Incorporating targeted protein structure in deep learning methods for molecule generation in computational drug...

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Artificial intelligence

Background:

• Traditional drug discovery faces challenges with high costs, low productivity, and frequent compound failures due to poor efficacy or off-target binding.
• Structure-based approaches integrate protein target information early in molecule design to mitigate late-stage failures.

Purpose of the Study:

• To review current deep learning (DL) methods for structure-based drug discovery.
• To explore how DL models utilize protein structural information for designing molecules with improved binding potential.
• To suggest future research directions in this field.

Main Methods:

• Review of existing literature on deep learning applications in structure-based drug discovery.
• Analysis of various methods for encoding and utilizing protein structural data.
• Discussion of co-folding models that predict protein and ligand structures simultaneously.

Main Results:

• Deep learning methods offer promising strategies for structure-based drug design.
• Incorporating structural information via DL can enhance molecular binding potential.
• These methods aim to maintain chemical and physical plausibility of designed molecules.

Conclusions:

• Deep learning significantly enhances structure-based drug discovery by leveraging protein structural data.
• Future directions include refining DL models for more accurate predictions and broader applications.
• This approach holds the potential to increase efficiency and success in developing novel therapeutics.