Abstract 6306: Topological arrangement of samples based on high dimensional feature similarities: the NetFlow framework

Abstract At the core of exploratory data analysis is the development of effective informatics tools to represent and visualize how samples are organized in a dataset. Cancer heterogeneity between and within cancer types poses challenges for many clustering and dimensional reduction approaches as biological data often vary continuously rather than form distinct clusters. It is therefore essential to develop a visualization tool that can expose the underlying clusters as well as the broader relational structure of the samples in the data. We introduce NetFlow, an open-source software tool designed to address this need. NetFlow serves as a comprehensive and interactive investigative platform to aid researchers in visualizing and analyzing complex biological data to support the generation and evaluation of hypotheses. NetFlow arranges samples in an undirected graph based on similarities of the underlying high dimensional feature set. It takes biological feature data and provides a visually informative relational topology where each node represents a sample. To incorporate ordering and clustering behavior, the relational topology is derived from the underlying metric structure and constructed by combining the pseudo-ordering of the samples, generated via lineage tracing techniques, with nearest-neighbor information. In addition to classic metrics, NetFlow uses Optimal Mass Transport which offers a network-level and robust metric to compare high dimensional sample profiles with respect to a feature network. NetFlow supports the computation of kernel transformations and diffusion distances to account for local and multi-scale geometry which influence aspects of the similarity structure that will be reflected in the extracted organizational representation. NetFlow is designed in a modular fashion to facilitate easy integration with diverse computational pipelines. The modules support data input/output, similarity structure derivation, visualization, and down-stream analyses. A key feature of NetFlow is its interactive visualization dashboard for exploring the resulting relational topology and its associations with biological or clinical features. The dashboard offers multiple capabilities for customizing the visualization and enhancing interpretability. For instance, the user can select features of interest to color-code the nodes (i.e., samples) and edges (relationships), and can select samples of interest to identify statistically significant characteristic features. These functionalities empower users to gain deeper insights into the underlying data structure and its potential biological or clinical implications. NetFlow results in an abstract sample-space network rather than a 2D spatial arrangement. Its topological nature enables interpretable, uncluttered visualizations of relations within and between clusters for identifying novel inter-sample similarities. Citation Format: Rena Elkin, Jung H. Oh, Joseph O. Deasy. Topological arrangement of samples based on high dimensional feature similarities: the NetFlow framework [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 6306.