AI Applications in Structural Biology

Software AI in Crystallography
1

Let’s catalogue applications of artificial intelligence and machine learning in structural biology spanning the whole ecosystem from experiment to data reduction to structure modeling and deposition. If you’ve got an application, respond here! Tell us

  1. The problem / use case
  2. The AI model that solved it
  3. Where to find out more in a publication or source repo or website
2
  1. Polychromatic diffraction data need to be corrected for wavelength and harmonic overlaps during scaling.
  2. Careless solves this problem by modeling harmonics and learning wavelength-dependent systematic errors
  3. The Careless repo, paper, review, and SBGrid talk.
3
  1. Solvent content estimation for protein crystals usually uses the Matthews Coefficient, which has approximately a 20% failure rate (especially for large numbers of molecules).
  2. ML models trained on Patterson maps do better, with the best model halving the prediction error. This model, called U-Solv, is based on the U-Net CNN architecture
  3. There is a preprint, a repo, and the model is in use as the default solvent content predictor in CCP4 Cloud
1 Like
4

from @graeme

  1. Identifying crystal hits in sitting-drop crystallization plates
  2. CHiMP which is a classifier based on the ConvNeXt architecture
  3. paper