Transfer learning of multicellular organization via single-cell and spatial transcriptomics.

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• Biological tissues display intricate gene expression and multicellular patterns crucial for understanding development and disease.
• Single-cell RNA sequencing (scRNA-seq) offers comprehensive gene coverage but lacks spatial context.
• Spatial transcriptomics (ST) provides spatial information but has limitations in gene profiling.

Purpose of the Study:

• To develop a novel computational method for deciphering cellular spatial organization by integrating scRNA-seq and ST data.
• To identify spatial-organizing genes (SOGs) that drive observed multicellular patterns.
• To infer pseudo-growth trajectories and analyze tissue development and disease mechanisms.

Main Methods:

• Developed iSORT, a transfer learning method utilizing a neural network to map gene expression to spatial locations.
• Integrated scRNA-seq and ST data to reconstruct multicellular organization at a single-cell scale.
• Applied iSORT to diverse biological systems including human cortex, mouse embryo, and atherosclerotic arteries.

Main Results:

• iSORT accurately reconstructs multicellular organization across various biological systems.
• Identified spatial-organizing genes (SOGs) critical for tissue patterning.
• Demonstrated that SOGs in atherosclerotic arteries are strongly linked to vascular structural anomalies.
• Inferred pseudo-growth trajectories using the SpaRNA velocity concept.

Conclusions:

• iSORT is an accurate and practical tool for integrating scRNA-seq and ST data to understand spatial biology.
• The identified SOGs provide insights into the molecular mechanisms underlying tissue development and disease pathology.
• This approach advances the study of multicellular organization and spatial genomics.