Explore drug-like space with deep generative models.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Drug discovery requires identifying novel molecules with specific properties to target diseases.
• Exploring the vast chemical space for potential drug candidates is a significant challenge.
• Current molecular libraries represent only a fraction of the total possible drug-like chemical space.

Purpose of the Study:

• To develop and evaluate a deep molecular generative model for efficient exploration of drug-like chemical space.
• To compare the generative model's performance against existing molecular generative models.
• To demonstrate the model's potential in designing specific drug inhibitors, such as for MDM2-p53 interactions.

Main Methods:

• Construction of a drug-like dataset.
• Generative design of molecules using a Conditional Randomized Transformer with MACCS fingerprints.
• Comparison with previously published molecular generative models.
• Utilizing quantitative estimation of drug-likeness (QED) and quantitative estimate of protein-protein interaction targeting drug-likeness (QEPPI).

Main Results:

• The deep molecular generative model successfully explored a wider drug-like chemical space compared to existing models.
• Generated molecules exhibited overlap with the chemical space of known drugs.
• The combined QED and QEPPI approach effectively captured a larger portion of the drug-like space.
• The model showed potential for designing inhibitors targeting the MDM2-p53 protein-protein interaction.

Conclusions:

• Deep molecular generative models offer a powerful alternative for molecular design and drug discovery.
• These models enable guided exploration of the chemical space for novel drug candidates.
• The developed approach enhances the efficiency and scope of identifying molecules with desired therapeutic properties.