Generative chemistry: drug discovery with deep learning generative models.

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Generative modeling

Background:

• Drug discovery costs are rising, necessitating innovative approaches.
• Deep learning generative models demonstrate advanced capabilities in creating novel molecular structures.
• Artificial intelligence is emerging as a transformative paradigm in the pharmaceutical industry.

Purpose of the Study:

• To review recent advancements in generative chemistry for drug discovery.
• To outline the history and evolution of AI in drug discovery.
• To provide an overview of the infrastructure supporting generative chemistry.

Main Methods:

• Review of artificial intelligence in drug discovery.
• Analysis of chemical databases, molecular representations, and cheminformatics tools.
• Detailed discussion of generative architectures: recurrent neural networks, variational autoencoders, adversarial autoencoders, and generative adversarial networks.

Main Results:

• Generative models, including RNNs, VAEs, AAEs, and GANs, are effectively utilized for de novo compound generation.
• Established infrastructure (databases, representations, tools) supports generative chemistry applications.
• The review highlights the potential of these models to expedite the drug discovery pipeline.

Conclusions:

• Deep learning generative models represent a significant advancement in de novo molecular design for drug discovery.
• The integration of AI, particularly generative chemistry, offers a promising avenue to overcome the challenges of new drug development.
• Future perspectives and challenges in generative chemistry are identified, paving the way for further innovation.