Genome visualization is an important analytical method for revealing complex variation patterns, supporting disease mechanism research, and advancing precision medicine. With the rapid increase in data volume and heterogeneity, traditional script-driven visualization methods face prominent bottlenecks in multi-omics integration, real-time interaction, and reusability.

Recently, the Advanced Interactive Technology and Application Development Division of our center innovatively developed AuraGenome, a large language model-driven intelligent framework for circular genome visualization. This framework breaks through the traditional paradigm of “manual–script–static” and establishes a new mode of “natural language–agent–interactive.” It enables genomic data to be rapidly transformed into high-quality, interactive visualization results, while supporting traceability and reusability throughout the entire workflow.

In the analysis of acute myeloid leukemia (AML) chromosomal translocations, scientists completed the visualization of structural variants and transcriptional levels within 20 minutes and identified as well as interactively annotated potential biomarker regions. In the replication of the melanoma (COLO-829) mutation map, scientists reproduced within only 7 minutes the classic interactive visualization results reported in a Nature paper, with high fidelity.

Further comparative experiments showed that, compared with the traditional tool Circos, AuraGenome improved efficiency by 69% and accuracy to 89%, truly realizing the shift of “allowing scientists to focus on genomic data rather than the tools themselves.”

This work has been published in IEEE Computer Graphics and Applications. The first author is Chi Zhang, a master’s student at CNIC; Yu Dong, assistant researcher, is the co-first author; and Yang Wang, senior engineer, is the corresponding author. This work was supported by the Strategic Priority Research Program of the Chinese Academy of Sciences, Grant No. XDA0460304.

The Visual Analytics System of LLM-driven Genome Circular Visualization Generation

Zhang, Chi, Yu Dong, Yang Wang^*, Yuetong Han, Guihua Shan, and Bixia Tang. "AuraGenome: An LLM-Powered Framework for On-the-Fly Reusable and Scalable Circular Genome Visualizations." IEEE Computer Graphics and Applications 01 (2025): 1-14.