scDM: A deep generative method for cell surface protein prediction with diffusion model.

Area of Science:

• Single-cell biology
• Computational biology
• Genomics and proteomics

Background:

• Proteins execute organismal functions, with RNA to protein transitions undergoing post-transcriptional regulation.
• Simultaneous analysis of RNA and surface protein expression offers deeper biological insights.
• Cellular indexing of transcriptomes and epitopes by sequencing (CITE-Seq) measures both RNA and protein expression but is costly and time-intensive.

Purpose of the Study:

• To develop a computational tool for predicting surface protein expression from RNA expression data.
• To leverage CITE-Seq datasets for designing a deep generative prediction model.
• To uncover biological discoveries through accurate protein expression predictions.

Main Methods:

• Designed scDM, a deep generative prediction method based on diffusion models.
• Utilized a novel data encoding strategy within the diffusion model.
• Employed Gaussian noise injection and gradual removal to learn data distribution for prediction.

Main Results:

• Demonstrated satisfactory prediction results across diverse datasets.
• Validated the effectiveness of integrating single-cell multiomics data with diffusion models.
• Showcased the potential of jointly analyzing predicted surface protein expression and cancer cell drug scores.

Conclusions:

• scDM accurately predicts surface protein expression from RNA data in single cells.
• The study highlights the utility of diffusion models in single-cell multiomics analysis.
• Findings suggest new avenues for identifying therapeutic drug targets by integrating protein expression and drug sensitivity data.