Atom-to-atom Mapping: A Benchmarking Study of Popular Mapping Algorithms and Consensus Strategies

Area of Science:

• Cheminformatics
• Computational Chemistry
• Data Science

Background:

• Atom-to-Atom Mapping (AAM) is crucial for reaction analysis.
• Several AAM tools exist, employing diverse algorithms.
• Benchmarking and optimization are needed for reliable AAM.

Purpose of the Study:

• To compare the performance of popular AAM tools.
• To optimize an open-source AAM tool and develop an automated error correction algorithm.
• To apply the best performing AAM strategy to a large reaction database.

Main Methods:

• Comparative analysis of ChemAxon, Indigo, RDTool, NameRXN, and RXNMapper.
• Optimization of the RDTool program and development of a consensus mapping strategy.
• Utilized the Condensed Graph of Reaction approach for chemical distance calculation and created the "AAM fixer" algorithm.
• Benchmarking performed on a curated Golden dataset of 1851 reactions.

Main Results:

• RXNMapper demonstrated superior performance among the evaluated AAM tools.
• The developed "AAM fixer" algorithm effectively corrected erroneous mappings.
• The optimized RDTool and consensus strategies showed improved results.
• The best performing AAM approach was successfully applied to map the USPTO database.

Conclusions:

• RXNMapper, in conjunction with the "AAM fixer", offers a highly accurate solution for reaction mapping.
• The optimized RDTool and developed algorithms provide valuable resources for the cheminformatics community.
• The study provides a benchmark dataset and optimized tools for advancing AAM research.