structure prediction: homology modeling lecture 8 structural bioinformatics dr. avraham samson...

Structure prediction: Homology modeling

Lecture 8Structural Bioinformatics

Dr. Avraham Samson81-871

Homology Modeling

• Presentation• Fold recognition• Model building

– Loop building– Sidechain modeling– Refinement

• Testing methods: the CASP experiment

Homology Modeling

• Presentation• Fold recognition• Model building

– Loop building– Sidechain modeling– Refinement

• Testing methods: the CASP experiment

Why do we need homology modeling ?

To be compared with:

Structural Genomics project

• Aim to solve the structure of all proteins: this is too much work experimentally!

• Solve enough structures so that the remaining structures can be inferred from those experimental structures

• The number of experimental structures needed depend on our abilities to generate a model.

Proteinswithknownstructures

Unknown proteins

Structural Genomics

Homology Modeling: why it works

High sequence identity

High structure similarity

Homology Modeling: How it works

o Find template

o Align target sequence with template

o Generate model:- add loops- add sidechains

o Refine model

Homology Modeling

• Presentation• Fold recognition• Model building

– Loop building– Sidechain modeling– Refinement

• Testing methods: the CASP experiment

Fold Recognition

Homology modeling refers to the easy case when the template structure can be identified using BLAST alone.

What to do when BLAST fails to identify a template?

•Use more sophisticated sequence methods•Profile-based BLAST: PSIBLAST•Hidden Markov Models (HMM)

•Use secondary structure prediction to guide the selection of a template, or to validate a template

•Use threading programs: sequence-structure alignments

•Use all of these methods! Meta-servers: http://bioinfo.pl/Meta

Fold Recognition

Blast for PDB search

Full homology modeling packages

Profile based approach

HMM

Structure-derived profiles

Fold recognition andSecondary structure prediction

Homology Modeling

• Presentation• Fold recognition• Model building

– Loop building– Sidechain modeling– Refinement

• Testing methods: the CASP experiment

Very short loops: Analytic Approach

Wedemeyer,ScheragaJ. Comput. Chem.20, 819-844(1999)

Medium loops: A database approach

Scan database and search protein fragments with correct number of residuesand correct end-to-end distances

Loop length

cRM

S (Ǻ

)

Method breaksdown for loopslarger than 9

Medium loops: A database approach

1) Clustering Protein Fragments to Extract a Small Set of Representatives (a Library)

data clustereddata

library

Long loops: A fragment-based approach

Generating Loops

Fragment library

Generating Loops

Fragment library

Generating Loops

Fragment library

Generating Loops

Fragment library

Generating Loops

Fragment library

Threading

Loop length

<cR

MS

(Ǻ)

Long loops: A fragment-based approach

Test cases: 20 loops for each loop length Methods: database search, and fragment building, with fragment libraries of size L

Loop building: Other methods

Heuristic sampling (Monte Carlo, simulated annealing)

Inverse kinematics

Relaxation techniques

Systematic sampling

http://www.cs.ucdavis.edu/~koehl/BioEbook/loop_building.html

Homology Modeling

• Presentation• Fold recognition• Model building

– Loop building– Sidechain modeling– Refinement

• Testing methods: the CASP experiment

Self-Consistent Mean-Field Sampling

P(J,1)

P(J,3)

P(J,2)

i,2i,1i,3

Self-Consistent Mean-Field Sampling

i,2i,1i,3

P(i,1)

P(i,3)

P(i,2)

P(i,1)+P(i,2)+P(i,3)=1

Self-Consistent Mean-Field Sampling

Multicopy Protein

i,1

i,2i,3

j,1

j,3

j,2

Self-Consistent Mean-Field Sampling

E(i,k) = U(i,k) + U(i,k,Backbone)

Multicopy Protein Mean-Field Energy

i,1

i,2i,3

j,1

j,3

j,2

Self-Consistent Mean-Field Sampling

Nrot(i)

1l

new

RTl)E(i,

exp

RTk)E(i,

expk)(i,P

E(i,k) = U(i,k) + U(i,k,Backbone)

Multicopy Protein Mean-Field Energy

Update Cycle

i,1

i,2i,3

j,1

j,3

j,2

(Koehl and Delarue, J. Mol. Biol., 239:249-275 (1994))

Self-Consistent Mean-Field Sampling

Another way is simply aligning the side chains as well

Dead End Elimination (DEE) Theorem

• There is a global minimum energy conformation (GMEC) for which there is a unique rotamer for each residue

• The energy of the system must be pairwise.

1

1 11

),(),()(N

i

N

ijsr

N

irTemplateTot jiETemplateiEECE

Each residue i has a set of possible rotamers. The notation ir means residue i has the conformation described by rotamer r.

The energy of any conformation C of the protein is given by:

CGMECECE TotTot )()(

Note that:

Dead End Elimination (DEE) Theorem

Consider two rotamers, ir and it, at residue i and the set of all other rotamer conformations {S} at all residues excluding i.

If the pairwise energy between ir and js is higher than the pairwise energy between it and js, for all js in {S}, then ir cannot exist in the GMEC and is eliminated. Mathematically:

}{,,,, SjiETemplateiEjiETemplateiEij

srrij

srr

If

then

ir does not belong to the GMEC

This is impractical as it requires {S}. It can be simplified to:

ij

sts

tij

srs

r jiETemplateiEjiETemplateiE ,max),(,min),(

Dead End Elimination (DEE) Theorem

If

then

ir does not belong to the GMEC

Iteratively eliminate high energy rotamers: proved to converge to GMEC

Desmet, J, De Maeyer, M, Hazes, B, Lasters, I. Nature, 356:539-542 (1992)

Other methods for side-chain modeling

-Heuristics (Monte Carlo, Simulated Annealing)SCWRL (Dunbrack)

-Pruning techniques

-Mean field methods

Loop building + Sidechain Modeling: generalized SCMF

Template

Add multi-copiesof candidate “loops”

Add multi-copiesof candidate side-chains

Final modelKoehl and Delarue.Nature Struct. Bio.2, 163-170 (1995)

Homology Modeling

• Presentation• Fold recognition• Model building

– Loop building– Sidechain modeling– Refinement

• Testing methods: the CASP experiment

Refinement ?

CASP6 assessors, homology modeling category:

“We are forced to draw the disappointing conclusion that, similarlyto what observed in previous editions of the experiment, no modelresulted to be closer to the target structure than the template toany significant extent.”

The consensus is not to refine the model, as refinement usually pulls the model away from the native structure!!

Homology Modeling: Practical guideApproach 2: Submit target sequence to automatic servers

- Fully automatic:

- 3D-Jigsaw : http://www.bmm.icnet.uk/servers/3djigsaw/

- EsyPred3D: http://www.fundp.ac.be/urbm/bioinfo/esypred/

- SwissModel: http://swissmodel.expasy.org//SWISS-MODEL.html

- Fold recognition:

- 3D-PSSM: http://www.sbg.bio.ic.ac.uk/~3dpssm/

- Useful sites:

- Meta server: http://bioinfo.pl/Meta

- PredictProtein: http://cubic.bioc.columbia.edu/predictprotein/

Comparative Modeling Server & Program

COMPOSER http://www.tripos.com/sciTech/inSilicoDisc/bioInformatics/matchmaker.html

MODELER http://salilab.org/modeler

InsightII http://www.msi.com/

SYBYL http://www.tripos.com/