Data processing

The KDO8PS diffraction patterns collected were identified and calibrated using Cheetah⁷¹. For hit-finding, a minimum of 2 connected pixels with a count above a threshold of 500 ADU with a minimum signal-to-noise ratio of 8 was considered a peak, and an image containing at least 10 such peaks was classified as a hit. From the ~4,100,000 images collected, 37,000 were classified as hits, which corresponds to an average hit rate of 0.9%.

Approximately 46% of the identified hits could be indexed and the Bragg reflections integrated using the software package CrystFEL (version 0.8.0)⁵⁵, based on the peak locations found by Cheetah. Indexing was performed by CrystFEL’s indexamajig sequentially trying XGANDALF⁷², DirAx⁷³, MOSFLM⁷⁴ and XDS⁷⁵ requiring a cubic body-centered lattice and unit cell parameters of a = b = c = 118 Å and α=β=γ=90° (Supplementary Fig. ⁵). The indexamajig integration radii were set to 2, 3, 4 pixels and indexing solutions were checked by ensuring that they accounted for at least 50% of the observed peaks (option “check-peaks”).

The indexed patterns were further processed with ambigator⁷⁶ to resolve the indexing ambiguity inherent to the cubic space group I23 for serially collected data. The number of correlation coefficients per crystal was limited to 1000 for speed, and 10 passes of ambiguity resolution were made over all crystals. This resulted in 8315 indexing assignments being changed, corresponding to roughly 50% of the data as expected. The indexed reflections were subsequently scaled and merged using partialator⁷⁷, applying the unity model (without partiality modeling) over 3 iterations. The intensities were converted to structure factor moduli using truncate (from the CCP4 suite⁷⁸), and a fraction of 0.1 reflections were included in the generated R_free set. The L-test implemented in truncate was used to ascertain successful de-twinning of the data (Supplementary Fig. ⁶).