Gex2SGen: Designing Drug-like Molecules from Desired Gene Expression Signatures.

Area of Science:

• Computational chemistry
• Genomics
• Drug discovery

Background:

• Drug-induced gene expression profiling is crucial for understanding drug mechanisms.
• Deep learning models can explore chemical space and design targeted drug molecules.
• Accessible transcriptomic data and deep learning enable designing drugs based on gene expression signatures.

Purpose of the Study:

• To develop a deep learning model, Gex2SGen, for generating novel drug-like molecules from desired gene expression profiles.
• To enable cell-specific drug design based on transcriptomic signatures.
• To create a generalized method for designing small molecules with specific drug-like properties.

Main Methods:

• Developed Gex2SGen, a deep learning model accepting cell-specific gene expression profiles as input.
• Trained the model on open-source drug-induced transcriptomic data.
• Validated the model by comparing generated molecules against known inhibitors and anti-cancer drugs.

Main Results:

• Gex2SGen successfully generated molecules similar to known inhibitors when tested against gene-knocked-out profiles.
• The model designed novel molecules highly similar to existing anti-breast cancer drugs for a triple-negative breast cancer signature.
• The generated molecules effectively elicited the desired transcriptomic profiles.

Conclusions:

• Gex2SGen provides a generalized deep learning framework for designing small molecules based on specific gene expression signatures.
• This method facilitates drug discovery by linking gene expression profiles to novel molecule generation.
• The approach holds potential for developing targeted therapeutics by designing molecules to achieve desired cellular responses.