Mouse-Geneformer: A deep learning model for mouse single-cell transcriptome and its cross-species utility.
Keita Ito1, Tsubasa Hirakawa2, Shuji Shigenobu3,4
1Graduate School of Engineering, Chubu University, Kasugai, Aichi, Japan.
Plos Genetics
|March 19, 2025
Summary
A new mouse-specific Geneformer model, trained on 21 million single-cell RNA sequencing profiles, accurately analyzes mouse transcriptomes and aids in disease gene discovery. This model also shows cross-species potential for human data analysis.
Area of Science:
- Computational biology and bioinformatics
- Genomics and transcriptomics
- Machine learning in biological research
Background:
- Single-cell RNA sequencing (scRNA-seq) generates complex transcriptome data, necessitating advanced analytical tools.
- Geneformer, a Transformer-based deep learning model, has shown success in human transcriptome analysis.
- A mouse-specific Geneformer is crucial due to the mouse's prominence as a model organism in research.
Purpose of the Study:
- To develop and evaluate a mouse-specific Geneformer (mouse-Geneformer) for analyzing mouse scRNA-seq data.
- To assess the performance of mouse-Geneformer in cell type classification and disease gene identification.
- To investigate the cross-species applicability of mouse-Geneformer to human transcriptome data.
Main Methods:
- Construction of a large-scale mouse transcriptome dataset comprising 21 million scRNA-seq profiles.
- Pre-training the Geneformer architecture on the mouse dataset to create mouse-Geneformer.
- Fine-tuning mouse-Geneformer for downstream tasks like cell type classification and in silico perturbation experiments.
- Cross-species analysis involving ortholog mapping and fine-tuning with human scRNA-seq data.
Main Results:
- Mouse-Geneformer effectively models the mouse transcriptome and improves cell type classification accuracy.
- In silico perturbation experiments identified disease-causing genes, validated in vivo.
- Mouse-Geneformer demonstrated cross-species analysis capabilities, achieving comparable accuracy to human Geneformer on human data after fine-tuning.
- Performance varied in cross-species disease modeling, highlighting the need for species-specific nuances.
Conclusions:
- Mouse-Geneformer is a robust tool for analyzing mouse scRNA-seq data, enhancing biological insights and disease gene discovery.
- The Geneformer architecture is adaptable to different species with sufficient transcriptome data.
- Cross-species analysis shows promise but underscores the importance of species-specific models for complex biological processes.
- Mouse-Geneformer offers potential benefits for human research, especially with data types inaccessible in humans, and for non-model organisms.
