Abstract High-dimensional biomedical studies increasingly require comprehensive exploratory tools that jointly model, integrate, and visualize sample organization across modalities within a single, interpretable framework. We therefore developed NetFlow, a computational framework that constructs a Pseudo-Organizational StructurE (POSE) graph of sample relationships from high-dimensional and multimodal data, preserving both continuous variation and local neighborhoods while enabling downstream analytics in the same space. Glioblastoma (GBM) exhibits pronounced heterogeneity across transcriptomic, epigenomic, and miRNA profiles that complicates subtype discovery and interpretation. To demonstrate NetFlow’s multi-modal integration capability, we applied it to TCGA GBM (n=213) with matched mRNA expression, DNA methylation, and miRNA profiles. Per-modality sample-sample distances were computed, each was transformed to a common range, and then fused into a unified similarity matrix. Next, diffusion-based, multi-scale metrics on the fused similarity seeded a lineage-tracing-inspired backbone augmented with mutual nearest neighbor edges to build the POSE. The NetFlow framework enabled POSE-aware clustering, survival analysis, and cross-omic differential feature testing. The POSE revealed a distinct lower-risk GBM subgroup with significantly improved overall survival (log-rank P=1.19x10-5) that was not reproducible from any single modality alone. This subgroup exhibited a cross-omic signature including reduced mRNA expression of EMP3 and TIMP1, CRIP1 hypermethylation, and decreased miR-222 expression. Additionally, the identified subgroup showed clinical correlates consistent with favorable prognosis including enrichment for IDH1 mutation and younger age. We further quantified edge-level modality influence, which indicated balanced contributions from each modality, supporting genuine multi-modal integration rather than dominance by any single data type. These results establish that NetFlow yields an interpretable, modality-balanced POSE. They further demonstrate that its unified modeling, integration, visualization, and analysis enables discovery of prognostically relevant subgroups and cross-omic biomarkers. More broadly, NetFlow provides a practical, extensible framework for comprehensive exploratory data analysis in multi-modal oncology cohorts that generalizes beyond GBM to accelerate hypothesis generation and subtype refinement in diverse settings. Citation Format: Rena Elkin, Jung Hun Oh, Anish K. Simhal, Joseph O. Deasy, . Multi-modal integration in the NetFlow framework for comprehensive exploratory data analysis 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 5520.
Elkin et al. (Fri,) studied this question.