CausalGeD: Blending Causality and Diffusion for Spatial Gene Expression Generation.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Integrating single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics (ST) data is vital for understanding gene expression within its spatial context.
• Current integration methods show limited performance, often failing to capture complex gene interactions and achieving structural similarity below 60%.

Purpose of the Study:

• To develop a novel computational framework, CausalGeD, for accurate integration of scRNA-seq and ST data.
• To leverage causal relationships between genes to enhance the performance of spatial transcriptomic data integration.

Main Methods:

• CausalGeD combines diffusion and autoregressive processes to model gene causal relationships.
• The model generalizes the Causal Attention Transformer, originally used for image generation, to gene expression data.
• This approach captures gene regulatory mechanisms without requiring predefined relationships.

Main Results:

• CausalGeD demonstrated superior performance across 10 diverse tissue datasets compared to state-of-the-art methods.
• Key metrics, including Pearson's correlation and structural similarity, showed improvements of 5-32% with CausalGeD.
• The model successfully captured complex gene regulatory mechanisms, advancing both technical integration and biological interpretation.

Conclusions:

• CausalGeD offers a significant advancement in integrating scRNA-seq and ST data by incorporating gene causal relationships.
• The method provides more accurate and biologically meaningful insights into spatial gene expression patterns.
• This work paves the way for deeper understanding of tissue architecture and cellular functions through integrated omics data.