GEP-DNN4Mol: automatic chemical molecular design based on deep neural networks and gene expression programming

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Molecular inverse design is crucial for discovering novel compounds.
• Current methods often lack sufficient diversity in generated molecules.
• Exploring vast chemical spaces efficiently remains a challenge.

Purpose of the Study:

• To introduce GEP-DNN4Mol, a method for generating diverse molecules with specific properties.
• To improve the exploration of chemical space in molecular design.
• To address limitations in diversity found in existing molecular generation techniques.

Main Methods:

• Utilizing Gene Expression Programming (GEP) as the core generator for molecular structures.
• Employing a Deep Neural Network (DNN) as an evaluator to guide GEP updates based on molecular features.
• Integrating both SMILES and SELFIES molecular representations for enhanced compatibility and expressiveness.

Main Results:

• GEP-DNN4Mol demonstrates superior performance compared to state-of-the-art methods.
• Achieved higher validity, novelty, and diversity in generated molecules.
• Showcased enhanced efficiency in exploring large chemical spaces.

Conclusions:

• GEP-DNN4Mol offers a powerful new approach for molecular inverse design.
• The method effectively balances molecular property optimization with diversity.
• This work advances the field of computational molecular design and discovery.