BDE

BDE predicts bond dissociation enthalpies (BDEs) for organic molecules using a graph neural network to provide near-chemical-accuracy BDE estimates for studies of reactivity, metabolism, and combustion.

Key Features:

• Machine Learning Integration: ALFABET is a graph neural network trained on 290,664 BDE values derived from automated density functional theory (DFT) calculations at the M06-2X/def2-TZVP level for 42,577 small organic molecules.
• High Accuracy: The model achieves a mean absolute error of 0.58 kcal/mol on unseen molecules.
• Rapid Computation: The method produces BDE predictions in sub-second time scales per query.

Scientific Applications:

• Chemical Reactivity Analysis: Predicting BDEs to assess bond strengths and inform reaction pathway analysis in synthetic chemistry.
• Metabolic Pathway Prediction: Identifying major sites of hydrogen abstraction in drug-like molecules to support metabolism prediction.
• Soot Formation Studies: Determining dominant molecular fragmentation pathways relevant to combustion and soot formation.

Methodology:

Graph neural network (ALFABET) trained on 290,664 BDE values obtained from automated DFT calculations at the M06-2X/def2-TZVP level for 42,577 small organic molecules.