SyntaLinker: automatic fragment linking with deep conditional transformer neural networks.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Artificial Intelligence in Drug Discovery

Background:

• Fragment-based drug design (FBDD) faces challenges in generating focused compound libraries for specific drug targets.
• Traditional methods for linking molecular fragments rely on predefined empirical rules.

Purpose of the Study:

• To introduce SyntaLinker, a novel program for automated molecular fragment linking in FBDD.
• To leverage deep conditional transformer neural networks and syntactic pattern recognition for fragment linking.

Main Methods:

• SyntaLinker employs deep conditional transformer neural networks to learn implicit rules for fragment linking.
• The program recognizes syntactic patterns within SMILES notations derived from medicinal chemistry databases like ChEMBL.
• It generates molecular structures based on specified fragment pairs and constraints.

Main Results:

• SyntaLinker demonstrates an automated approach to linking molecular fragments, moving beyond predefined empirical rules.
• The system successfully learns linking strategies from existing chemical structure data.
• Case studies confirm the utility and advantages of SyntaLinker in the context of FBDD.

Conclusions:

• SyntaLinker offers a powerful, data-driven solution for fragment linking in FBDD.
• The use of deep learning and syntactic pattern recognition represents a significant advancement in automated molecular design.
• This approach enhances the efficiency and focus of compound library generation for drug discovery.