Prediction of Single-Cell perturbation response based on Direction-Constrained diffusion Schrödinger Bridge

Area of Science:

• Computational Biology
• Systems Biology
• Genomics

Background:

• Single-cell transcriptional responses to perturbations are crucial for understanding gene regulatory networks, drug discovery, and personalized medicine.
• Increasing perturbation conditions lead to data sparsity, challenging the capture of dynamic responses and necessitating advanced computational modeling.

Purpose of the Study:

• To develop a generative framework for modeling single-cell transcriptional responses to perturbations.
• To address data sparsity and improve the prediction of dynamic gene expression changes under various conditions.

Main Methods:

• Introducing Direction-Constrained Diffusion Schrödinger Bridge (DC-DSB), a generative framework learning probabilistic trajectories between unperturbed and perturbed states.
• Utilizing hierarchical representations from experimental variables and biological priors for enhanced conditional control.
• Implementing a direction-constrained conditioning strategy to inject condition signals along biologically relevant trajectories, improving model quality and stability.

Main Results:

• DC-DSB demonstrates improved accuracy and generalization in predicting gene expression compared to existing methods.
• The framework models dynamic expression trajectories and co-expression structures, enabling the discovery of synergistic and antagonistic gene interactions.
• DC-DSB facilitates the progressive reconstruction of gene regulatory pathways.

Conclusions:

• DC-DSB offers a biologically consistent and generalizable approach to single-cell perturbation modeling.
• The trajectory-based and condition-aware architecture overcomes limitations of static models.
• This framework supports advanced downstream analyses in gene regulation and accelerates drug discovery.