scDiffusion: conditional generation of high-quality single-cell data using diffusion model.

Area of Science:

• Computational Biology
• Genomics
• Artificial Intelligence

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular processes but acquiring sufficient high-quality data remains a challenge.
• Existing generative models struggle to produce realistic synthetic scRNA-seq data, especially under controlled conditions.

Purpose of the Study:

• To develop a novel generative model, scDiffusion, for producing high-fidelity synthetic scRNA-seq data.
• To enable the generation of scRNA-seq data with controlled conditions and explore continuous cell development trajectories.

Main Methods:

• scDiffusion integrates diffusion models with foundation models.
• Multiple classifiers guide the diffusion process for multi-condition data generation.
• The Gradient Interpolation strategy enables the generation of continuous cell development trajectories.

Main Results:

• scDiffusion generates synthetic scRNA-seq data that closely mimics real data.
• The model can conditionally generate data for specific cell types, including rare ones.
• scDiffusion successfully generated a continuous developmental trajectory of mouse embryonic cells, demonstrating its utility in cell fate research.

Conclusions:

• scDiffusion is an effective tool for augmenting scRNA-seq datasets.
• The model provides valuable insights into cell development and fate determination.
• scDiffusion facilitates the generation of novel cell types not present in the training data.