Abstract 6246: Graph-based integration of multi-modal data uncovers cell-type specific vulnerabilities and advances toward a multi-scale digital twin

Abstract Background: While recurrent cancer drivers have facilitated drug discovery, strong driver signals are elusive in rare and heterogeneous cancers. Importantly, therapeutically actionable molecular vulnerabilities often extend beyond conventional cancer drivers. We previously developed a graph-based AI approach that identified key molecular vulnerabilities across 19 solid tumor types leveraging bulk sequencing data, outperforming standard methods in target hypothesis generation by three fold (https://doi.org/10.1101/2024.06.04.597409). However, a critical factor in cancer progression and therapeutic resistance lies in the interplay between tumor cells with their microenvironment, surrounding tissue and beyond. There is a real need for a generalizable multiscale communication model that captures molecular and cellular interactions within cells, between cells, and across cellular niches and organs. Such a model could pave the way for constructing a comprehensive multiscale digital twin, offering a transformative framework for understanding cancer dynamics and identifying novel therapeutic cancer targets. Methods: Graph-based computational models provide a powerful framework for analyzing complex biological systems, making them especially effective for any nodal communication, where a node can represent an individual molecule, a cell, a group of cells or an organ. We introduce a probabilistic graph-based approach that integrates data across multiple modalities and scales—from bulk to single-cell resolution—through network fusion to identify cell-specific vulnerabilities. This method, FN-MVP (Fusion Net-Molecular Vulnerability Picker), is, to our knowledge, the first molecular resolution multiscale model able to model intracellular molecular networks together with cell-cell, cell-niche communication. It unifies data from both single cell and bulk data to provide a comprehensive view of these combined modalities. Results: By leveraging the cooperativity of weak signals within cancer molecular networks, FN-MVP amplifies true molecular vulnerabilities with greater accuracy (significantly higher than bulk MVP, p = 1.3e-3), as validated by genetic perturbation data. Applying FN-MVP to lung cancer patients revealed cell subtype-specific vulnerabilities and identified critical ligands and receptors that drive TME dynamics in a context-specific manner. Conclusions: These findings underscore the power of this approach in uncovering actionable molecular interactions. FN-MVP represents a foundational step toward building a generalizable multiscale communication model, capable of integrating molecular, cellular, and tissue-level interactions. This progress lays the groundwork for developing a multiscale digital twin, a transformative tool for understanding cancer and advancing therapeutic strategies. Citation Format: Ying Zhu, Li Zhao, David Maxwell, Michael Darkoh, Bissan Al-Lazikani. Graph-based integration of multi-modal data uncovers cell-type specific vulnerabilities and advances toward a multi-scale digital twin [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 6246.