Prediction & validation
Chemical shift prediction
Predict chemical shifts from a structure — to validate a model, to check an assignment, or to seed one.
Predicted shifts are useful in both directions. Against a finished structure they are a validation: systematic disagreement over a region means the model is wrong there, and it is often the only independent evidence you have. Against an assignment in progress they are a prior, narrowing which spin system can belong to which residue.
LEGOLAS is a machine-learned predictor, fast and strong where its training data represents your protein well and weaker where it does not. The useful signal is disagreement: a residue whose predicted shift departs sharply from the measured one is worth your attention, whether the model is wrong there or the structure is.
What it needs
- Structure in PDB or mmCIF
- Optionally, measured shifts to compare
What you get
- Predicted shifts per atom
- Deviations from measured values where supplied
What does the work
LEGOLAS
Open source, redistributed under its licenceMITMachine-learned shift prediction
Related in Prediction & validation
- Backbone dihedral predictionPredict backbone φ and ψ angles from chemical shifts, with per-residue reliability.
- Secondary structure predictionDetermine secondary structure from chemical shifts or from sequence alone.
- Structure validation against shiftsScore how well a structure explains the chemical shifts measured for it — a check that geometry-only validation cannot make.