1. Common Coot
(Fulica atra)

2. Validation in Coot

3. Validation with Coot Bernhard Lohkamp Karolinska Institutet
Mar 2011 Shanghai
Validation with Coot
Bernhard Lohkamp
(Paul Emsley)
(University of Oxford)
Karolinska Institutet

4. Feature Integration Refinement Validation External Internal
e.g. REFMAC
Internal
External
e.g. MolProbity

7. What is Validation?
Comparison of various aspects of the model with pre-conceived notions of “good quality”
Includes unrestrained and restrained criteria
Many aspects of validation overlap with refinement and model-building
Global and local indicators

8. Why Validate? Model-building is error-prone
(although automated methods seem to do better)
Map interpretation is subjective
Someone else did the model-building
The model was built several years ago
and the notion of “good quality” has changed
Deposition requires validation

9. A “good” model Makes statistical sense
The reciprocal space representation agrees tolerably well with the observations (R-factor)
No meaningful difference map peaks (no under- or over-fitting)

10. A “good” model Makes statistical sense
The reciprocal space representation agrees tolerably well with the observations (R-factor)
No meaningful difference map peaks (no under- or over-fitting)
Makes chemical sense
Model geometry is consistent with the restraints
Ramachandran Plot has less than 1% outliers
A good clashscore

11. A “good” model Makes statistical sense
The reciprocal space representation agrees tolerably well with the observations (R-factor)
No meaningful difference map peaks (no under- or over-fitting)
Makes chemical sense
Model geometry is consistent with the restraints
Ramachandran Plot has less than 1% outliers
A good clashscore
Make physical sense
Crystal packing and contacts

12. A “good” model Makes statistical sense
The reciprocal space representation agrees tolerably well with the observations (R-factor)
No meaningful difference map peaks (no under- or over-fitting)
Makes chemical sense
Model geometry is consistent with the restraints
Ramachandran Plot has less than 1% outliers
A good clashscore
Make physical sense
Crystal packing and contacts
Makes biological sense
Residues in chemically sensible environment
Is consistent (on the whole) with external biochemistry observations (active site residues)

13. Quick Bayes Bayes Eq: Pr(model|data)  Pr(data|model) * Pr(model)
Pr(data|model) is also called the model likelihood, LM(model|data)

14. Validation Tools - Pr(model)
Ramachandran Plot
Kleywegt Plot (NCS differences)
Geometry Analysis
Peptide  Analysis
Temperature Factor Analysis
Rotamer Analysis
Clashes

15. Validation Tools - Pr(data|model)
Density Fit Analysis
Difference Map Peaks
Variance analysis at water positions
Unmodelled blobs

17. Find difference map peaks

18. Check/Delete Waters

19. Difference Map Sampling
Detects “Anomalous” Waters

22. Find unmodelled blobs

23. Pr(model)... Ramachandran Plot Kleywegt Plot (NCS differences)
Geometry Analysis
Peptide  Analysis
Temperature Factor Analysis
Rotamer Analysis
Sequence analysis
Clashes

25. Ramachandran Plot for residues with CB

26. Ramachandran Plot for GLY

27. Ramachandran Plot for PRO

28. Top500-based distribution

29. Not all outliers are bad

30. [*] Named by George Sheldrick
Kleywegt Plots[*]
[*] Named by George Sheldrick

32. Peptide  Needed to check the planarity of the peptide link
At low resolutions it is possible to give the protein lots of (too much) freedom to optimize the fit to the density
Can accidentally create CIS peptides
When discovered they are easily reconverted using the CIS<- >TRANS peptide tool
Less accidents happen when peptide plane restraints are applied

33. B-factor variance

34. Rotamers
Side-chains have certain preferred combinations of torsions round their rotatable bonds
An analysis (batched around the staggered conformations) will give rotamer occurrence

35. Torsion-based Validation

37. Not all outliers are bad
Correct rotamer(?!)

38. Sequence validation
Assign sequence (Extensions  Dock sequence  Associate sequence)
Sequence alignment of model to given sequence (Validate Alignment vs PIR)

39. More Validation Pr(model)
Coot has interface to Molprobity
(Molprobity is the widely regarded as the best model validation suite)
Uses identical Ramachandran plot
Uses identical Rotamer library
Coot reads probe dots directly

41. Molprobity’s Reduce & Probe

43. Other Programs Molprobity Suite molprobity.biochem.duke.edu WHATCHECK
VERIFY-3D

44. Handling NCS...

45. What is Non-Crystallographic Symmetry?
2 or more copies of a molecule in the unit cell not related by crystallographic symmetry
Crystallographic copies of molecules are (of course) treated as if they were exactly the same across the unit cell – and indeed across the whole crystal
Non-crystallographically related molecules provide different representations of the same molecule
This can be useful for model-building
But difficult to use in practice

46. Handling NCS What are the Problems? Strict NCS:
NCS should appear like crystallographic symmetry does [exact copies]
Non-Strict NCS:
Molecules are different
How to cope with differences, but minimize unnecessary rebuilding?

47. Handling NCS Typical Scenario:
I have done an LSQ overlap of my NCS-related molecules and from the graph, have seen significant deviations in the positions of some side-chains.
Why are they different?

48. [*] Named by George Sheldrick
…or Kleywegt Plots[*]
[*] Named by George Sheldrick

49. ...or NCS Differences graph

53. NCS Model-modification Tools
Automatic detection of NCS
And their operators
Copy Master NCS molecule to others
Applies NCS transformation
Copy NCS Master residue-range
Change NCS Master chain
NCS Skipping

54. Acknowledgements Paul Emsley Kevin Cowtan Eleanor Dodson Keith Wilson or
Google: Coot
or for WinCoot
Paul Emsley
Kevin Cowtan
Eleanor Dodson
Keith Wilson
Libraries, dictionaries
Alexei Vagin, Eugene Krissinel, Stuart McNicholas
Dunbrack, Richardsons
Coot Builders and Testers
William Scott, Ezra Peisach
York YSBL, Dundee, Glasgow (early adopters)
Coot Mailing List subscribers