scPerb: Predict single-cell perturbation via style transfer-based variational autoencoder.

Area of Science:

• Computational biology
• Single-cell genomics
• Systems biology

Background:

• Traditional methods for analyzing cellular responses to perturbations are labor-intensive and costly.
• Existing computational approaches often fail to adequately distinguish perturbation effects from cell-type specific patterns.
• Accurate prediction of cellular responses is vital for advancing computational biology and drug discovery.

Purpose of the Study:

• Introduce scPerb, a novel computational framework for predicting cellular responses to perturbations at the single-cell level.
• Explicitly extract and transfer perturbation-related variances from unperturbed to perturbed cells.
• Overcome limitations of existing methods in distinguishing perturbation effects.

Main Methods:

• Utilizes a style transfer strategy within a variational autoencoder architecture.
• Incorporates a style encoder to capture latent representation differences between cell states.
• Enables accurate prediction of gene expression post-perturbation.

Main Results:

• scPerb demonstrates superior performance and accuracy compared to existing methods.
• Achieved high R-squared values (0.98, 0.98, 0.96) on benchmarking datasets.
• Effectively separates and transfers perturbation-related variances for enhanced prediction.

Conclusions:

• scPerb represents a significant advancement in predicting cellular responses to perturbations.
• The framework provides a robust tool for computational biology, improving prediction accuracy.
• Addresses key limitations of current methodologies in analyzing perturbation effects.