Abstract P33: Leveraging probabilistic causal disease models to understand molecular pathways and target resistance mechanisms in Multiple Myeloma with the CREBBP and EP300 bromodomain inhibitor, pocenbrodib

Abstract To explore and devise innovative approaches for the treatment of Multiple Myeloma (MM) – a disease often deemed incurable due to the development of resistance mechanisms against available therapies – we adopted an integrative data approach, constructing causal AI models of MM that integrate co-expression relationships, DNA variation, longitudinal clinical data, and expansive knowledge graphs. Diverging from traditional methods, our approach centers on constructing probabilistic causal models to elucidate the regulatory mechanisms of disease and capture existing known causal relationships published in the scientific literature or curated in pathway databases. We employ generative AI and other advanced machine learning approaches to stratify patients into mechanistically-informed subtypes that we can directly associate with clinical outcomes. These models, and the ability to map driver mechanisms to outcomes, serve as a foundation for analyzing our real-world data, a dataset made up of more than 200, 000 cancer patients with multimodal molecular data generated on tumor tissues and matched to longitudinal clinical data. Finally, we can validate and refine therapeutically relevant hypotheses using a scalable high-content organoid and cell line screening platform to characterize drug and/or targeted genetic perturbation signatures. We analyzed our models in MM to reveal sub-populations of relapsed refractory patients that could be susceptible to specific pathway inhibition. This analysis revealed a focused causal subnetwork that predicts poor overall survival in standard MM treatment protocols. We demonstrate that not only does this subnetwork show promise in predicting response to a novel CREBBP and EP300 bromodomain inhibitor, pocenbrodib, but it sheds light on the molecular underpinnings of MM, beyond the currently known IRF4, MYC, and immune cell regulation mechanisms that the inhibition of CREBBP/EP300 is known to target. Our network models reveal that this novel subnetwork is intricately linked to a range of driver mechanisms that strongly predict disease progression and are associated with known mechanisms of resistance. Notably, the signature persists across various MM stages, from initial diagnosis to relapsed/refractory states, and intensifies with each additional line of therapy. Further, signature scores are higher in patients with known high-risk cytogenetic abnormalitiesOur findings suggest that stratifying patients based on this gene signature could be instrumental in identifying individuals at high risk of MM progression. Such stratification could transform the selection process for pocenbrodib treatment. By leveraging an integrated, AI-driven approach, we advance the understanding of MM and pave the way for more targeted and effective therapeutic strategies. Citation Format: Allison E Roder, Alice M Walsh, Taylor K Krebs, Veronica Calvo, Bonnie V Dougherty, Preston Smith, Jens Renstrup, Eric E Schadt. Leveraging probabilistic causal disease models to understand molecular pathways and target resistance mechanisms in Multiple Myeloma with the CREBBP and EP300 bromodomain inhibitor, pocenbrodib abstract. In: Proceedings of the Blood Cancer Discovery Symposium; 2024 Mar 4-6; Boston, MA. Philadelphia (PA): AACR; Blood Cancer Discov 2024;5 (2Suppl): Abstract nr P33.