1. Reduce the need for human intervention in protein model building
Xpleo1 : Modeling missing protein fragments in weak or ambiguous density
Consensus Modeler : a tool to combine experimental protein models
1 Acta Cryst. (2005). D61, 2-13

2. How to Model Missing Fragments?
Disorder and Dynamics in Proteins
X-ray Crystallography: well-ordered crystals for diffraction
Mobile fragments challenging
Dynamic disorder: weak density
Multi-modal disorder: ambiguous/overlapping density
Existing modeling techniques break down:
-> Gaps in main-chain
High potential pay-off:
Mobile regions often facilitate binding
Impact on structural/functional studies,
drug design: virtual screening, lead optimization
How to Model Missing Fragments?

3. Modeling a Missing Fragment
Partial model of the molecule:
Missing fragments residues in length
‘Anchoring’ residues of the fragment correctly positioned
Known sequence
Start with attaching an open conformation to an anchor
Method consists of two stages:
Generate a set of promising, closed conformations that match the electron density reasonably well.
Torsion angle refinement of stage-1 conformers while maintaining closure
Stage 1: Main chain only.
Stage 2: Side chains added from backbone dependent rotamer library.
c-angle ‘free’ refinement.

4. Results aaRMSD = 0.6Å Final model Xpleo conformation
TM0813: residues 83-96
initial model: RESOLVE
resolution: 2.0Å, truncated at 2.8Å
contour: 1.0s
PDB: 1J5X

5. Consensus Modeler: an Algorithm to Combine Protein Models
Goal:
Obtain optimal starting model for manual refinement from combining all traces obtained from automated data processing
Idea:
Mix and match multiple incomplete inputs to increase completeness
Error reduction: Compare input models to identify and correct errors
Obtain a ‘globally optimal’ model: DP algorithm

6. Algorithm
Shift origin, superimpose all traces using crystallographic symmetry operations
Superimpose NCS related molecules using SS matching algorithm1
Represent models as a graph, with each vertex a residue connected by weighted edges.
Edge weight represents ‘goodness of fit’ of its pair of residues in the final model
Determine path through graph that globally maximizes the goodness of fit
traces
Output: - Single, optimally complete protein model
- Consensus reflection file; combined HL coefficients
- All superposed input traces, all origin-shifted reflection files
residues
1 Krissinel and Henrick, 2003

7. Edge Weights Problems in Models: Gaps
‘Dummy’ residues in graph represent gaps in output
Gaps ≤ 3 residues filled automatically
Frame Shift
Two models place different fragments at same location in density
Divergence
Input models disagree (peptide flip, main-chain in side-chain density, etc.)
Undocked Fragments
Solvent chains, GLY chains
Edge Weights Represent:
Agreement with same residue in other models
Density Fit (CC), Ramachandran values, Geometry
Continuity of the output model (Dummy Residues, Discontinuities in Sequence)

8. Results Target/CrystalID rsds res SG mol models best trace Consensus
PC07317D/22317
253
2.25 C2 1 2
78%
86%
PC02663D/22977
203
2.0 8
88%
93%
TM0771/20687
310 I4 3
69%
79%
TM1622/13219
160
1.9
P3221 37
84%
PC04261E/24045
311
2.56
P43212 6
73%1
1 31% docked