AetherCell: Pre-trained Weights

This repository contains the official pre-trained weights (model checkpoints) for AetherCell, a generative foundation model for virtual cell perturbation and in vivo drug discovery.

📄 Project Links

• Code Repository: GitHub - AetherCell (Primary entry point for usage)
• Training Data: Zenodo Record
• Paper: (Coming Soon)

💡 Model Description

AetherCell is a deep generative foundation model that unifies context-rich clinical RNA-seq with perturbation-dense L1000 assays. It utilizes a Satellite-Backbone VAE architecture to:

1. Construct a unified transcriptomic manifold from 519,609 RNA-seq samples.
2. Predict mechanism-specific cellular responses to chemical and genetic perturbations.
3. Support downstream tasks like drug sensitivity prediction (IC50) and virtual screening.

📂 Files Included

• Backbone VAE: Weights for the global RNA-seq manifold.
• Satellite VAE: Weights for the L1000 platform anchoring.
• Generative Modules: Conditioned perturbation predictors.

📚 Citation

If you use this model in your research, please cite our work: (Pending publication details)

For questions or issues, please open an issue on our GitHub Repository.