Predicting electrochemical stability of redox-active molecules is needed to achieve durable flow batteries for electric grid storage. Here, Sure Independence Screening and Satisfying Operator (SISSO) is used to construct a binary classifier for predicting chemical stability of charged organic molecules. SISSO is an artificial intelligence (AI) expert that derives and tests heuristic algebraic formulas to rationalize data trends. Two unrelated families of redox-active molecules were classified by the same empirical parameter, which is a nonlinear combination of molecular descriptors found by SISSO without human guidance.
Significance and Impact
Experimental characterization of electrochemical stability is challenging and time consuming. In silico prediction can accelerate materials discovery, but each family of redox-active molecules needs to be studied separately. Can there be a cross-platform indicator that applies to numerous unrelated families? Most chemists would find that unlikely, yet a non-intuitive equation discovered by SISSO confidenty classifies two disparate redoxmer families. Statistical tests suggest that this AI-found formula is not a data fluke, yet the reasons for its unexpected success remain elusive.
- Positively charged pyridinium anolytes and neutral dimethoxybenzene catholytes were studied
- SISSO uses computed molecular properties to create a feature space to classify and predict electrochemical stability of charged molecules