Annotating Spatial Multi-Omics Spot-Level Niche Types Using Bi-View Retrieval-Augmented Generation With SpotTypeLLM.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial multi-omics generates complex data without labels, requiring manual annotation.
• Current large language models (LLMs) struggle with high-dimensional omics data and biological context.

Purpose of the Study:

• To develop an automated framework (SpotTypeLLM) for spatial multi-omics spot-level niche type annotation.
• To overcome limitations of manual annotation and existing computational methods.

Main Methods:

• SpotTypeLLM utilizes a Bi-view Retrieval-Augmented Generation (BiRAG) approach tailored for LLMs.
• Data encoding involves scLLM-based embeddings, graph convolutional networks (GCNs), and multi-view attention.
• Spot-specific genes are selected to construct LLM prompts.

Main Results:

• SpotTypeLLM demonstrates superior accuracy compared to ten state-of-the-art methods on labeled datasets.
• The framework effectively handles data sparsity and heterogeneity for niche assignment.
• Validated practical utility in enhancing spatial multi-omics analysis.

Conclusions:

• SpotTypeLLM offers an effective solution for automated spatial multi-omics annotation.
• The framework improves upon existing methods, addressing LLM limitations in biological data analysis.
• Enables more efficient and accurate biological discovery from spatial omics data.