Cell Type Prediction for Single-Cell RNA Sequencing Utilizing Unsupervised Domain Adaptation and Semi-Supervised...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression measurement at the individual cell level.
• Deep learning methods are increasingly used for cell-type identification in scRNA-seq data.
• Integrating multiple scRNA-seq datasets often introduces batch effects, hindering accurate cell-type prediction.

Purpose of the Study:

• To develop a novel deep learning model, scUDAS, for robust cell-type prediction in scRNA-seq data.
• To address and mitigate batch effects arising from the integration of multiple scRNA-seq datasets.
• To improve the accuracy and reliability of cell-type identification across diverse experimental conditions.

Main Methods:

• scUDAS employs unsupervised domain adaptation and semi-supervised learning (SSL) for batch effect correction.
• The model is pre-trained on a labeled source dataset and then adapted to a target dataset using adversarial training.
• SSL with consistency regularization is utilized to further enhance performance by leveraging both datasets.

Main Results:

• scUDAS effectively reduces distribution differences between scRNA-seq datasets, mitigating batch effects.
• The proposed model demonstrates superior performance compared to existing deep learning-based batch correction methods.
• Accurate cell-type prediction is achieved even when integrating data from different laboratories and experimental setups.

Conclusions:

• scUDAS provides an effective solution for batch effect correction in scRNA-seq data analysis.
• The integration of unsupervised domain adaptation and SSL enhances the robustness of cell-type prediction.
• scUDAS represents a significant advancement for reliable cell-type identification in large-scale scRNA-seq studies.