Completing and balancing database excerpted chemical reactions with a hybrid mechanistic - machine learning approach

Abstract

Computer Aided Synthesis Planning (CASP) development of reaction routes requires understanding of complete reaction structures. However, most reactions in the current databases are missing reaction co-participants. Although reaction prediction and atom mapping tools can predict major reaction participants and trace atom rearrangements in reactions, they fail to identify the missing molecules to complete reactions. This is because these approaches are data-driven models trained on the current reaction databases which comprise of incomplete reactions. In this work, a workflow was developed to tackle the reaction completion challenge. This includes a heuristic-based method to identify the balanced reactions from reaction databases and complete some imbalanced reactions by adding candidate molecules. A machine learning masked language model (MLM) was trained to learn from reaction SMILES sentences of these completed reactions. The model predicted missing molecules for the incomplete reactions; a workflow analogous to predicting missing words in sentences. The model is promising for prediction of small and middle size missing molecules in incomplete reaction records. The workflow combining both the heuristic and the machine learning methods completed more than half of the entire reaction space.

Reaction Rebalancing: A Novel Approach to Curating Reaction Databases

“…In addition, we evaluated the computational efficiency of our methods, observing an average processing time of 46 seconds per 1000 reactions on an average workstation where one-third of the reactions were solved by MCS. In our comparative analysis, our method surpassed the current state-of-the-art, ChemMLM [29], demonstrating superior performance in both success rate and accuracy. The reported outcomes for ChemMLM showed a success rate fluctuating between 4.1% to 42.7% on the USPTO dataset.…”

“…CGRTools offers a rule-based method for rebalancing reactions by adding small molecules, which however has limited success in achieving perfect balance [28]. A hybrid workflow [29] combines ChemBalancer's heuristic methods and ChemMLM's machine learning to enhance molecule prediction. While ChemBalancer focuses on reaction completion, lacking precise accuracy metrics, ChemMLM shows promise with small molecules but struggles with complex structures [29].…”

“…A hybrid workflow [29] combines ChemBalancer's heuristic methods and ChemMLM's machine learning to enhance molecule prediction. While ChemBalancer focuses on reaction completion, lacking precise accuracy metrics, ChemMLM shows promise with small molecules but struggles with complex structures [29].…”