CFM-GP: unified conditional flow matching to learn gene perturbation across cell types

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Understanding gene perturbations is key for functional genomics and drug discovery.
• Current computational models struggle with continuous dynamics and shared biological processes across cell types.

Purpose of the Study:

• To introduce CFM-GP, a conditional flow-matching framework for modeling gene perturbation effects.
• To develop a unified computational approach that captures both common and cell-type-specific responses.

Main Methods:

• CFM-GP uses a conditional flow-matching framework to learn a continuous vector field.
• This model transforms control expression profiles into perturbed states, conditioned on cell type.
• The unified architecture integrates common regulatory programs and type-specific responses.

Main Results:

• CFM-GP outperformed existing methods in predictive accuracy and distributional alignment across five datasets.
• The model demonstrated strong cross-species generalization capabilities.
• Inferred flow trajectories revealed interpretable biological pathways and transcriptional cascades.

Conclusions:

• CFM-GP provides a scalable foundation for modeling cellular perturbation responses.
• The framework enables data-driven exploration of gene function and therapeutic strategies in diverse cellular systems.
• This approach enhances our understanding of gene function and intervention strategies.