This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Membrane Bioinformatics SS091
V9 – orientation of TM helices
- Modelling 3D structures of helical TM bundlesPark, Staritzbichler, Elsner & Helms, Proteins (2004), Park & Helms, Proteins (2006)
Aim: construct structural model for a bundle of ideal transmembrane
helices.
(1) Construct 12 good geometries for every helix pair AB, BC, CD, DE, EF, FG
(2) overlay ABCDEFG
„thin out“ solution space containing ca. 126 models
(a) remove „solutions“ where helices collide with eachother
(b) delete non-compact „solutions“
(3) score remaining 106 solutions by sequence conservation
(4) cluster 500 best solutions in 8 models
(5) rigid-body refinement, select 5 models with best sequence conservation.
Ab initio structure prediction of TM bundles
Membrane Bioinformatics SS09
25
Rigid-body refinement
Membrane Bioinformatics SS09
26
dark: Model
light: X-ray structure
Additional input:
known connectivity of the
helices A-B-C-D-E-F-G.
Otherwise, the search
space would have been
too large.
Compare best models with X-ray structures
HalorhodopsinBacteriorhodopsin Sensory Rhodopsin
Rhodopsin NtpK
Membrane Bioinformatics SS09
27
Comparing the best models with X-ray structures
Membrane Bioinformatics SS09
28
These are our 4 best
non-native models of bR.
Because contact between
A and E was not imposed,
very different topologies
were obtained.
In 2006, our methods
could not distinguish
between these models.
but they could serve as
input for further
experiments.
Can one select the best model?
Membrane Bioinformatics SS09
29
“Success case”: True de novo model of 4-helix bundle
Membrane Bioinformatics SS09
30
Predicting lipid-exposure
Membrane Bioinformatics SS09
31
Predicting lipid-exposure
Aim: derive optimal scale to predict exposure of residues
to hydrophobic part of lipid bilayer.
Scale should optimally correlate with SASA minimize quadratical error.
Y: SASA values of the training set (N = 2901 residue positions)
X: profile of residue frequencies from multiple sequence alignment ( N 21 matrix)
: wanted propensity scale for 20 amino acids + 1 intercept value (21)
Solution for minimization task
Membrane Bioinformatics SS09
32
What does MO scale capture?
Membrane Bioinformatics SS09
33
Improved prediction of exposure by statistical learning
Prediction method Prediction accuracy [%]
Beuming & Weinstein 68.7
TMX 78.7
Yuan ... Teasdale 71.1
Beuming & Weinstein(2004) method
Membrane Bioinformatics SS09
34
Improved method by statistical learning
The theory of Support Vector Classifiers evolves from a simpler case of optimal
separating hyperplanes that, while separating two separable classes, maximize
the distance between a separating hyperplane and the closest point from either
class.
A: The two classes can be fully separable by a hyperplane, and the optimal separating hyperplane can be obtained by solving Eq. 9. B: It is not possible to separate the two classes with a hyperplane, and the optimal hyperplane can be obtained by solving Eq. 17.