Deep Learning Applied to Ligand-Based De Novo Drug Design.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• The emergence of deep generative models offers a powerful new paradigm for accelerating drug discovery.
• De novo drug design aims to create novel molecular structures with desired properties.
• Ligand-based methods are crucial for identifying compounds that interact with biological targets.

Purpose of the Study:

• To provide an updated overview of de novo design approaches utilizing artificial intelligence (AI).
• To focus specifically on ligand-based deep generative models in drug discovery.
• To review the literature on AI-driven de novo design from 2017-2020.

Main Methods:

• Review of historical de novo design techniques predating AI.
• Description of common neural network architectures for ligand-based de novo design.
• Compilation of over 100 deep generative models reported in recent literature (2017-2020).

Main Results:

• Identification and categorization of numerous deep generative models for de novo drug design.
• Analysis of studies where AI-generated compounds were synthesized and biologically evaluated.
• Demonstration of the practical application of deep generative models in drug discovery.

Conclusions:

• Deep generative models, particularly ligand-based AI approaches, represent a significant advancement in de novo drug design.
• Synthesized and tested compounds validate the potential of these AI methods in identifying viable drug candidates.
• Future directions include further refinement and application of deep generative models for optimized drug discovery pipelines.