How do quantum chemical descriptors shape hydrogen atom abstraction reactivity in cupric-superoxo species? A combined DFT and machine learning perspective†‡
Abstract
Oxygen activation, a crucial function performed by enzymes, prompts the synthesis of biomimetic models utilised to investigate structure–activity relationships, with a particular focus on metal-superoxo species resulting from O2 interaction with the metal centre. Among others, cupric-superoxo species have been extensively studied, showcasing diverse examples and potent catalytic capabilities. While quantum chemical calculations have helped in understanding the mechanistic aspect of their reactivity, recent advances in machine learning (ML) tools have substantiated this further and offered potent predictive power. The development of machine learning tools and associated quantum descriptors for open-shell paramagnetic catalysts is rarely pursued due to the complexity involved. However, if achieved, it has the potential to fundamentally change the existing paradigm in catalytic design and development. In making this connection, a detailed hydrogen atom transfer (HAT) reaction instigated by [(TMPA)Cu(II)–O2˙−] species and its analogues gains relevance as they offer a unique set of diverse reactivity pathways among structurally similar cupric-superoxo species. In this study, we embark on a comprehensive exploration of reactivity mechanisms employing the DFT method (B3LYP/TZVP) with five distinct catalysts and three varied substrates, resulting in combinations that lead to fifteen different reactions for the HAT reaction. The reactivity of cupric-superoxo species was found to be correlated not only with the rate-limiting HAT barrier but also with the competitive dimerization barrier. Our comprehensive analysis of mechanisms offered a rationale for the experimentally observed reactivity and the setting of goals for developing suitable ML models. In making this connection, we have arrived at fifteen quantum chemical descriptors, including exchange interaction (J), sterics, hydrogen bonding, and various thermodynamic parameters derived from DFT calculations. Our multivariate linear regression (MLR) model accurately predicts catalytic reactivity towards HAT using these quantum chemical descriptors based simply on ground state geometry. The H-bonding interactions, along with the free energy of the HAT/PT/ET reaction (ΔGPCET/ΔGPT/ΔGET), were found to yield excellent results for accuracy (R2 = 0.90), setting a stage to study multinuclear paramagnetic catalysts. For the first time, this study provides valuable insights not only into the reactivity of metalloenzymes but also offers design clues to enhance the reactivity of transient species using the ML approach.