[modeller_usage] Regarding Model Clustering

[Ashish Runthala <>]

Dear Modellers,
What is the best way to cluster the high scoring model decoys into a well refined conformation with almost correct domain topology, as existing in the considered models. 
I tried using transfer_xyz to pool 2 best complete models(being best for each of the 2 individual domains, considered on TM-score,GDT).
I also tried building the new model through these 2 models using them as templates in different salign modules, but i failed to improve the model for both the individual domains and the overall model together. For 1 domain my prediction with a set of templates and my representative alignment had GDT-TS of 96.875 for Domain1. For D2 best model, it was 99.645. 
The moment i used Salign, topology got messed up.
So when i just pooled (copied) the individual correct domain coordinates from best models for D1 and D2 into the complete model, RMSD became 2.433 with GDT-TS of just 63.309. WHY IT HAPPENED? So only refinement is used by many CASP9 teams to further refine this topology. But this big difference of GDT-TS can't be corrected by refinement. 

WHY 2 CORRECT DOMAIN MODELS PUT OR COPIED TOGETHER CAN'T GIVE AN OVERALL CORRECT MODEL.
HOW TO USE SPICKER. TELL ME THE CORRECT PATH TO CLUSTER THE MODELS FOR BUILDING AN OVERALL CORRECT MODEL TOPOLOGY. 

Thanks for support,
Ashish



Ashish Runthala,
Lecturer, Structural Biology Cell,
Biological Sciences Group,
BITS, Pilani
Rajasthan, INDIA