The phase-seeding method for solving non-centrosymmetric crystal structures: a challenge for artificial intelligence.

Area of Science:

• Crystallography
• Artificial Intelligence
• Computational Chemistry

Background:

• Determining crystal structures requires solving the phase problem, where structure factor phases are experimentally inaccessible.
• Traditional methods like direct methods and Patterson approach have limitations with large or low-resolution data.
• Current AI methods are successful for centrosymmetric structures but not non-centrosymmetric ones.

Purpose of the Study:

• To propose a new AI-integrated phasing method for solving crystal structures, applicable to both centrosymmetric and non-centrosymmetric cases.
• To reduce the complexity of the phase problem from continuous regression to multi-class classification.
• To enable AI-based structure solution for a wider range of crystal structures.

Main Methods:

• Discretizing continuous phase values for non-centrosymmetric structures into a few distinct values (phase seeding).
• Transforming the phase problem into a multi-class classification task suitable for deep learning.
• Utilizing a smaller training dataset due to discretization, reducing computational complexity.

Main Results:

• The proposed phase-seeding method effectively solves small, medium, and large crystal structures.
• Feasibility demonstrated with minimal phase seeds (3-4 values) and 10-30% seed symmetry-independent reflections.
• Successful application to non-centrosymmetric structures, expanding AI capabilities.

Conclusions:

• The phase-seeding method offers a potential breakthrough for ab initio crystal structure solution using AI.
• Combines AI classification with classical phasing, applicable to structures of any complexity or symmetry.
• Significantly enhances the scope and efficiency of AI in solving the phase problem.