Tokenvizz: GraphRAG-Inspired Tokenization Tool for Genomic Data Discovery and Visualization

Abstract

SummaryOne of the primary challenges in biomedical research is the interpretation of complex genomic relationships and the prediction of functional interactions across the genome. Tokenvizz is a novel tool for genomic analysis that enhances data discovery and visualization by combining GraphRAG-inspired tokenization with graph-based modeling. In Tokenvizz, genomic sequences are represented as graphs, where sequence k-mers (tokens) serve as nodes and attention scores as edge weights, enabling researchers to visually interpret complex, non-linear relationships within DNA sequences. Through a web-based visualization interface, researchers can interactively explore these genomic relationships and extract biologically meaningful insights about regulatory patterns and functional elements. Applied to promoter-enhancer interaction prediction tasks, Tokenvizz outperformed traditional sequential models while providing interpretable insights into genomic features, demonstrating the advantage of graph-based representations for biological discovery. Availability and ImplementationTokenvizz, along with its user guide, is freely accessible on GitHub at: https://github.com/ceragoguztuzun/tokenvizz. ACM Reference FormatCera[g] O[g]uztuzun, Zhenxiang Gao, and Rong Xu. 2024. Tokenvizz: GraphRAG Inspired Tokenization Tool for Genomic Data Discovery and Visualization. In Proceedings of (Bioinformatics). ACM, New York, NY, USA, 7 pages. https://doi.org/XXXXXXX.XXXXXXX