Molecular Generation for Desired Transcriptome Changes With Adversarial Autoencoders.

Area of Science:

• Computational Biology
• Drug Discovery
• Genomics

Background:

• Gene expression profiling is crucial for evaluating drug efficacy and toxicity.
• Predicting drug effects on cellular processes remains a challenge in pharmaceutical research.

Purpose of the Study:

• To develop a novel generative model for inferring drug molecules capable of inducing specific gene expression changes.
• To create a computational tool for designing targeted therapeutics based on desired transcriptional responses.

Main Methods:

• Proposed a Bidirectional Adversarial Autoencoder (BiAAE) model.
• Separated gene expression features into drug-related and unrelated components.
• Validated the model using the LINCS L1000 dataset and generated molecular structures in SMILES format.

Main Results:

• The BiAAE model successfully generated novel molecular structures predicted to induce specific gene expression changes.
• The model demonstrated the ability to predict gene expression differences resulting from molecular structure incubation.
• Experimental validation confirmed the model's capability in both generating and predicting drug-induced transcriptional responses.

Conclusions:

• The Bidirectional Adversarial Autoencoder offers a powerful approach for in silico drug design.
• This method can accelerate the discovery of new therapeutics by predicting molecular structures for desired gene expression outcomes.