Explainable AI could make breast cancer drug predictions safer and clearer

By Vijay Kumar MalesuReviewed by Susha Cheriyedath, M.Sc.May 22 2026

By combining genomics, explainable AI, and drug-repurposing strategies, researchers outline a pathway for finding faster, cheaper, and more targeted treatment options for complex breast cancer subtypes.

Study: AI-genomics synergy for drug repurposing in breast cancer: an interpretability-driven framework. Image Credit: ProStockStudio / Shutterstock

In a recent review published in the journal npj Genomic Medicine, a group of authors examined how artificial intelligence (AI)-driven genomic analysis could support drug repurposing strategies for precision treatment in breast cancer.

Background

Breast cancer remains one of the most common and deadly cancers, with an estimated 2.3 million new cases and 670,000 deaths worldwide in 2022. Developing new medications to treat breast cancers generally requires more than 10 years and billions of dollars, which creates challenges for bringing new therapies to patients quickly. AI and genomic technologies help reveal how genes, drugs, and pathways are associated with breast cancer. These assist researchers in identifying existing drugs that could be utilized in the treatment of the more aggressive breast cancer subtypes and provide patients with quicker and less expensive treatment options. However, further research is needed to ensure these predictions are accurate, clinically interpretable, and applicable across diverse patient populations.

Understanding the genomic complexity of breast cancer

Breast cancer is a complex disease characterized by great variability between different tumors in different patients and also between tumors in the same patient. The primary molecular subtypes of breast cancer are luminal A and B, human epidermal growth factor receptor 2 (HER2)-enriched, and triple-negative breast cancer (TNBC). Advancements in genomics are allowing scientists to study multiple layers of biological information simultaneously, including deoxyribonucleic acid (DNA) mutations, gene expression, protein expression, and epigenetic regulation.

AI in breast cancer research

AI has emerged as a powerful tool for managing the large volumes of genomic and clinical data generated in cancer research. Machine learning and deep learning models can identify patterns in large datasets, allowing researchers to group tumors, predict how tumors will respond to therapies, and identify new targets for developing therapeutic agents.

Another application of machine learning in cancer research is to determine whether genetic variants are pathogenic. Machine learning models developed using large genomic databases are being developed to help classify variants of uncertain significance (VUS) and link pathogenic variants to potential therapeutic options.

Researchers can now use Explainable AI (XAI) methods such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) to better understand which genomic variables most affect AI model predictions. However, these post-hoc tools do not prove causal mechanisms, and their outputs still require biological validation.

Drug repurposing as a faster therapeutic strategy

Drug repurposing involves finding new uses for existing medications. As approved drugs already have established safety profiles, they can move into clinical testing more rapidly than entirely new compounds. This strategy is especially attractive in breast cancer, where treatment resistance remains a major challenge.

Anticancer properties have been found in many frequently prescribed medications used to treat conditions other than cancer. Metformin, a medication prescribed to treat diabetes, activates AMP-activated protein kinase (AMPK) and can inhibit PI3K/AKT/mTOR signaling in breast cancer models. Statins, prescription drugs used to alter lipid levels, inhibit HMG-CoA reductase and modulate the mevalonate pathway, which has been linked to cancer-cell proliferation and metastasis.

AI improves drug repurposing by integrating genomic, pharmacologic, and clinical data. Computational methods can compare tumor gene expression patterns with drug-induced molecular signatures to identify compounds that can reverse cancer-associated pathways.

Graph neural networks and transformer-based AI systems may predict that an existing drug could effectively target a specific genomic abnormality. Importantly, interpretability methods help explain the biological mechanisms behind these predictions, increasing confidence in their potential clinical relevance.

Integrating AI, genomics, and precision oncology

Researchers propose an integrated framework that combines AI, genomics, and drug repurposing into a continuous precision oncology pipeline. In this model, patient-specific molecular data are analyzed using advanced AI systems that identify potential therapeutic targets and rank candidate drugs according to biological plausibility and subtype relevance.

The framework emphasizes mechanistic validation rather than relying solely on statistical associations. In the proposed framework, drug candidates would undergo pathway analysis, molecular docking, and experimental validation before clinical trials.

A key feature of the proposed framework is a feedback loop. Data from experimental and clinical research would be reintegrated into the AI models to enhance future predictions and support adaptive learning. The result is an iterative process in which AI models are refined using new biological and clinical data to improve predictions of cancer biology and patient response to therapy.

Challenges limiting clinical translation

Despite promising advances, several barriers continue to limit clinical implementation. Many genomic datasets lack diversity and are heavily biased toward populations of European ancestry, reducing the reliability of predictions for underrepresented groups. The authors warn that such bias could reduce model performance in African, Asian, Latin American, and other underrepresented populations and worsen inequities in breast cancer care.

Validation is another key challenge for the AI-generated drug predictions. Most of them do not progress from computation to laboratory testing; therefore, experimental testing with cell cultures, patient-derived organoids, and animal models will remain essential for establishing biological activity and pharmacological safety. Some computationally predicted drugs that have been shown to be very "promising," including metformin and statins, have not consistently translated into clear clinical benefit in randomized or clinical studies.

Conclusion

The review concluded that integrating AI with genomic analysis offers a promising strategy for accelerating drug repurposing in breast cancer. Combining multi-omics data with interpretable machine learning methods helps researchers prioritize candidate repurposed drugs and therapeutic hypotheses for validation. The advancements could lead to faster drug development cycles, reduced costs, and more efficient personalized treatments. However, before using these technologies in clinical settings, it is essential to address challenges related to data diversity, reproducibility, experimental validation, and ethical governance.

Download your PDF copy by clicking here.