DigFrag as a digital fragmentation method used for artificial intelligence-based drug design.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Fragment-Based Drug Design (FBDD) is crucial for drug discovery.
• Current FBDD methods use rigid rules, limiting adaptability.
• Artificial Intelligence (AI) offers new approaches to FBDD.

Purpose of the Study:

• To develop a novel digital fragmentation method (DigFrag) for FBDD.
• To compare AI-generated fragments with human-curated ones.
• To evaluate the impact of fragment source on deep generative models.

Main Methods:

• Developed DigFrag, a local molecular graph-based fragmentation technique.
• Utilized a deep generative model to assess fragments from AI and human expertise.
• Generated and analyzed compounds based on different fragment sets.

Main Results:

• DigFrag produced fragments with higher structural diversity.
• Compounds generated from DigFrag fragments were more desirable.
• AI-generated data showed better compatibility with AI models.

Conclusions:

• DigFrag offers an advanced approach to fragment library construction in FBDD.
• AI-driven methods can generate superior fragments for drug discovery.
• The MolFrag platform supports diverse molecular segmentation techniques.