Matrix factorization and transfer learning uncover regulatory biology across multiple single-cell ATAC-seq data sets

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Existing single-cell ATAC-seq (scATAC-seq) methods excel at cell type clustering but lack robust strategies for integrating multiple datasets or modalities.
• The challenge of harmonizing scATAC-seq data across different experiments and sequencing types remains a significant hurdle in the field.

Purpose of the Study:

• To develop and validate an analysis framework for integrating multiple scATAC-seq datasets.
• To identify common regulatory patterns across diverse scATAC-seq data.
• To integrate scATAC-seq with scRNA-seq data to provide orthogonal evidence for predicted transcriptional regulators.

Main Methods:

• Application of the CoGAPS Matrix Factorization algorithm for pattern identification.
• Utilization of the projectR transfer learning program for data integration.
• Integration with single-cell RNA sequencing (scRNA-seq) data for cross-validation.

Main Results:

• The developed framework successfully identifies common regulatory patterns within and across multiple scATAC-seq datasets.
• These patterns accurately characterize cell types, demonstrating the framework's efficacy.
• The identified patterns align with existing biological knowledge and reveal novel regulatory insights.

Conclusions:

• The proposed analysis framework provides a powerful tool for integrating multiple scATAC-seq datasets, overcoming current limitations.
• This approach enhances the ability to identify cell types and discover regulatory mechanisms.
• The integration with scRNA-seq data strengthens the biological interpretation and validation of findings.