Page 1
CONFOLD: Residue-Residue Contact-guided ab initio Protein Folding
Badri AdhikariStudent of Dr. Jianlin Cheng
Department of Computer ScienceUniversity of MissouriColumbia MO 65211
3/13/2015
The 12th Annual MCBIOS Conference,Little Rock Marriott Conference and Convention
CenterLittle Rock, AR
Page 2
predict contacts
Introduction
contacts
Machine-learning based
DNcon, SVMcon
Coevolution-derivedPSICOV, CCMpred, DCAfold
HybridMetaPSICOV
ab initioEVFOLD
fragment-basedFRAGFOLD, ROSETTA
improvement
protein.rnet.missouri.edu/confold/
build 3D models
CONFOLD
Page 3
MetaPSICOVDr. David Jones at University College London (UCL)
mfDCADr. José Onuchic at UC San Diegohttp://dca.rice.edu/portal/dca
GREMLINDr. David Baker at University of Washington
CCMpredDr. Johannes Söding at University of Munich (LMU Munich)
FreeContactDr. Burkhard Rost at Technische Universität München (TUM)
DNconDr. Jianlin Cheng at University of Missouri Columbia
EVFOLDDr. Chris Sander at Memorial Sloan Kettering Cancer Center
EVFOLDDr. Debora MarksHarvard Medical School
Page 4
predict contacts
Introduction
contacts
Machine-learning based
DNcon, SVMcon
Coevolution-derivedPSICOV, CCMpred, DCAfold
HybridMetaPSICOV
ab initioEVFOLD
fragment-basedFRAGFOLD, ROSETTA
improvement
protein.rnet.missouri.edu/confold/
build 3D models
CONFOLD
Page 5
residue-residue contacts in proteins
Page 6
a contact in a proteinResidues i and j are in contact if
distance(Cβi, Cβj) ≤ 8 Å
Variations:- use of Cα instead of Cβ- other thresholds like 7 Å or 12 Å- minimum sequence separation of
5 or 6 residues
Page 7
# of contacts and threshold
at 8 Å threshold at 12 Å threshold at 20 Å threshold
Page 8
building 3D models using predicted contacts
Page 10
EVFOLD vs FRAGFOLD
to use, or not to use fragments,- that is the question.
http://en.wikipedia.org
Page 11
1. Reconstruct secondary structuresreconstruct Helices and beta-sheets
2. Handle noise in predicted contactsIf we try to satisfy all contacts,the model can easily jumble up
ab initio - Challenges
Page 12
CONFOLD method- convert secondary structures in to distance,
angle and h-bond restraints- combine contact restraints and secondary
structure restraints using weighting schemes- use of customized CNS suite v1.3 for modelling- two stage model-building method
- improves beta-sheet quality- improves accuracy of models
Page 13
build 3D
models
translate using derived restraints
restraints - dihedral - distance - h-bonds
restraints - distance
secondary
structure
contacts
select all or top-xL contacts
build 3D
models
restraints (updated) - dihedral - distance - h-bonds
restraints (filtered) - distance
weights
detect β-sheets
filter unsatisfied contacts
Page 15
Reconstructing Secondary Structures
Helix residues- 77 out of 79 on averageStrand residues- 33 out of 42 on average
Data Set:24 proteins in Tc category of CASP11
1YPI 2QOM
Page 16
Evaluation of best models
SPTB2_HUMAN
RMSD calculations
Page 17
Evaluation of all models
Page 18
How does the two-stage idea work?
stage 1 stage 2TM-score = 0.5 TM-score = 0.61
Page 19
Conclusion• We can reconstruct secondary structures, helices
and beta-sheets, in proteins from scratch, very well
• Existing ab inito methods for building 3D models using contacts can be improved
• Two-stage model building approach can improve models’ accuracy by removing noisy contacts
Page 20
Acknowledgementsto my advisor Dr. Cheng and
to my friends Deb, Renzhi, Jilong and Jie in our research lab.