Generative AI (GenAI) in cancer research: toward addressing clinical and translational challenges

Dear Editor, Cancer remains a prominent cause of morbidity and mortality globally, despite breakthroughs in diagnostics, treatments, and precision medicine. Despite notable advancements in molecular oncology, immunotherapy, and targeted therapies, long-lasting remission and recurrence prevention remain substantial clinical challenges. The effectiveness of long-term treatment is nevertheless limited by tumor heterogeneity, dynamic molecular evolution, therapeutic resistance, and disease recurrence. In this regard, the rapid development of generative artificial intelligence (GenAI) offers promising opportunities to revolutionize translational oncology, clinical decision-making, and cancer research. Biomedical research and healthcare analytics are evolving due to recent advances in artificial intelligence (AI), including machine learning (ML), deep learning (DL), large language models (LLMs), and multimodal large language models (MLLMs). These computational tools enhance data-driven precision oncology techniques by facilitating the large-scale integration and analysis of diverse medical datasets[1,2]. AI-assisted approaches for prognosis prediction, treatment stratification, early cancer diagnosis, biomarker discovery, and clinical workflow optimization are being increasingly investigated. Furthermore, as noted by Agha et al, the increasing focus on reproducibility and transparency in AI-driven healthcare research emphasizes the significance of strong reporting systems and data governance norms[3]. Among emerging AI technologies, LLMs and MLLMs have garnered significant attention for their ability to process and synthesize complex biomedical information at an unprecedented scale. LLMs are deep learning-based architectures trained on extensive textual datasets. They are capable of identifying clinically relevant patterns from electronic health records, pathology reports, scientific literature, and genomic annotations. These models have shown promise in enhancing prognosis estimation, facilitating personalized treatment planning, enabling early diagnosis, and improving clinical decision support systems. Simultaneously, MLLMs enhance these capabilities by integrating multiple modalities, including radiological imaging, histopathology, genomics, laboratory parameters, and clinical narratives. This multimodal integration holds significant value in oncology, given that cancer biology is inherently multidimensional and dynamic. MLLMs can improve tumor categorization, risk stratification, and disease progression prediction by combining radiological images with patient-specific clinical and genomic data. These systems automate medical record analysis, report generation, and extraction of information from unstructured datasets, reducing clinicians’ workload[4,5]. However, poor clinical validation, algorithmic bias, limited interpretability, high computational requirements, and concerns about data privacy, transparency, and regulatory control limit their transformational potential. In this context, the Delphi-2 M model, which analyzes a patient’s lifestyle and medical records to support early diagnosis and disease risk assessment, highlights the vital role of AI-assisted models. This model can be trained on cancer datasets, including patients’ medical records, lifestyle data, and cancer databases, for early detection[6,7]. GenAI is becoming more important in oncology research and treatment, according to recent studies. Concerning the complexity and advancement of the effort made by the research, Ruma et al, highlighted the importance of GenAI and LLMs for the development of personalized medicine and also discussed the limitations, followed by the data aspects and their integration with AI, which will assist with real-time clinical decisions[8]. Odhisho et al, highlighted the use of GenAI and designed a framework that will assist in expediting medical data extraction for clinical and research use cases by using OpenAI’s GPT-4.0, following the prostate MRI report[9]. Similarly, Rizkala et al, have also highlighted the comparison of ChatGPT-4.0 versus a DiCE summary, discussing the requirement for readability optimization and the need for future text to follow the prompts and model[10]. The system may incorporate lifestyle factors, genomic profiles, imaging biomarkers, and clinical histories for precision screening and customized cancer management. GenAI technology may improve cancer drug discovery, biomarker identification, therapeutic target prediction, and understanding of tumor biology. These models help identify novel carcinogenic pathways and accelerate translational research by efficiently analyzing large-scale genomic, transcriptomic, proteomic, imaging, and clinical data. AI can optimize experimental design, data interpretation, manual processing time, and the construction of robust predictive models for cancer susceptibility, treatment response, and patient survival. Despite these potential developments, the incorporation of GenAI into oncology must address ethical, technical, and therapeutic issues. Safe and reliable implementation requires high-quality non-redundant datasets, transparent model creation, external validation, interpretability, and rigorous prospective clinical evaluation. To ensure that GenAI technologies lead to clinically important and equitable advances in cancer care, oncologists, computer scientists, bioinformaticians, ethicists, and regulatory bodies must collaborate.