Entangled Conditional Adversarial Autoencoder for de Novo Drug Discovery

Area of Science:

• Medicinal Chemistry
• Computational Biology
• Artificial Intelligence

Background:

• Traditional drug discovery is time-consuming, often taking months.
• Machine learning and generative models offer accelerated approaches to identify novel drug candidates.
• Generative models can design molecules with specific desired properties.

Purpose of the Study:

• To propose a novel generative artificial intelligence architecture for accelerated drug discovery.
• To design molecules with specific properties, including therapeutic activity, solubility, and synthesizability.
• To generate a novel inhibitor for Janus kinase 3, a target relevant to autoimmune diseases.

Main Methods:

• Development of an entangled conditional adversarial autoencoder (ECAAE) architecture.
• Application of the ECAAE model to generate molecular structures with targeted properties.
• In vitro testing of the generated Janus kinase 3 inhibitor.

Main Results:

• The ECAAE model successfully generated novel molecular structures based on specified properties.
• A novel inhibitor targeting Janus kinase 3 was discovered using the proposed model.
• The discovered molecule demonstrated significant activity and selectivity in in vitro assays.

Conclusions:

• The proposed entangled conditional adversarial autoencoder accelerates the identification of novel drug candidates.
• This AI-driven approach can effectively generate potent and selective inhibitors for therapeutic targets.
• The generated Janus kinase 3 inhibitor holds potential for treating conditions like rheumatoid arthritis, psoriasis, and vitiligo.