High‐Throughput Screening of Flavan‐3‐Ol Antioxidants Through DFT ‐Guided Machine Learning
TL;DR
An integrated computational strategy that combines density functional theory with graph neural network‐based machine learning to predict three key thermodynamic descriptors governing antioxidant reactivity is introduced, confirming that aromatic hydroxylation enhances predicted reactivity and that the energetic ordering of PA supports the sequential proton loss electron transfer (SPLET) pathway as the predominant mechanism in aqueous media.
Abstract
Flavan‐3‐ols, a subclass of dietary flavonoids bearing multiple hydroxyl (OH) groups, are well‐recognized for their antioxidant potency, yet systematic exploration of their derivatives across the large combinatorial space of possible hydroxylation patterns remains a computational challenge. Here, we introduce an integrated computational strategy that combines density functional theory (DFT) with graph neural network‐based machine learning (ML) to predict three key thermodynamic descriptors governing antioxidant reactivity: ionization potential (IP), OH bond dissociation enthalpy (BDE), and proton affinity (PA). Starting from a virtual library of more than 5600 unique flavan‐3‐ol variants, an active learning workflow was employed to iteratively guide DFT calculations, culminating in a high‐quality dataset of 1235 compounds for model development. Our DFT analysis provides a quantitative basis for structure–property relationships (SPR), confirming that aromatic hydroxylation enhances predicted reactivity and that the energetic ordering of PA mol < BDE mol < IP supports the sequential proton loss electron transfer (SPLET) pathway as the predominant mechanism in aqueous media. The final ML models achieved high predictive accuracy, particularly for molecular‐wide properties representing the most reactive site (MAEs: 1.02 for IP, 0.68 for BDE mol , and 0.79 kcal/mol for PA mol ). Prospective experimental testing supports the utility of the ML models, confirming a significant lead which demonstrated a 70% increase in antioxidant activity over the (+)‐Afzelechin benchmark. This highlights the potential of the model in prioritizing promising antioxidant candidates for targeted experimental validation. In addition, the work establishes a scalable computational framework for accurately predicting fundamental reactivity descriptors, providing deep mechanistic and structural insights for accelerating the discovery of novel flavonoid‐based compounds.