Xtricorder: a Likelihood-Enhanced Self-Rotation Function and Application to a Machine Learning-Enhanced Matthews Prediction of Asymmetric Unit Copy Number.

Analysis of crystallographic diffraction data after collection and integration but before phasing gives the crystallographer a `first-look' assessment of data quality and flags potential challenges in subsequent structure determination. We here report the development of Xtricorder, a `first-look' application specifically targeted at likelihood-based phasing. Xtricorder incorporates the full array of analyses previously available in the Phaser codebase, with some enhancements and updates, in a more streamlined and accessible implementation. In addition, Xtricorder offers a likelihood-enhanced self-rotation function. A novel graphical representation of the self-rotation function, the `composite-section diagram', presents the results for user inspection and has the added advantage that, in an adapted form, it is appropriate for training a convolutional neural network to enhance the standard Matthews analysis and double the accuracy of asymmetric unit copy-number prediction. We investigate the usefulness of the likelihood-enhanced self-rotation function in `first-look' analyses, exploring the circumstances under which the self-rotation function results are useful, and discuss the application to AI-generated structure prediction.