Integration and transfer learning of single-cell transcriptomes via cFIT.

Area of Science:

• Computational Biology
• Genomics
• Neuroscience

Background:

• Single-cell RNA sequencing (scRNA-seq) enables comprehensive cell type characterization.
• Integrating diverse scRNA-seq datasets is crucial for understanding cellular identity and function.
• Existing methods face challenges in handling batch effects and transferring knowledge across datasets.

Purpose of the Study:

• To introduce a novel computational method, common factor integration and transfer learning (cFIT), for scRNA-seq data integration.
• To address the challenge of batch effect correction across experiments, technologies, subjects, and species.
• To enable knowledge transfer from high-quality to low-quality datasets for improved cell identification.

Main Methods:

• Developed cFIT, a method modeling shared information in a common factor space.
• Utilized iterative nonnegative matrix factorization (NMF) for parameter learning.
• Employed low-rank matrix for knowledge transfer between datasets.

Main Results:

• cFIT effectively captures batch effects across diverse scRNA-seq datasets.
• The method demonstrated superior reliability in preserving biological variations compared to existing approaches.
• Applied cFIT to human and mouse developing brain scRNA-seq data, revealing cross-species transcriptional similarities.

Conclusions:

• cFIT provides a robust framework for integrating and transferring knowledge from scRNA-seq data.
• The method facilitates the creation of comprehensive cell-type landscapes, particularly in neuroscience.
• cFIT offers valuable insights into developmental processes like brain development.