Related Experiment Video
Updated: Jul 2, 2025
10:12
Droplet Barcoding-Based Single Cell Transcriptomics of Adult Mammalian Tissues
Published on: January 10, 2019
18.5K
Toward universal cell embeddings: integrating single-cell RNA-seq datasets across species with SATURN.
Yanay Rosen1, Maria Brbić2, Yusuf Roohani3
1Department of Computer Science, Stanford University, Stanford, CA, USA.
Nature Methods
|February 17, 2024
Summary
SATURN, a novel deep learning method, integrates cross-species single-cell data by using protein language models. This approach enables universal cell embeddings, facilitating evolutionary studies and gene function analysis across diverse species.
Area of Science:
- Computational biology
- Evolutionary biology
- Genomics
Background:
- Single-cell analysis across species reveals cell type evolution.
- Interspecies genomic differences hinder joint analysis to homologous genes.
Purpose of the Study:
- To develop a deep learning method for integrating cross-species single-cell datasets.
- To enable joint analysis regardless of genomic similarity and identify conserved/divergent gene functions.
Main Methods:
- SATURN (Single-cell ATlas Universal Representation Network) uses protein language models to generate gene embeddings.
- Combines protein embeddings with RNA expression data for universal cell embeddings.
- Applies method to whole-organism atlases and embryogenesis datasets.
Main Results:
- SATURN successfully integrates datasets from evolutionarily distant species.
- Enables effective transfer of cell type annotations across species.
- Identifies functionally related genes coexpressed across species and detects divergent gene functions, such as in glaucoma-associated genes.
Conclusions:
- SATURN overcomes genomic barriers in cross-species single-cell analysis.
- Provides a powerful tool for studying cell type conservation and diversification.
- Facilitates comparative genomics and disease gene function studies across species.
