Learning to encode cellular responses to systematic perturbations with deep generative models.

Area of Science:

• Systems Biology
• Artificial Intelligence
• Genomics

Background:

• Cellular signaling networks maintain homeostasis against perturbations.
• Perturbations induce transcriptomic patterns reflecting network organization.
• Understanding signal encoding is crucial for systems biology.

Purpose of the Study:

• Investigate deep generative models (DGMs) for modeling cellular signaling systems.
• Learn representations of cellular states from transcriptomic responses to perturbations.
• Assess DGMs' capability in understanding signal encoding.

Main Methods:

• Applied variational autoencoder (VAE) and supervised vector-quantized VAE (SVQ-VAE).
• Modeled gene expression data from perturbagen treatments.
• Analyzed learned representations for relationships between perturbagens and gene targets.

Main Results:

• DGMs accurately regenerated gene expression data.
• Learned representations revealed relationships between perturbagen classes.
• Models enabled mapping between drugs and their target genes.

Conclusions:

• Deep generative models effectively learn and depict cellular signal encoding.
• Learned representations offer broad applications in systems biology.
• Demonstrates the power of AI in precision medicine and biological research.