Algorithm for reaction classification

Area of Science:

• Computational chemistry
• Cheminformatics
• Chemical reaction analysis

Background:

• Reaction classification is crucial in chemistry, with various algorithms developed.
• Existing datasets for evaluating these algorithms are limited and not fully representative of real-world reactions.
• Reproducible comparison of maximum common substructure (MCS) algorithms is needed.

Purpose of the Study:

• To develop and release a comprehensive and challenging dataset for evaluating reaction classification algorithms.
• To provide a standardized benchmark for comparing different MCS algorithms.
• To facilitate the development of more accurate and consistent atom mapping techniques.

Main Methods:

• Developed a novel algorithm to identify atom mappings by finding the minimum chemical distance (MCD) between reactants and products using all maximum common substructures (MCS).
• Created a diverse and challenging reaction test set drawn from extensive chemical reaction databases.
• Organized example reactions into difficulty levels to demonstrate requirements for reaction center detection.

Main Results:

• A publicly available, challenging reaction dataset has been created.
• The dataset is designed for use with InfoChem's software and other classification algorithms.
• The dataset demonstrates essential requirements for consistent reaction center detection by mapping algorithms.

Conclusions:

• The developed dataset addresses the need for a representative and reproducible benchmark in reaction classification research.
• This resource will aid in the objective comparison and advancement of MCS algorithms.
• The scientific community is invited to contribute to the dataset's ongoing development for a common standard.