scDTL: enhancing single-cell RNA-seq imputation through deep transfer learning with bulk cell information.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution transcriptome views but suffers from dropout events.
• Dropout events in scRNA-seq data present significant challenges for accurate downstream analysis.
• Existing imputation methods may not fully leverage inherent gene relationships.

Purpose of the Study:

• To develop an advanced imputation method for scRNA-seq data.
• To address dropout events by integrating bulk RNA-sequencing information.
• To improve the accuracy of gene expression profiles in single-cell data.

Main Methods:

• Proposed scDTL, a deep transfer learning approach for scRNA-seq data imputation.
• Utilized a denoising autoencoder trained on bulk RNA-seq data as an initial model.
• Employed a domain adaptation framework to transfer knowledge from bulk to single-cell data.
• Incorporated a 1D U-Net denoising model for multi-granularity gene representations.
• Applied a cross-channel attention mechanism to fuse features from both models.

Main Results:

• scDTL demonstrated superior performance compared to existing state-of-the-art imputation methods.
• Quantitative comparisons confirmed the effectiveness of scDTL.
• Downstream analyses showed improved results using scDTL-imputed data.

Conclusions:

• scDTL effectively imputes scRNA-seq data by leveraging bulk RNA-sequencing information.
• The proposed deep transfer learning framework successfully addresses dropout events.
• scDTL offers a robust solution for enhancing scRNA-seq data quality and downstream analysis accuracy.