Transfer learning efficiently maps bone marrow cell types from mouse to human using single-cell RNA sequencing.

Area of Science:

• Computational Biology
• Immunology
• Genomics

Background:

• Biomedical research frequently uses model organisms to understand human biology, but direct translation can be challenging.
• Previous cross-species comparative studies were limited by bulk-cell data, hindering detailed biological insights.
• Single-cell RNA sequencing (scRNA-seq) offers high-resolution cellular data for more accurate comparisons.

Purpose of the Study:

• To investigate the utility of transfer learning for mapping bone marrow biology between mice and humans.
• To develop and evaluate a machine learning model for cell type identification across species.
• To demonstrate efficient cross-species biological data reconstruction using limited datasets.

Main Methods:

• Trained a multiclass logistic regression model on mouse bone marrow scRNA-seq data for cell type classification.
• Applied the trained model to identify human bone marrow cells, assessing accuracy and limitations.
• Re-trained the model using human data and explored zero-shot learning for cross-species inference.

Main Results:

• The logistic regression model achieved performance comparable to complex neural networks in classifying mouse bone marrow cells.
• The model identified human bone marrow cells with 83% overall accuracy, though some cell types were challenging.
• Re-training with minimal human data (<10 cells/type) rapidly improved classifier accuracy for human cell types.
• Zero-shot learning successfully inferred some human cell identities directly from the mouse-trained model.

Conclusions:

• Transfer learning provides an efficient method for mapping bone marrow biology between species using scRNA-seq data.
• Simple machine learning models can effectively reconstruct complex biological systems from limited cross-species data.
• This approach has significant implications for advancing comparative biomedical research and drug discovery.