CoupleVAE: coupled variational autoencoders for predicting perturbational single-cell RNA sequencing data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell sequencing enables detailed genetic analysis of individual cells.
• Understanding cellular responses to perturbations is vital for biological insights.
• Experimental acquisition of post-perturbation cellular states is often cost-prohibitive.

Purpose of the Study:

• To develop a novel deep learning method for predicting single-cell RNA sequencing (scRNA-Seq) data after perturbations.
• To address the challenge of accurately forecasting cellular responses computationally.

Main Methods:

• Proposed CoupleVAE, a deep learning architecture using coupled variational autoencoders.
• CoupleVAE employs two encoders to extract latent features, a coupler for latent space translation, and two decoders for data generation.
• The method enables intricate state transformations within the latent space.

Main Results:

• CoupleVAE demonstrated superior performance in predicting scRNA-Seq data for perturbed cells across three real-world datasets (infection, stimulation, cross-species).
• The model surpassed existing comparative methods in predictive accuracy.
• Validation on diverse datasets highlights the robustness and effectiveness of CoupleVAE.

Conclusions:

• CoupleVAE offers a powerful computational approach for predicting single-cell perturbation responses.
• This method can significantly reduce the cost and time associated with experimental analysis.
• CoupleVAE advances the field of computational biology by providing accurate scRNA-Seq data prediction.