Abstract 7461: Identifying molecular vulnerabilities in rare cancers using a novel graph-based therapeutic discovery model

Abstract Background: Rare cancers, as defined by the WHO criteria with an incidence of fewer than 6 cases per 100, 000 people annually, pose significant challenges in diagnosis and treatment due to their low prevalence, heterogeneity, and limited therapeutic options. Patients with rare cancers represent a historically neglected population. Traditional statistical approaches predominantly focus on altered genes, overlooking molecular vulnerabilities that are not explicitly mutated but could still be exploited therapeutically. Methods: We analyzed a cohort of over 1, 000 patients spanning more than 120 distinct rare cancer types (Futreal, A., "Rare cancer patient profiling and engagement", AACR Annual Meeting 2022). By systematically evaluating whole-exome sequencing and RNA sequencing data, we identified genes representing key alterations across the cohort. To uncover both altered and latent molecular vulnerabilities, we applied an integrative, generalizable graph-based AI model explicitly designed for therapeutic discovery. Unlike traditional methods, this model integrates multimodal patient data into a network-based framework, connects genomic alterations with latent dependencies, and identifies hidden signals of molecular vulnerabilities, irrespective of their mutational status. Results were validated using public resources, including the DepMap dataset, without incorporating cell line data as inputs or for model training. Results: Unsupervised clustering revealed 15 molecular subgroups across the cohort. Some clusters, such as CNS (Brain and Spine), exhibited high molecular homogeneity (e.g., 92% of cluster members), while heterogeneous diagnoses like sarcoma were distributed across four clusters, reflecting distinct molecular subtypes. For each cluster, we constructed networks to predict molecular vulnerabilities. Each subgroup yielded at least five prioritized targets associated with FDA-approved or late-stage investigational drugs. Additionally, we identified novel targets that hold promise for innovative drug discovery, offering new avenues for therapeutic exploration. Validation using DepMap data confirmed significant enrichment of these predicted vulnerabilities. Further integration with PRISM and the Genomics of Drug Sensitivity in Cancer (GDSC) datasets refined and prioritized key therapeutic opportunities. Conclusions: This study demonstrates the power of a graph-based therapeutic discovery model to identify key molecular vulnerabilities in rare cancers. By uncovering both altered and latent dependencies, we have laid the foundation for safe and effective therapeutic development in these neglected malignancies. This integrative approach provides a scalable framework for advancing precision oncology in rare tumors, addressing a critical unmet need for novel and repurposed therapeutic options. Citation Format: Li Zhao, Ying Zhu, Randy Chu, Haifeng Zhu, Stephanie T. Schmidt, Yuntao Yang, Chunjie Jiang, Mark Knafl, Xingzhi Song, Xiaogang Wu, Jianhua Zhang, Sabitha Prabhakaran, Scott Woodman, Timothy Heffernan, Elizabeth Burton, Rare Tumor Initiative Team, Andy Futreal, Bissan Al-Lazikani. Identifying molecular vulnerabilities in rare cancers using a novel graph-based therapeutic discovery model [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 7461.