ChemGenerator: a web server for generating potential ligands for specific targets.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Drug discovery is limited by experimental exploration, necessitating in silico methods to expand chemical databases and identify drug-target interactions.
• Developing novel, biologically active molecules remains a critical challenge in pharmaceutical research.

Purpose of the Study:

• To introduce ChemGenerator, a web server designed for generating novel active molecules tailored to specific biological targets.
• To provide an efficient in silico alternative for the initial stages of drug discovery, complementing traditional experimental methods.

Main Methods:

• ChemGenerator employs a Recurrent Neural Network (RNN) with Long Short-Term Memory (LSTM) autoencoder architecture.
• The model is trained on 7 million molecular Simplified Molecular-Input Line-Entry System (SMILES) strings and utilizes transfer learning for target-guided generation.
• Performance is evaluated using loss metrics and demonstrated on the Epidermal Growth Factor Receptor target.

Main Results:

• ChemGenerator achieved a significantly lower loss (<0.01) compared to a reference model (0.2–0.4).
• The system demonstrated effective performance in generating molecules for the Epidermal Growth Factor Receptor.
• The ChemGenerator web server is publicly accessible at http://smiles.tcmobile.org.

Conclusions:

• ChemGenerator offers an efficient in silico method for virtual screening in drug discovery.
• This approach accelerates the identification of potential drug candidates, reducing reliance on extensive experimental screening.
• The tool represents a valuable advancement for expanding chemical databases and profiling drug-target linkages.