Modeling interpretable correspondence between cell state and perturbation response with CellCap.

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Single-cell transcriptomics combined with perturbations reveals cell-state-specific responses, crucial for understanding molecular mechanisms.
• Current computational methods often overlook response heterogeneity, focusing instead on average cellular responses.
• Identifying cell-state-specific responses is key to discovering novel regulatory pathways and therapeutic targets.

Approach:

• We introduce CellCap, a deep generative model for end-to-end analysis of single-cell perturbation experiments.
• CellCap utilizes sparse dictionary learning in a latent space to deconstruct perturbation responses into transcriptional programs.
• Model interpretability is enhanced through dot-product cross-attention and a linearly-decoded latent space.

Key Points:

• CellCap effectively models cell-state-specific and heterogeneous responses to perturbations.
• The model successfully uncovers relationships between cell state and perturbation response in real datasets.
• Interpretability is demonstrated through simulated scenarios and application to complex experimental data.

Conclusions:

• CellCap provides valuable insights into cellular behaviors during perturbation experiments.
• The model's ability to capture heterogeneous responses advances understanding of complex biological systems.
• CellCap offers a powerful tool for uncovering novel regulatory pathways and potential therapeutic targets.