Single-cell RNA-seq data augmentation using generative Fourier transformer.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity but is often limited by small sample sizes.
• Data scarcity hinders statistically reliable conclusions, especially for rare cell types or diseases.
• Existing deep learning generative models (GMs) struggle with data inadequacy due to pre-training dependencies.

Purpose of the Study:

• To introduce scGFT (single-cell Generative Fourier Transformer), a train-free generative model for synthesizing realistic single cells.
• To address the challenge of limited cell numbers in scRNA-seq data analysis.
• To provide a scalable solution for data augmentation in cell-targeted research.

Main Methods:

• Developed scGFT, a cell-centric, train-free generative model utilizing a Fourier Transformer architecture.
• Validated scGFT using both simulated and experimental scRNA-seq data.
• Compared scGFT's performance against leading neural network-based generative models.

Main Results:

• scGFT successfully synthesizes single cells with natural gene expression profiles, preserving intrinsic data characteristics.
• Demonstrated the mathematical rigor and effectiveness of scGFT in data augmentation.
• scGFT outperformed existing generative models in synthesizing realistic single-cell data.

Conclusions:

• scGFT offers a robust and scalable approach to mitigate data scarcity in single-cell genomics.
• The train-free nature of scGFT overcomes limitations of traditional deep learning models.
• This method enhances the statistical power and reliability of cell-targeted research.