spatiAlytica: Viewer-Grounded Multimodal Agentic System for Interactive Spatial Omics Analysis

Area of Science:

• Computational Biology
• Bioinformatics
• Spatial Omics

Background:

• Spatial transcriptomics and proteomics offer insights into tissue architecture and cellular interactions.
• Current analysis methods are hindered by programming complexity and limitations of text-based AI.

Purpose of the Study:

• To introduce spatiAlytica, a viewer-centric AI system for non-programmer biologists to conduct spatial omics analysis.
• To enable hypothesis-driven exploration and interpretation of spatial omics data through natural language interaction.

Main Methods:

• spatiAlytica integrates with the Napari viewer, featuring viewer-state serialization, agentic memory, and biological concept mapping.
• The system supports code generation, debugging, Spatial Visual Question Answering (VQA), and grounded interpretation.
• A benchmark, spatiAlyticaBench, was created for evaluating spatial analytical capabilities.

Main Results:

• spatiAlytica demonstrated superior performance compared to baseline agents, utilizing less time and computational resources.
• Case studies on Kaposi's sarcoma, colorectal cancer, and ovarian cancer successfully identified known spatial patterns.
• Progressive CD8 T-cell dysfunction during Kaposi's sarcoma progression was uncovered.

Conclusions:

• spatiAlytica provides an accessible and efficient platform for spatial omics data analysis.
• The system facilitates exploratory analysis and interpretive reasoning for biologists without extensive programming skills.
• spatiAlytica has the potential to advance discoveries in cancer research and other fields utilizing spatial omics data.