AI is no longer just a support tool in drug development. This paper shows how it can help build faster, smarter, and more personalized cancer treatments, while also highlighting the scientific and regulatory hurdles that still stand.
Research: 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 reportOur group investigated how artificial intelligence (AI) is reshaping drug discovery and its application to precision oncology.
background
What if life-saving cancer drugs could be discovered in months instead of decades? New drugs typically take more than a decade to develop and are expensive, making it difficult for patients to get new drugs in time. However, recent advances in AI may help accelerate the identification of promising drug candidates, improve patient stratification, and provide more predictive information about how well some patients will respond to treatment. Data bias, regulatory uncertainty, and insufficient clinical validation limit the use of AI. To realize its full potential, further research is needed to ensure that AI-based treatments can be implemented safely, equitably, and clinically meaningfully.
Transitioning to AI-driven drug development
Drug discovery has traditionally relied on long experimental processes, often characterized by high failure rates and soaring costs. Thus, several challenges in oncology arise from tumor heterogeneity, therapeutic resistance, clonal evolution, and physiologically complex disease biology. Using AI to address these challenges has created new opportunities for researchers through computational modeling, predictive analysis, and automated compound design.
A major milestone in this transition is the development of AI-generated small molecules that have advanced to clinical trials. For example, tumor necrosis factor receptor-associated factor 2 and NCK-interacting kinase (TNIK) inhibitors designed using generated AI demonstrated safety, tolerability, and pharmacodynamic evidence of target engagement in human studies. Importantly, although this study was conducted in idiopathic pulmonary fibrosis rather than cancer, it provides an early translational reference point that is methodologically relevant to oncology.
Examples of treatments using AI
AI is now part of early clinical therapeutic applications. For example, the generated AI-derived compound INS018_055 is currently in Phase II clinical studies focused on the treatment of fibrotic diseases. One of the immuno-oncology drugs (EXS21546) was enhanced using AI to counteract immunosuppression 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 examples demonstrate significant benefits, including accelerating early-stage drug discovery by automating target identification and compound optimization, reducing failure rates by predicting toxicity and off-target effects before clinical testing, and lastly, improving clinical trial success rates by selecting appropriate patients and using biomarkers.
The technological drivers behind AI success
AI in drug discovery has benefited from several technological advances. For example, new techniques for predicting protein structure, including techniques for modeling complex biomolecular interactions, are providing a more detailed understanding of how drugs interact with their targets. This allows scientists to create more effective and precise drugs.
Self-supervised learning (SSL) allows AI to find useful patterns in large unlabeled datasets. This is particularly valuable in oncology, where vast genomic and multi-omics data are available but annotation is often lacking. SSL enables the identification of new drug targets and disease mechanisms.
Federated learning allows institutions to collaborate without sharing sensitive patient data. This helps protect privacy while improving the generalizability of the model. This is important for developing effective treatments for diverse populations.
Reduced experimental burden and ethical concerns
AI has the potential to reduce reliance on traditional laboratory and animal testing. Computational tools can simulate important aspects of how drug candidates behave, predict how drugs are metabolized in the body, and assess toxicity through absorption, distribution, metabolism, excretion, and toxicity (ADMET) modeling.
By testing drugs on virtual patients using “digital twins,” we can simulate how individuals will respond to treatments under development. This could help refine personalized treatment strategies and reduce unnecessary early-stage experimentation, although these approaches still require empirical validation.
Challenges in clinical translation
A key issue in AI-driven drug discovery is generalizability, as AI is trained on specific data that may not perform well in different patients. For AI-generated treatments to gain regulatory approval and clinical approval, decisions must be clear and biologically plausible. Clinicians and regulators need to understand how these systems reach their conclusions.
Another challenge is that biased AI training data can lead to inequitable medical outcomes and widen disparities between patients. Ensuring data diversity and incorporating bias mitigation strategies are important steps toward equitable care.
Fragmented regulations are also a major challenge, as different countries have different standards for validating AI models. More harmonized standards for validation, reproducibility, data interoperability, and monitoring are needed to speed up the approval process.
The future of AI and 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 could also help interpret real-time clinical data, including circulating tumor DNA (ctDNA), to track progression and detect resistance early. Therefore, continuously adjusting a patient’s treatment plan based on updated data may support adaptive treatment strategies.
AI development is also supported by federated validation and adaptive clinical trial design, enabling scalable AI-based approaches by enabling continuous learning from existing data and refining treatment strategies to match actual patient needs.
conclusion
This perspective shows that AI-driven drug discovery has moved from theoretical possibility to early clinical reality, demonstrating evidence of safety, target engagement, and preliminary efficacy. However, these advances represent early feasibility rather than definitive validation. Sustained clinical effectiveness requires a hybrid framework that integrates computational modeling with experimental and clinical validation.
Addressing the trust, equity, and regulatory governance limitations of AI are essential if these approaches are to help accelerate, improve, and personalize cancer treatment, thereby improving future patient care.
