Cross-species cell-type assignment from single-cell RNA-seq data by a heterogeneous graph neural network

Area of Science:

• Computational Biology
• Genomics
• Evolutionary Biology

Background:

• Comparative single-cell RNA sequencing (scRNA-seq) analysis reveals cellular diversity and evolution.
• Accurate cell-type assignment is critical but challenging for nonmodel species due to poor genome annotation and limited biomarkers.

Purpose of the Study:

• To develop a novel computational model for robust cross-species cell-type assignment and gene module extraction from scRNA-seq data.
• To improve the understanding of cellular diversity and evolutionary mechanisms across species.

Main Methods:

• Development of CAME, a heterogeneous graph neural network model.
• Learning aligned and interpretable cell and gene embeddings.
• Utilizing non-one-to-one homologous gene mapping for enhanced cross-species analysis.

Main Results:

• CAME significantly outperforms five classical methods in cell-type assignment accuracy and robustness.
• The model successfully transfers human brain cell types to mice and discovers shared cell-type-specific functions.
• CAME aligns spermatogenesis trajectories in humans and macaques, revealing conserved expression dynamics.

Conclusions:

• CAME provides accurate cross-species cell-type assignments, particularly for nonmodel organisms.
• The model effectively uncovers shared and divergent characteristics between species using scRNA-seq data.
• CAME facilitates the exploration of cellular evolution and function across diverse species.