Reaction rebalancing: a novel approach to curating reaction databases

Area of Science:

• Computational chemistry and biochemistry
• Cheminformatics
• Bioinformatics

Background:

• Reaction databases are crucial for computational chemistry and biochemistry.
• Incomplete reaction data, with missing reactants/products, limits their utility.
• Accurate and complete reaction datasets are essential for applications like Computer-aided Synthesis Planning (CASP) and metabolic network analysis.

Purpose of the Study:

• To address the urgent need for curation and correction of incomplete chemical reaction entries.
• To develop a computational framework for rebalancing chemical reactions and enhancing dataset completeness.
• To provide an open-source solution for improving the accuracy of chemical reaction databases.

Main Methods:

• The SynRBL framework employs a dual strategy for reaction rebalancing.
• A rule-based method using atomic symbols and counts predicts missing non-carbon compounds.
• A Maximum Common Subgraph (MCS)-based technique aligns carbon compounds to infer missing entities.

Main Results:

• The rule-based method achieved over 99% accuracy for non-carbon compounds.
• MCS-based accuracy ranged from 81.19% to 99.33% for carbon compounds.
• The overall framework demonstrated high efficacy with success rates from 89.83% to 99.75% and accuracy from 90.85% to 99.05%.

Conclusions:

• The SynRBL framework offers a novel and accurate solution for recalibrating chemical reactions.
• It significantly enhances reaction completeness and provides groundbreaking accuracy in reaction rebalancing.
• This work paves the way for future improvements in atom-atom mapping and automated synthesis planning.