18 Dec 2025

Attendees:

  1. Petrovic, D. McDonagh, D. Waterman, E. Krissinel

Results

  • Created this website to document progress.

  • Tried to process data from the original paper ([Clabbers et al., 2019]). There is a problem with rotation data (missing starting angle and oscillation). Problem with reading CBF header.

  • Created a global fitter function. Model built using these parameters:

    • Spot property: H, K, L, intensity, sigma, resolution

    • Dataset property: average \({\rm F}_o\), median \({\rm Fo}\)

First Second

Global prediction on training dataset.

Global prediction on new dataset

  • Created a local fitter function. Model built using these parameters

    • Spot property: H, K, L, intensity, sigma, resolution

    • Dataset property: average \({\rm F}_o\), median \({\rm Fo}\)

    • Neighbor property: min neighbor intensity, max neighbor intensity, 90 percentile neighbor intensity, IQR neighbor intensity, max neighbor resolution, min neighbor resolution, average neighbor resolution, number of neighbors, average neighbor intensity,

  • Neighbors are computed in real space (pixel distance).

First Second

Local prediction on training dataset.

Local prediction on new dataset

  • Comparing R1 factors (per image and for the whole dataset).

First Second

\(R_1\) across datasets.

\(R_1\) per image.

Discussion

  • In addition to \(R_1\), use maximum likelihood and entropy to assess the fit quality.

  • The GBT model should use spot coordinates in the inverse space (i.e. \(k_x\), \(k_y\), \(k_z\)) instead of Miller indices.

  • Use excitation error as a spot property.

  • Use distance on the Ewald sphere (instead of the real/detector space) to determine nearest neighbors of a spot.

  • Another approach might be to train CNNs on images of spots on the Ewald sphere.

  • Vary the GBT model size to get a better fit.