Prediction of protein structure

aim Structure prediction tries to build models of 3D structures of proteins that could be useful for understanding structure-function relationships.

Genbank/EMBL 105.000.000 Uniprot 5.200.000 PDB 47.000

The protein folding problem The information for 3D structures is coded in the protein sequence Proteins fold in their native structure in seconds Native structures are both thermodynamically stables and kinetically available

ab-initio prediction Prediction from sequence using first principles AVVTW...GTTWVR

Ab-initio prediction “In theory”, we should be able to build native structures from first principles using sequence information and molecular dynamics simulations: “Ab-initio prediction of structure” Simulaciones de 1 ms de “folding” de una proteína modelo (Duan-Kollman: Science, 277, 1793, 1998). Simulaciones de folding reversible de péptidos (20-200 ns) (Daura et al., Angew. Chem., 38, 236, 1999). Simulaciones distribuidas de folding de Villin (36-residues) (Zagrovic et al., JMB, 323, 927, 2002).

... the bad news ... It is not possible to span simulations to the “seconds” range Simulations are limited to small systems and fast folding/unfolding events in known structures steered dynamics biased molecular dynamics Simplified systems

typical shortcuts Reduce conformational space 1,2 atoms per residue fixed lattices Statistic force-fields obtained from known structures Average distances between residues Interactions Use building blocks: 3-9 residues from PDB structures

Results from ab-initio Average error 5 Å - 10 Å Function cannot be predicted Long simulations Some protein from E.coli predicted at 7.6 Å (CASP3, H.Scheraga)

comparative modelling The most efficient way to predict protein structure is to compare with known 3D structures

Protein folds

Basic concept In a given protein 3D structure is a more conserved characteristic than sequence Some aminoacids are “equivalent” to each other Evolutionary pressure allows only aminoacids substitutions that keep 3D structure largely unaltered Two proteins of “similar” sequences must have the “same” 3D structure

Possible scenarios 1. Homology can be recognized using sequence comparison tools or protein family databases (blast, clustal, pfam,...). Structural and functional predictions are feasible 2. Homology exist but cannot be recognized easily (psi-blast, threading) Low resolution fold predictions are possible. No functional information. 3. No homology 1D predictions. Sequence motifs. Limited functional prediction. Ab-initio prediction

fold prediction

3D struc. prediction

1D prediction Prediction is based on averaging aminoacid properties AGGCFHIKLAAGIHLLVILVVKLGFSTRDEEASS Average over a window

1D prediction. Properties Secondary structure propensitites Hydrophobicity (transmembrane) Accesibility ...

Propensities Chou-Fasman Biochemistry 17, 4277 1978 a b turn

Some programs (www.expasy.org) BCM PSSP - Baylor College of Medicine Prof - Cascaded Multiple Classifiers for Secondary Structure Prediction GOR I (Garnier et al, 1978) [At PBIL or at SBDS] GOR II (Gibrat et al, 1987) GOR IV (Garnier et al, 1996) HNN - Hierarchical Neural Network method (Guermeur, 1997) Jpred - A consensus method for protein secondary structure prediction at University of Dundee nnPredict - University of California at San Francisco (UCSF) PredictProtein - PHDsec, PHDacc, PHDhtm, PHDtopology, PHDthreader, MaxHom, EvalSec from Columbia University PSA - BioMolecular Engineering Research Center (BMERC) / Boston PSIpred - Various protein structure prediction methods at Brunel University SOPM (Geourjon and Deléage, 1994) SOPMA (Geourjon and Deléage, 1995) AGADIR - An algorithm to predict the helical content of peptides

1D Prediction Original methods: 1 sequence and uniform parameters (25-30%) Original improvements: Parameters specific from protein classes Present methods use sequence profiles obtained from multiple alignments and neural networks to extract parameters (70-75%, 98% for transmembrane helix)

Methods for remote homology Homology can be recognized using PSI-Blast Fold prediction is possible using threading methods Acurate 3D prediction is not possible: No structure-function relationship can be inferred from models

Threading Unknown sequence is “folded” in a number of known structures Scoring functions evaluate the fitting between sequence and structure according to statistical functions and sequence comparison

ATTWV....PRKSCT .......... SELECTED HIT 10.5 > .......... 5.2

ATTWV. PRKSCT. Sequence HHHHH. CCBBBB. Pred. Sec. Struc. eeebb. eeebeb ATTWV....PRKSCT Sequence HHHHH....CCBBBB Pred. Sec. Struc. eeebb....eeebeb Pred. accesibility .......... Sequence GGTV....ATTW ........... ATTVL....FFRK Obs SS BBBB....CCHH ........... HHHB.....CBCB Obs Acc. EEBE.....BBEB ........... BBEBB....EBBE

Threading accurancy

Comparative modelling Good for homology >30% Accurancy is very high for homology > 60% Reminder The model must be USEFUL Only the “interesting” regions of the protein need to be modelled

Expected accurancy Strongly dependent on the quality of the sequence alignment Strongly dependent on the identity with “template” structures. Very good structures if identity > 60-70%. Quality of the model is better in the backbone than side chains Quality of the model is better in conserved regions

Quality test No energy differences between a correct or wrong model The structure must by “chemically correct” to use it in quantitative predictions

Analysis software PROCHECK WHATCHECK Suite Biotech PROSA

Prediction software SwissModel (automatic) SwissModel Repository http://www.expasy.org/swissmod/ SwissModel Repository http://swissmodel.expasy.org/repository/ 3D-JIGSAW (M.Stenberg) http://www.bmm.icnet.uk/servers/3djigsaw/ Modeller (A.Sali) http://salilab.org/modeller/modeller.html MODBASE (A. Sali) http://alto.compbio.ucsf.edu/modbase-cgi/index.cgi

Final test The model must justify experimental data (i.e. differences between unknown sequence and templates) and be useful to understand function.