scFTAT: a novel cell annotation method integrating FFT and transformer.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) studies are advancing rapidly.
• Current annotation methods struggle with high data sparsity and manual effort.
• Automating cell annotation is crucial for large-scale scRNA-seq data analysis.

Purpose of the Study:

• To develop a novel model for automatic single-cell data annotation.
• To address challenges of data sparsity and manual annotation in scRNA-seq.
• To improve the accuracy and efficiency of cell type identification.

Main Methods:

• Proposed scFTAT model integrating Fast Fourier Transform (FFT) and an enhanced Transformer.
• Utilized Linear Discriminant Analysis (LDA) to reduce data sparsity.
• Incorporated kernel approximation, position encoding enhancement, and attention enhancement for fine-tuning.

Main Results:

• scFTAT achieved high accuracy and robustness on six typical scRNA-seq datasets.
• Demonstrated superior performance compared to existing annotation tools.
• Achieved 0.93 accuracy and 0.89 MCC on human kidney data, outperforming others (0.92 accuracy, 0.85 MCC).

Conclusions:

• scFTAT effectively integrates FFT and enhanced Transformer for automatic feature learning.
• The model successfully addresses sparsity and manual annotation challenges in scRNA-seq.
• Experimental results validate the efficiency and robustness of scFTAT for scRNA-seq data analysis.