Abstract 4195: OncoGraphDB: Rapid projections of patient-level multimodal graph database facilitates efficient exploration of tumor subtypes and identifies new combinatorial biomarkers of patient outcomes and gene essentiality

Abstract Cancer ‘omics data are often analyzed in isolation of other related data modalities despite the potential for added benefit when analyzing a multimodal model. One strategy is to include all available data, but this approach may obfuscate the underlying signal if data modalities exist in the model that do not significantly contribute to the analysis. To enable efficient exploration of multimodal data, we developed the OncoGraphDB framework to rapidly project subgraphs of a larger, all inclusive, cancer omics knowledge graph. This enabled us to identify subspaces that maximally separate patients of interest using a user-defined metric (i.e. patient survival, drug response, or predicted gene essentiality). We found that across cancer indications, gene expression data provided the most single-mode information, which is consistent with several recently published studies. However, performance was improved when predicting prognosis and shared gene essentiality scores when additional data modalities were included in the analysis, including somatic mutation signatures and DNA methylation profiles. Agentic AI was then applied to the Neo4j graph framework to support interactive interrogation of the cancer -omics knowledge graph providing domain experts an exploratory data portal for hypothesis generation. Finally, we describe the gene expression and image classifiers for projecting new patient samples onto the graph to support N-of-1 reverse translation studies. The OncographDB platform facilitates efficient exploration of complex cancer databases and provides data analysis solutions for precision medicine applications. All Authors were or are employees of AbbVie. The design, study conduct, and financial support for this research were provided by AbbVie. AbbVie participated in the interpretation of data, review, and approval of the publication. No honoraria or payments were made for authorship. Citation Format: Jacob Pfeil, Amber Tse, Severiano Villarruel, Emily Rossi, Liqian Ma, Xi Zhao, Josue Samayoa, Kyle Halliwill. OncoGraphDB: Rapid projections of patient-level multimodal graph database facilitates efficient exploration of tumor subtypes and identifies new combinatorial biomarkers of patient outcomes and gene essentiality [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4195.