MVCLST: A spatial transcriptome data analysis pipeline for cell type classification based on multi-view comparative...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics offers gene expression, location, and imaging data for biological insights.
• Integrating these diverse data types for cell classification remains a computational challenge.

Purpose of the Study:

• To develop an accurate cell type classification method for spatial transcriptomics data.
• To address the challenge of integrating gene expression, spatial location, and tissue image information.

Main Methods:

• Proposed MVCLST (Multi-View Comparative Learning for Spatial Transcriptomics), a multi-view comparative learning method.
• Constructed two views using gene expression, cell coordinates, and image features.
• Employed four encoders for shared/unique features, contrastive loss for consistency, and decoders for feature fusion.
• Utilized the Leiden algorithm for cell type identification and established the MVCLST-CCFS framework.

Main Results:

• MVCLST enhances feature extraction effectiveness and mitigates erroneous data impacts.
• Achieved excellent clustering results on human prefrontal cortex and mouse brain data.
• Outperformed state-of-the-art methods in identifying highly variable genes across cell types in mouse olfactory bulb data.

Conclusions:

• MVCLST provides an effective computational framework for spatial transcriptomics data analysis.
• The method demonstrates robust performance in cell type classification and gene discovery.
• This approach advances the integration of multi-modal data in single-cell biology.