Phase seeding may provide a gateway to structure solution by deep learning.

The phase-seeding method proposed by Carrozzini et al. [(2025), Acta Cryst. A81, 188-201] introduces a strategy for integrating artificial intelligence (AI) with established ab initio phasing techniques. Rather than presenting an AI-based phasing solution itself, the authors demonstrate how traditional crystallographic methods can be significantly enhanced if provided with a small subset of approximate phase values - a `phase seed' - that could, in principle, be generated by a machine learning model. By discretizing phase values into a few angular bins, the method transforms the continuous phase problem into a classification task, thereby reducing the computational burden on AI training. This hybrid approach shows promise for improving structure solution, particularly for large and complex non-centrosymmetric crystals, and opens a pathway for future AI-assisted crystallographic workflows.