Cell-to-cell distance that combines gene expression and gene embeddings

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Large-language models (LLMs) are increasingly applied to biological data.
• LLMs generate gene-embedding matrices alongside experimental gene-expression matrices.
• Integrating these diverse data types presents analytical challenges.

Purpose of the Study:

• To develop a method for combining gene-expression and gene-embedding matrices.
• To improve the definition of cell-to-cell distance in single-cell data analysis.
• To enhance the accuracy of cell type clustering.

Main Methods:

• Development of a computationally feasible approach to integrate gene-expression and gene-embedding data.
• Application of the developed method to six real-world single-cell datasets.
• Evaluation of the method's performance in defining cell-to-cell distance and clustering.

Main Results:

• The proposed method effectively combines information from both gene-expression and gene-embedding matrices.
• The integrated approach demonstrates superior performance in clustering cells of the same type across all tested datasets.
• The new cell-to-cell distance measure shows significant advantages over existing methods.

Conclusions:

• The integration of gene-embedding and gene-expression data provides a powerful new approach for single-cell analysis.
• The developed method offers a computationally efficient and effective solution for improving cell-to-cell distance metrics.
• This advancement facilitates more accurate cell type identification and biological interpretation from complex single-cell datasets.