SAVI Space-combinatorial encoding of the billion-size synthetically accessible virtual inventory

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Drug Discovery

Background:

• The Synthetically Accessible Virtual Inventory (SAVI) previously generated over a billion synthetically accessible compounds using LHASA transform rules.
• Explicitly handling large molecular libraries like SAVI is computationally intensive for drug discovery.
• Existing methods faced challenges with memory and search efficiency for massive virtual compound collections.

Purpose of the Study:

• To design and implement SAVI-Space-2024, a novel computational approach to manage and explore vast virtual chemical libraries.
• To address the computational demands and memory limitations associated with large-scale virtual screening in drug discovery.
• To improve search functionalities for similarity and substructure identification within extensive molecular databases.

Main Methods:

• Development of a reaction-driven combinatorial data structure for SAVI-Space-2024.
• Encoding transformation rules as reaction SMARTS for combinatorial application.
• Utilizing Enamine Building Blocks as the foundation for molecule generation.
• Implementing efficient algorithms for similarity and substructure searches.

Main Results:

• SAVI-Space-2024 generates 7.5 billion molecules from Enamine Building Blocks.
• The new approach requires significantly less memory (1.4 GB) compared to the explicitly enumerated SAVI library (210 GB).
• Enhanced search capabilities, including fast similarity and substructure searches, are demonstrated on standard hardware.

Conclusions:

• SAVI-Space-2024 offers a computationally efficient and memory-saving solution for exploring large virtual chemical spaces.
• The reaction-driven combinatorial data structure represents a significant advancement for drug discovery applications.
• Improved search performance enables faster identification of potential drug candidates from massive molecular libraries.