scGen

scGen predicts single-cell transcriptional responses to genetic and environmental perturbations by modeling high-dimensional single-cell gene expression for in silico screening across cell types, studies, and species.

Key Features:

• Generative Modeling: Employs variational autoencoders combined with latent space vector arithmetic to capture complex patterns in gene expression.
• Cross-Study and Cross-Species Generalization: Generalizes predictions across different cell types, independent studies, and multiple species.
• Accurate Modeling of Perturbation and Infection Responses: Distinguishes responding and non-responding genes and cells and captures cell-type– and species-specific response features.
• Implementation: Implemented in TensorFlow and distributed with Python scripts and Jupyter Notebooks for reproducing figures and results.

Scientific Applications:

• Experimental Design: Enables in silico screening of perturbation responses to inform experimental planning in disease research and drug treatment development.
• Integration with Large-Scale Data Atlases: Screens potential perturbations against atlases of healthy organ states to aid investigation of disease mechanisms and therapeutic interventions.

Methodology:

Uses variational autoencoders to learn latent representations of high-dimensional single-cell gene expression and applies latent space vector arithmetic to predict perturbation responses.