De novo Molecular Design with Generative Long Short-term Memory

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Artificial intelligence in drug discovery

Background:

• Drug discovery increasingly utilizes computational models for identifying novel chemical entities.
• Generative machine learning models, particularly deep neural networks, have been adapted from natural language processing for de novo drug design.
• These models learn molecular structures and generate new entities without predefined building blocks or transformations.

Purpose of the Study:

• To review de novo molecular design approaches employing artificial intelligence, specifically deep generative models.
• To highlight the potential of long short-term memory models in hit and lead identification for medicinal chemistry.

Main Methods:

• Review of deep generative models in de novo drug design.
• Focus on implicit representation of chemical knowledge.
• Exploration of long short-term memory models' application.

Main Results:

• Deep generative models offer a powerful alternative to traditional virtual molecule construction.
• Implicit chemical knowledge representation simplifies molecular design tasks.
• Long short-term memory models show promise for hit and lead discovery.

Conclusions:

• Artificial intelligence, particularly deep generative models, significantly advances de novo drug design.
• The implicit representation of chemical knowledge is a key innovation.
• Long short-term memory models are prospective tools for accelerating hit and lead finding in medicinal chemistry.