Structural biology should be computable!

• Protein structures determined by amino acid sequences

• Protein structures and complexes correspond to global free energy minima

• Fundamental test of understanding and huge practical relevance

Model of energetics of inter and intramolecular

interactions

Design(Given Structure, OptimizeSequence)

Prediction(Given Sequence, OptimizeStructure)

Ab initio structure Protein Structure Protein designprediction

Protein-protein docking Protein-protein Interface design interactions

ROSETTA

Model of macromolecular interactions

• Removal of single methyl groups can destabilize proteins --> jigsaw puzzle-like packing crucial

• Buried polar atoms almost always hydrogen bonded --> treat hydrogen bonding as accurately as possible

• Exposed charge substitutions generally have little effect --> damp long range elctrostatics

• Focus on short range interactions!

Random Start

Low-Resolution Monte Carlo Search

(integrate out sidechain degrees of

freedom)

High-Resolution Refinement with full atomic detail

105

Predictions

Conformational sampling

Select lowest energy models

Jeff Gray (Hopkins),Ora Furman (Hebrew University), Chu Wang

Docking Low-Resolution Search

• Monte Carlo Search• Rigid body translations and

rotations• Residue-scale interaction

potentials

Protein representation: backbone atoms + average centroids

N

O

OO

N

O

N

O

N

N

O

......

Docking Protocol

(Target 12: cohesin-dockerin; unbound-bound) 1. Initial Search 2. Refinement

RMSD to arbitrary starting structure

Ene

rgy

RMSD to starting structure of refinement

(Å)

red,orange– xrayblue – model; green – unbound

0.46Å interface rmsd 87% native contacts 6% wrong contacts

Target 12Cohesin-Dockerin

Side Chain Flexibility

dockerin

cohesinOra Furman,Chu Wang

Details of T12 Interface

D39

N37

S45

L83

E86

Y74

L22

R53

dockerin

cohesin

red,orange– xrayblue - model

red,orange– xrayblue - model

0.23Å interface rmsd

Target 15immunity protein D-colicin D tRNase

Accurate Side Chain Modeling

colicin

immunity proteinScience 310, 638-642

Details of T15 Interface

H611

red,orange– xrayblue - model

E56

K610K608

K607

E68

E59D61

colicin

immunity protein

red,orange– xrayblue – model; green – unbound

2.34Å interface rmsd 36% native contacts

Target 20HemK-RF1

Modeling Backbone Movement

RF1

HemK

Loop with methylated Gln

Chu Wang

QuickTime™ and aYUV420 codec decompressor

are needed to see this picture.

CASP6 T0198: PhoU domain repeat

Model 2: 4A over 210 rsds

(Model 1: 3.94 over 198)

Phil Bradley

CASP6 T0212

Model 2: 3.97 over 109 rsds(Model 1: 4.0 over 104)

QuickTime™ and aDV/DVCPRO - NTSC decompressor

are needed to see this picture.

T0281 ab initio prediction (1.59Å)Phil Bradley

1r69

1ubq

Science 309, 1868-1871

2REB

Boinc.bakerlab.org/rosettaDavid Kim

High resolution ab initio structure prediction from single sequences by enhanced diversity “barcode” directed samplingOutreach!

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

High Resolution Refinement of CASP target 199 - remote homology model

Calculations performed on SDSC teragrid clustersBin Qian

High Resolution NMR Model Refinement

Vatson Raman

Disulfide Bond Formation Protein

Blue - X-ray structure Green - NMR models Red - Rosetta models

Computing Structural Biology• Free energy function reasonable => Computing simple

protein structures and interactions now appears to be within reach

• Implications for structural genomics? • More cpu power => more accurate predictions for larger

proteins• For larger complexes, experimental data essential (low

resolution electron density!).• Symmetry helps!

Modeling accuracy also illustrated by structures of designed proteins

Top7 X-ray structure has correct topology. Backbone RMSD to design only 1.2Å!!

C- Backbone Overlay

Red : X-ray structure

Blue : Design modelBrian Kuhlman, Gautam Dantas;Science 302 1364-8

Design of novel H bond network

interface

G177

Q51

Q180

Q169

Y35

G177Q180

Q169

Q51

Y35

G177

Y35

Design X-ray

Lukasz Joachimiak

Design of new protein functions

• Design of new protein-protein interactions• Design of enzymes catalyzing novel

chemical reactions • Design of new transcription factor and

endonuclease specificities• Design of HIV vaccine

HIV vaccine design

• Present HIV coat protein epitopes locked into conformation observed in complexes with neutralizing antibodies using designed scaffolds

• Preliminary results: designed proteins fold and bind neutralizing antibodies (5nM affinity). One design confirmed crystallographically.

Bill Schief in collaboration with Peter Kwong

Crystal structure of Mab 2F5in complex with its HIV epitope

Model of non-HIV scaffold-epitope (red)

Computational design of non-HIV immunogens to elicit broadly-neutralizing antibodies

Bill Schief

WT-WT Design-WT

WT-Design Design-Design

Redesign ofDNA cleavage specificity of MsoIhoming endonuclease using ROSETTA

Justin Ashworth,Jim HavranekNature in press

Specific DNA cleavage by designed nuclease

wild-type I-Mso

Design

-1/2n1

wild-type

design

wild-type

design

Cleavage

targets

½ ¼

-

1/29

5uMnuclease

Acknowledgements

Design• Brian Kuhlman (UNC)• Gautam Dantas• Justin Ashworth• Jim Havranek

Robetta.bakerlab.org

prediction and design server: David Kim (domain parsing, boinc) and Dylan Chivian

Rosetta software freely available for academic use

Boinc.bakerlab.org/rosetta

Protein structure prediction• Phil Bradley (MIT) • Rhiju Das• Lars Marlstrom• Bin Qian• Vatson RamanProtein-protein docking• Ora Furman (Hebrew

University)• Chu Wang• Jeff Gray (Johns

Hopkins)