SCDD: a novel single-cell RNA-seq imputation method with diffusion and denoising.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity and evolution.
• Dropout events, common in scRNA-seq data, pose challenges for accurate analysis.
• Existing imputation methods often suffer from over-smoothing, limiting their effectiveness.

Purpose of the Study:

• To develop a novel imputation method for large-scale scRNA-seq data.
• To address the over-smoothing problem prevalent in current imputation techniques.
• To improve the accuracy and utility of scRNA-seq data for downstream biological insights.

Main Methods:

• A two-stage diffusion-denoising approach named SCDD was proposed.
• Stage 1: Diffusion imputation leverages similar cell expression profiles for dropout sites.
• Stage 2: A joint model with graph convolutional neural networks and contractive autoencoders refines imputation and removes noise.

Main Results:

• SCDD effectively suppresses the over-smoothing issue in scRNA-seq data imputation.
• The method demonstrates significant improvements in downstream analyses, including clustering and trajectory analysis.
• Experimental results validate the efficacy of SCDD for large-scale scRNA-seq data.

Conclusions:

• SCDD offers a robust solution for imputing dropout events in scRNA-seq data.
• The proposed method enhances the reliability of scRNA-seq data for biological discovery.
• SCDD represents a significant advancement in computational approaches for single-cell data analysis.