TransAnno-Net: A Deep Learning Framework for Accurate Cell Type Annotation of Mouse Lung Tissue Using Self-supervised...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for biological research.
• Accurate cell type annotation is essential but costly with traditional methods.
• Self-supervised learning offers a cost-effective alternative by leveraging unlabeled data.

Purpose of the Study:

• To develop an efficient and accurate cell type annotation method for large-scale scRNA-seq data.
• To reduce the reliance on expert labels and associated costs.
• To enhance the transferability and efficiency of cell type annotation models.

Main Methods:

• TransAnno-Net, a deep learning framework utilizing transfer learning and a Transformer architecture.
• Self-supervised pre-training on ~100,000 mouse lung scRNA-seq cells to learn gene-gene similarities.
• Fine-tuning on smaller datasets for specific cell type annotation tasks.
• Application of random oversampling to address class imbalance in scRNA-seq data.

Main Results:

• TransAnno-Net achieved superior performance on three mouse lung datasets with AUCs of 0.979, 0.901, and 0.982.
• Outperformed eight state-of-the-art (SOTA) cell type annotation methods.
• Demonstrated robust performance on cross-organ and cross-platform datasets, competitive with fully supervised methods.

Conclusions:

• TransAnno-Net is a highly effective method for cross-platform and cross-dataset cell type annotation of mouse lung tissues.
• The method supports cross-organ cell type annotation.
• Expected to accelerate research into the biological mechanisms of complex systems and diseases.