De Novo Design of Molecules with Multiaction Potential from Differential Gene Expression using Variational...

Area of Science:

• Computational chemistry
• Genomics
• Drug discovery

Background:

• Drug effects on gene transcription are well-studied for clinical and research applications.
• Drug-induced transcriptional changes offer potential for de novo molecular design.
• Phenotype-matching molecular structures to gene expression signatures is an emerging concept.

Purpose of the Study:

• To develop an autoencoder-based generative model, BiCEV, for de novo molecular design.
• To leverage gene expression data for creating novel molecular structures.
• To assess the model's performance in generating valid, unique, and diverse molecules.

Main Methods:

• An autoencoder-based generative model (BiCEV) was constructed.
• The model was trained using gene expression data as input.
• Model validation involved testing on gene-knockdown profiles and synergistic drug pair signatures.

Main Results:

• BiCEV generated novel molecules with high validity (96%), uniqueness (98%), and internal diversity (0.77).
• Designed molecules demonstrated consistent high collective quality.
• Similarity analysis to functional equivalents based on shared targets yielded mixed results.

Conclusions:

• The BiCEV model shows potential as an assistive tool for drug discovery.
• Further development and in vitro validation are recommended.
• The model could support early-stage hit identification and lead optimization.