Artificial intelligence (AI) is reshaping industries worldwide and oncology is emerging as one of its most promising frontiers. According to a 2024 survey by ICON plc, professionals involved in oncology drug development have high expectations for AI, already applying it across target identification, drug discovery, and clinical trial design.
While only 16% of oncology researchers currently use AI for biomarker detection and assessment, nearly half (49%) believe this is where machine learning (ML) could have the biggest impact on accelerating oncology drug development. Their optimism highlights a critical need: identifying biomarkers that are both clinically meaningful and broadly applicable to target patient populations.
The Biomarker Bottleneck in Oncology
Biomarkers are measurable indicators of biological processes, disease states, or responses to therapy. In oncology, they are essential at every stage of the disease journey, from early diagnosis and prognosis to therapy selection and monitoring. Many targeted cancer therapies depend on companion diagnostics to identify the right patients.
However, cancer biology is rarely simple. A single genetic mutation seldom provides sufficient information to guide treatment. Tumours exist within complex microenvironments, with multiple molecular and cellular factors interacting dynamically. Understanding these interactions requires analysing vast amounts of data from genetics, proteomics, transcriptomics, and molecular profiling.
Technologies like next-generation sequencing have helped, but they cannot fully capture the intricate patterns driving cancer progression and treatment response. This complexity makes biomarker discovery labor intensive, expensive, and often inconclusive.
Why AI Is a Game-Changer
AI excels at recognising patterns in large, multidimensional datasets, something traditional statistical methods struggle to achieve. By integrating diverse data sources, AI can uncover subtle, non-linear associations and provide a more holistic understanding of tumour biology. This, in turn, could lead to more accurate biomarker discovery and better patient stratification.
Different branches of AI offer different advantages:
- Machine learning (ML) uses algorithms to detect patterns and adapt as new data become available.
- Deep learning (DL), a subset of ML, relies on layered neural networks capable of extracting features from unstructured data with minimal manual input.
Researchers are already demonstrating AI’s power in this space. For instance, a study in Journal of Hepatology applied deep learning to RNA-seq, miRNA-seq, and DNA methylation data to identify molecular signatures linked to survival in hepatocellular carcinoma. These findings may eventually inform therapy choices and prognostic models.
Another study published in Cancer Cell showed that deep learning can detect genetic biomarkers, such as microsatellite instability, directly from routine colorectal cancer histology slides, bypassing traditional, more resource-intensive methods like PCR or immunohistochemistry. This approach could make genetic pre-screening faster and more accessible.
Emerging Frontiers: Immuno-Oncology and Multimodal Integration
Although AI is already showing promise in oncology research, its clinical adoption remains limited. One particularly promising frontier is immuno-oncology, an area where treatment responses vary widely and validated biomarkers are scarce. Immunotherapies benefit only a subset
of patients, and identifying who those patients are remains a major clinical challenge.
Here, AI offers potential solutions. Researchers are increasingly using AI to discover predictive biomarkers for immunotherapy response, drawing on diverse data sources like genomics, radiomics, and pathomics. According to a 2024 systematic review in Annals of Oncology, radiomics, the extraction of quantitative features from medical imaging, is currently the most widely explored modality.
Another exciting direction is multimodal data integration, where AI models combine molecular characterisation, imaging, and clinical history. This comprehensive approach enables AI to identify complex relationships between variables, leading to “meta-biomarkers” that offer richer predictive power than single-modality biomarkers.
Despite its promise, multimodal integration faces practical hurdles, including limited data standardisation, insufficient validation in clinical studies, and the logistical complexity of data collection. Nonetheless, researchers see this as a critical pathway for future biomarker development.
Challenges on the Road to Clinical Adoption
For AI-based biomarkers to move from research labs into clinical trials and practice, several major challenges must be addressed:
Data Quality and Diversity
AI thrives on large, high-quality, and representative datasets. But oncology data often come from small or homogeneous cohorts, which can lead to bias and limit generalisability. Building robust, diverse datasets that reflect real-world patient populations is essential.
Validation and Regulatory Acceptance
To be clinically useful, AI tools must work not only on their training datasets but across independent cohorts and demographic groups. Regulators will likely require evidence from prospective trials demonstrating clinical utility, safety, and reliability.
Explainability and Transparency
Deep learning models often operate as “black boxes,” offering little insight into how they arrive at predictions. Improving explainability is crucial for clinician trust, regulatory approval, and reproducibility.
Infrastructure and Integration
Incorporating AI into clinical workflows requires technical infrastructure, interoperability with hospital systems, and training for clinicians, factors that can slow adoption.
Looking Ahead: A New Era of Precision Oncology
The convergence of AI and biomarker science represents a transformative shift in oncology drug development. AI has the potential to:
- Accelerate biomarker discovery, revealing patterns invisible to traditional methods.
- Enable more precise patient stratification, improving trial design and therapy targeting.
- Reduce costs by minimising the need for complex or invasive testing.
- Enhance treatment outcomes by matching patients with therapies more effectively.
While significant challenges remain, especially around data quality, validation, and regulatory acceptance, progress is rapid. As AI tools mature, they could help usher in an era of personalised oncology, where treatment decisions are guided not just by single biomarkers but by integrated molecular signatures and dynamic disease models.
For now, oncology stakeholders are watching closely. “The promise of AI is clear,” said one drug development executive in the ICON survey. “The question is how quickly we can move from potential to practice.”
If developers, researchers, regulators, and clinicians can overcome current barriers, AI could redefine how we discover, validate, and use biomarkers, bringing precision oncology one step closer to reality.
References
ICON plc. Innovation in oncology: Accelerating R&D in an evolving landscape. 2024.
Calderaro J. et al. Artificial Intelligence for the Prevention and Clinical Management of Hepatocellular Carcinoma. Journal of Hepatology (2022).
Wagner S. J. et al. Transformer-Based Biomarker Prediction from Colorectal Cancer Histology. Cancer Cell (2023).
Prelaj A. et al. Artificial Intelligence for Predictive Biomarker Discovery in Immuno-Oncology. Annals of Oncology (2024).
Ligero M. et al. Artificial Intelligence-Based Biomarkers for Treatment Decisions in Oncology. Trends in Cancer (2025).
