A transferable data driven framework for electrolyte discovery

Abstract Widespread electrification in transportation and grid storage demands rapid development in batteries. In particular, electrolytes that deliver fast ion transport, wide electrochemical stability windows, durable electrode interfaces, safety under abuse, and scalable manufacturing. Machine learning (ML) combined with chemical experimentation can accelerate electrolyte development. Unsupervised clustering, supervised learning, and Bayesian optimization recover structure-property relations and navigate multi-objective trade-offs across diverse battery chemistries, even under small-data regimes. However, the dominant barrier is data quality and comparability: dispersed reports, missing or inconsistent metadata, and nonstandard test conditions undermine benchmarking and cross-system transfer, weakening the predictive value of ML models. In this perspective, a data-centric framework highlights (i) minimum information standards for electrolyte reporting, (ii) Findability, Accessibility, Interoperability, and Reuse (FAIR) community databases, (iii) Natural Language Processing (NLP) in literature-scale conversion of results, and (iv) shared benchmark tasks with reference cells for reproducible comparison. A unified feature space that respects system-specific constraints is proposed to enable transfer across chemistries. Implemented together, these elements establish a reproducible pathway toward autonomous, closed-loop electrolyte discovery.