scDCT: a conditional diffusion-based deep learning model for high-fidelity single-cell cross-modality translation.

Area of Science:

• Computational biology
• Genomics
• Molecular biology

Background:

• Single-cell multi-omics technologies provide deep insights into cellular heterogeneity.
• Existing cross-modality translation methods face challenges with high-dimensional, noisy, and sparse single-cell data.

Purpose of the Study:

• To develop a robust probabilistic framework for bidirectional cross-modality translation in single-cell data.
• To improve the integration of diverse single-cell molecular profiles, including scRNA-seq, scATAC-seq, and protein expression.

Main Methods:

• Proposed single-cell Diffusion models for Cross-modality Translation (scDCT).
• Integrated modality-specific autoencoders with conditional denoising diffusion probabilistic models.
• Developed a framework to map inputs to latent spaces and perform probabilistic translation across modalities.

Main Results:

• scDCT effectively captures cell-type heterogeneity and accounts for data sparsity.
• The model successfully models uncertainty during cross-modality translation.
• Demonstrated superior performance over state-of-the-art methods across various datasets and settings (paired, unpaired, cross-type, cross-tissue).

Conclusions:

• scDCT offers a robust and interpretable solution for single-cell multi-omics integration.
• The proposed framework advances the field of single-cell data analysis by enabling accurate cross-modality translation.
• Highlights the potential of diffusion models in addressing complex challenges in single-cell data integration.