Towards causal inference in materials informatics: methods, applications and opportunities

Machine learning (ML) has transformed materials screening, yet it optimises for statistical association rather than causation. This communication introduces causal inference frameworks, including structural causal models (SCMs), directed acyclic graphs (DAGs), and the potential-outcomes framework, as a principled complement to ML in materials informatics. We clarify how these tools align with the process–structure–property–performance (PSPP) paradigm, make explicit the assumptions required for valid causal inference in materials datasets, contrast causal inference with explainable AI, and provide a concrete worked example of causal modelling in laser powder-bed fusion, together with pseudo-code for constraint-based causal discovery. We review emerging applications spanning nanocatalysis, additive manufacturing, welding, composites, coatings, and degradation modelling, outline methodological strategies integrating causal discovery with Bayesian uncertainty quantification and graph neural networks (GNNs), and close with a prioritised research roadmap for autonomous causal materials discovery.