug discovery is generalizability, because AI is trained on specific data that may not perform well across different patients. For AI-generated therapies to gain regulatory approval and clinical acceptance, their decisions must be clear and biologically plausible. Clinicians and regulators need to understand how these systems arrive at their conclusions.
Another challenge is that biased AI training data can lead to unfair healthcare outcomes and increase disparities among patients. Ensuring diversity in data and incorporating bias mitigation strategies are essential steps toward equitable care.
Regulatory fragmentation is also a huge challenge, as different countries have varying standards for validating AI models. More harmonized standards for validation, reproducibility, data interoperability, and monitoring are needed to speed up the approval process.
AI and the future of precision oncology
AI is uniquely positioned to advance precision oncology by integrating multi-omics data, including genomic, transcriptomic, proteomic, and metabolomic information. This integration allows for more accurate disease classification and personalized treatment strategies.
AI may also help interpret real-time clinical data, including circulating tumor DNA (ctDNA), for tracking progression and early detection of resistance; therefore, it could support adaptive treatment strategies through continuous adjustment of the patient’s treatment regimen based on updated data.
AI development is also aided by federated validation and adaptive clinical trial design, enabling scalable AI-based approaches by enabling continuous learning from existing data and refining therapeutic strategies to fit actual patient needs.
Conclusion
The Perspective demonstrates that AI-driven drug discovery has moved from theoretical potential to early clinical reality, with evidence of safety, target engagement, and preliminary efficacy. However, these advances represent initial feasibility rather than definitive validation. Sustained clinical impact will require hybrid frameworks that integrate computational modeling with experimental and clinical validation.
Addressing limitations in AI reliability, fairness, and regulatory governance will be essential if these approaches are to help accelerate, improve, and personalize cancer treatments, thereby improving patient care in the future.