Journal: Advanced science (Weinheim, Baden-Wurttemberg, Germany)
This publication is a narrative review of how artificial intelligence and machine learning are being applied to advance cancer immunotherapy, with emphasis on current capabilities, limitations, and translational potential.
Key points:
1. Scope of AI in cancer immunotherapy
- Focuses on four major application areas:
- • Patient stratification – using ML models to predict which patients are likely to respond or develop resistance to immunotherapies (e.g., checkpoint inhibitors), typically by integrating clinical, imaging, and multi-omics data.
- • Biomarker discovery – leveraging high-dimensional data (genomics, transcriptomics, proteomics, single-cell, spatial data) to find predictive and prognostic biomarkers of response, toxicity, and resistance.
- • Treatment strategy optimization – applying AI to guide regimen selection, dosing, treatment sequencing, and combination strategies (e.g., immunotherapy plus chemotherapy, targeted therapy, or radiation).
- • Foundation models and large language models – reviewing emerging work using large, pre-trained models to integrate diverse biomedical datasets and knowledge sources for hypothesis generation, decision support, and more robust prediction tools in immuno-oncology.
2. Role of multi-omics and data integration
- • Emphasizes that the greatest gains come from combining heterogeneous data types rather than relying on single biomarkers.
- • Describes how deep learning architectures are being used to integrate multi-omics, pathology images, and clinical data to better capture tumor–immune interactions and microenvironmental features relevant to immunotherapy response.
3. Clinical translation and limitations
- • Notes that, despite promising performance in research settings, many AI models face barriers to routine clinical use:
- • Limited generalizability across cohorts, institutions, and sequencing platforms.
- • Small, biased, or non-representative training datasets.
- • Lack of interpretability and biological plausibility.
- • Regulatory, ethical, and data-sharing challenges.
- • Highlights the gap between methodological innovation and validated clinical decision-support tools actually used in practice.
4. Future directions and recommendations
- • Calls for:
- • Larger, more diverse, and better-annotated datasets, ideally from multi-center collaborations.
- • Prospective validation and clinical trials embedding AI-based tools for patient selection and treatment guidance.
- • Methods that improve interpretability and link model outputs to testable biological mechanisms.
- • Continued development of foundation models and LLM-based systems tailored to immuno-oncology, with rigorous evaluation of their real-world utility.
For a practicing oncologist, the take-home message is that AI/ML is rapidly reshaping how we identify immunotherapy candidates, discover biomarkers, and design treatment strategies, but most tools remain early-stage. The review serves mainly to map the field, highlight promising directions, and outline what is needed to move from research models to clinically reliable decision support in precision immunotherapy.