Modelling proteomes Ram Samudrala University of Washington

Modelling proteomes Ram Samudrala University of Washington

What is a “proteome”? All proteins of a particular system (organelle, cell, organism) What does it mean to “model a proteome”? For any protein, we wish to: ANNOTATION { - figure out what it looks like (structure or form) - understand what it does (function) Repeat for all proteins in a system Understand the relationships between all of them } EXPRESSION + INTERACTION

Figuring out protein structure Experimental methods: - X-ray crystallography, NMR, electron diffraction - Accurate but slow Predictive methods: - De novo prediction, comparative modelling - Automated and relatively fast; not as accurate experimental methods

CASP 5 prediction for T 138 4. 6 Å Cα RMSD for 84 residues

CASP 5 prediction for T 146 5. 6 Å Cα RMSD for 67 residues

CASP 5 prediction for T 170 4. 8 Å Cα RMSD for all 69 residues

CASP 5 prediction for T 129 5. 8 Å Cα RMSD for 68 residues

CASP 5 prediction for T 172 5. 9 Å Cα RMSD for 74 residues

CASP 5 prediction for T 187 5. 1 Å Cα RMSD for 66 residues

CASP 5 prediction for T 129 1. 0 Å Cα RMSD for 133 residues (57% id)

CASP 5 prediction for T 182 1. 0 Å Cα RMSD for 249 residues (41% id)

CASP 5 prediction for T 150 2. 7 Å Cα RMSD for 99 residues (32% id)

CASP 5 prediction for T 185 6. 0 Å Cα RMSD for 428 residues (24% id)

CASP 5 prediction for T 160 2. 5 Å Cα RMSD for 125 residues (22% id)

CASP 5 prediction for T 133 6. 0 Å Cα RMSD for 260 residues (14% id)

Figuring out protein function Experimental methods: - A variety of single molecule and genomic/proteomic techniques - Usually not complete; slow Predictive methods: - Structure and sequence based methods using experimental knowledge of known functions - Automated and relatively fast; not as accurate as experimental methods; may be comprehensive

Structure and sequence based functional studies Using structure to rationalise mutagenesis data: - Dr adhesin study with Steve Moseley (Molec. Micro. 2002) - Invb study with Sam Miller Using structure to rationalise putative function: - Bacterial tubulin study with Jim Staley (PNAS 2003) Using structure and sequence to predict inhibitor binding: - Protease inhibitor binding studies using docking (BMC 2003) - Sequence-based prediction using regression models for HIV protease and RT with John Mittler (submitted) Structure and sequence based methods for functional classification: - In progress - General method to assign function from sequence - Analyse binding of 130 K compounds to structures - Build machine learning models of functional families

Prediction of SARS Co. V proteinase inhibitors Ekachai Jenwitheesuk

Integrated structural and functional annotation of proteomes structure based methods microenvironment analysis Bioverse structure comparison * * homology zinc binding site? * * function? + sequence based methods assign function to entire protein space sequence comparison motif searches phylogenetic profiles domain fusion analyses + experimental data single molecule + genomic/proteomic } EXPRESSION + INTERACTION

Bioverse – explore relationships among molecules and systems http: //bioverse. compbio. washington. edu Jason Mc. Dermott

Bioverse – explore relationships among molecules and systems Jason Mcdermott

Bioverse – prediction of protein interaction networks Target proteome Interacting protein database protein α 85% experimentally determined interaction protein A predicted interaction protein B protein β 90% Assign confidence based on similarity and strength of interaction Jason Mcdermott

Bioverse – E. coli predicted protein interaction network Jason Mc. Dermott

Bioverse – M. tuberculosis predicted protein interaction network Jason Mc. Dermott

Bioverse – C. elegans predicted protein interaction network Jason Mc. Dermott

Bioverse – H. sapiens predicted protein interaction network Jason Mc. Dermott

Example uses of the Bioverse annotations and predicted networks Network-based corroboration of functional annotation: - Proteins that interact with each other have related or identical functions - Highly accurate Identification of essential proteins: - analyse key positions (central or articulation points) in the networks - 30 -40% accuracy; useful for hypothesis validation Identification of virulence factors and molecular mimicry: - analyse host-pathogen protein interaction networks - analyse host-pathogen protein similarity networks Evolutionary analysis and genetic engineering: - compare networks from different organisms/strains - mix and match proteins and sub-networks

Bioverse – network-based annotation for C. elegans Jason Mc. Dermott

Bioverse – identifying key proteins on the anthrax predicted network Articulation point proteins Jason Mc. Dermott

Bioverse – identifying key proteins on the rice predicted network Defense-related proteins Jason Mc. Dermott

Bioverse – viewer Aaron Chang

Take home message Prediction of protein structure, function, and networks may be used to model whole genomes to understand organismal function and evolution

Acknowledgements Aaron Chang Ashley Lam Ekachai Jenwitheesuk Gong Cheng Jason Mc. Dermott Kai Wang Ling-Hong Hung Lynne Townsend Marissa La. Madrid Mike Inouye Stewart Moughon Shing-Chung Ngan Yi-Ling Cheng Zach Frazier National Institutes of Health National Science Foundation Searle Scholars Program (Kinship Foundation) UW Advanced Technology Initative in Infectious Diseases