Multi-domain translation between single-cell imaging and sequencing data using autoencoders.

Area of Science:

• Single-cell biology
• Computational biology
• Immunology

Background:

• Single-cell methods capture diverse data modalities like imaging and sequencing, crucial for identifying heterogeneous cell states.
• Integrating these modalities is key to understanding cellular heterogeneity and function.
• Coupling imaging and sequencing data at the single-cell level remains a significant challenge.

Purpose of the Study:

• To develop a computational framework for integrating disparate single-cell data modalities, specifically imaging and sequencing.
• To address the challenge of combining information from modalities not measurable simultaneously within the same cell.
• To identify distinct cell subpopulations by integrating single-cell RNA sequencing and chromatin imaging data.

Main Methods:

• Utilized autoencoders to learn a probabilistic coupling between different data modalities.
• Mapped diverse data modalities into a shared latent space for integration.
• Applied the approach to single-cell RNA sequencing and chromatin images of human naive CD4+ T-cells.

Main Results:

• Successfully integrated single-cell RNA sequencing and chromatin imaging data.
• Identified distinct subpopulations of human naive CD4+ T-cells poised for activation.
• Demonstrated the framework's ability to translate between and integrate different data modalities.

Conclusions:

• The developed approach provides a robust framework for integrating and translating between diverse single-cell data modalities.
• This method enables deeper insights into cellular heterogeneity and function for biomedical discovery.
• Facilitates the study of cell states using complementary data types not measurable in the same cell.