How can we tell how good our model is FINAL

Views:
 
Category: Entertainment
     
 

Presentation Description

No description available.

Comments

Presentation Transcript

Slide1:

can we tell how good our model is? - How-

Slide2:

To find DIFFERENCES between crystallographic ‘model’ and real structure of the protein. It is only REPRESENTATION of the real structure. There are many sources of ERROR during the structure solution process . In order to detect INCONSISTENCIES and mistakes. Why do we test our model?

R = (Fo- Fc ) (Fo) :

5 factors that test our protein model are… B factor and occupancy R factor R free R.M.S.D Ramachandran plot R =  (  F o - F c  )  ( F o )

Slide4:

FACTOR #1: B factor and occupancy Termini Flexible loops S olvent-exposed regions Long regions H igh B factors are found in flexible regions How certain are we of atoms positions? Is there potential flexibility in the structure? Do atoms have any static or dynamic disorder?

Slide5:

Therefore … . Low B factor = GOOD! High B factor = uncertainty low overall B factor = high resolution

Slide6:

Is refinement valid? FACTOR #2: R factor How close is our model to the data? How close is our model to reality? Does our protein model have expected chemical characteristics?

R = (Fo- Fc ) where Fo = F observed (Fo) Fc = F calculated :

Calculating R factor …. OR in words… sum of differences between observed and calculated structure factors sum of observed structure factors R =  (  F o - F c ) where F o = F observed  ( F o ) F c = F calculated

Therefore…:

Therefore … The R factor indicates the QUALITY of model Shows the models DEVIATION from reality Low R factor = GOOD! High R factor = BAD! Effectively MEASURES models ERRORS

Slide9:

FACTOR #3: R free Has the noise been wrongly interpreted? Has the model been over fitted to the data? Therefore it’s a cross-validation parameter These reflections are excluded from refinement And is similar to R-factor ~ 1000 randomly selected subset of reflections So are statistically independent

Slide10:

R free checks quality of model – BEST INDICATION! Same as R factor but calculated for small percentage of reflections. Unbiased measure of how similar our model is to the data. Therefore … . High value may indicate over-fitting or serious model defect Low R free = GOOD! High R free = BAD!

Slide11:

Is our structure valid chemistry? FACTOR #4: RMSD How much do our model’s bond angles and lengths differ from typical parameters? Root-mean-square deviation of atomic positions – measures the average distance between the atoms of superimposed proteins

Slide12:

Therefore … R.m.s.d . values indicate how close the model’s bond angles and lengths are to those expected for small molecules Dominated by the amplitude of errors – affected by flexible and poorly defined regions RMSD bond length <0.02Å = GOOD ! RMSD bond length >0.02Å= BAD! RMSD Bond angle <2°= GOOD! RMSD Bond angle >2°= BAD! Doesn’t reflect models accuracy

Slide13:

Are the pairs of phi/psi angles of the polypeptide backbone mapped as expected? Ramachandran plot FACTOR #5: Plot of phi Φ and psi Ψ angles for each residue

Slide14:

> 90% of the angles are found in the expected areas of the plot = GOOD! Many residues not in the expected areas of the plot= BAD! Useful because Ramachandran plot values not restrained in refinement process. Therefore …

Conclusion:

Conclusion R free !! Ramachandran plot – EXPECTED REGIONS GOOD!! B factor – LOW IS GOOD!! R factor (%) – LOW IS GOOD!! Rfree (%) – LOW IS GOOD!! RMSD – LOW IS GOOD!! Which is the best indicator of model quality?

Slide16:

References Fisch , F., 2011. The Link to X-Ray Protein Crystallography: Validation [Online]. Available from: http ://my.yetnet.ch/dergutemensch/crystallography/ validation.htm [ Accessed 15/11/16]. Good for a brief overview of the reasoning behind, and the purpose of, validation of a model. Garwood, J., 2009. A Guardian of Structural Integrity. Lab Times, 5, pp. 18- 23. Goes into detail about the pitfalls and experimental errors in determining 3D protein structures, giving information on reasoning behind and aims of model validation. Jones, T., A., and Kleywegt , G., J., 1997. Good Model-building and Refinement Practice [Online]. Uppsala university, Uppsala. Available from: http://xray.bmc.uu.se/gerard/gmrp/ gmrp.html [Accessed 15/11/16]. Goes into detail about problems that arise during refinement as part of the model building process. Brief overview of “quality control” including in the geometry of the main chain ( Ramachandran plot) and the fit of model and map (R factor). Detailed information on Rfree as a cross-validation scheme. http://reference.iucr.org/dictionary/... Detailed information about R factor as a measure of structure quality and what calculations are required to obtain it . Wlodawer , A., Minor, W., Dauter , Z., Jaskolski , M., 2008. Protein crystallography for non-crystallographers, or how to get the best (but not more) from published macromolecular structures. The FEBS Journal, 275(1), pp. 1- 21. Very good, well-explained, detailed information on the indicators of structure quality – R factor, Rfree , R.m.s.d , Ramachandran plots. Kufareva , I., Abagyan , R., 2012. Methods of protein structure comparison. Methods in Molecular Biology, 857, pp. 231-257. Detailed information on R.M.S.D., what calculations are required to obtain it, and its drawbacks. https://en.wikipedia.org/wiki/Root-mean-square_deviation_of_atomic_positions Good for basic definition and overview of RMSD. Not necessarily reliable information. Kleywegt , G.L. Quality Control and Validation. Methods in Molecular Biology, 364, pp. 255-256. Overview of validation, and quality indicators such as R factor and R amachandran plots. Detailed information on ‘weak’ and ‘strong’ quality indicators, and overview of requirements of validation methods in terms of model and experimental data . g

authorStream Live Help