A Comprehensive Clinical Inference and Decision Support Tool for Multiple Myeloma using Agentic Large Language Models

There is a critical need for advanced, explainable AI tools to support therapeutic decision-making in Multiple Myeloma (MM), a biologically complex, incurable, and highly heterogeneous cancer. Despite extensive molecular and clinical profiling, treatment selection remains uncertain, particularly for pre-malignant, high-risk, or therapy-refractory patients. The diversity in patient-specific genetic and functional profiles, combined with a lack of consensus on optimal therapeutic strategies, highlights the limitations of current predictive models. To address this, we developed a generative AI pipeline tailored for MM that integrates molecular, clinical, and functional data with contextual knowledge from biomedical literature. By leveraging a fine-tuned large language model enhanced with Retrieval-Augmented Generation (RAG), this system provides patient-specific, literature-based insights into disease biology and treatment options. This approach enables explainable, data-driven predictions that not only improve therapeutic stratification but also pave the way for precision medicine in MM, especially in complex or under-characterized patient subsets.