Proteomics: focus on protein interaction

Report
Abstracts
List of participants
Photos

An extensive meeting report and selected papers are being published in the October issue of Comparative and Functional Genomics.

Report ...

The central theme of the workshop was the confrontation of a variety of methods, which aim to describe the "complete" protein interaction network that operates in a living cell. This goal poses two problems: one is the collection of experimental data by high throughput methods and the second is its organization in user friendly annotated databases.

The workshop addressed both problems as well as their integration. The sixteen invited speakers and the participants included experts in fields such as genetic methods of identifying protein partners, protein and peptide arrays, mass spectrometry and computational biology.

The main purpose of the workshop was to bring these very different expertises together, with the objective of defining strategies and bottlenecks in the effort of collecting reliable protein interaction information. Both the retrieval of information from the large but poorly accessible repository of scientific literature and the implementation of new experimental approaches, which makes use of recently developed high throughput technologies, received attention during the workshop.

The workshop was organised into four sessions
Genetic Methods
Protein Arrays
Mass spectrometry
Computational Methods.

A brief description of the topics discussed is presented here.

Genetic Methods

Pierre Legrain (Hybrigenics, France) discussed recent reports on whole genome protein interaction maps obtained by the two hybrid method and compared the results of two efforts aimed at characterising the complete interaction map of S. cerevisiae. Surprisingly he noted that there is little overlap between the two proposed maps. The different strategies were compared and the "superiority" of an approach that tests interaction of protein domains instead of whole proteins was proposed. Andreas Pluckthun (Biochemisches Institut, Switzerland) illustrated the advantages of PCA which, by working in E.coli, permits the testing of larger libraries. The feasibility of a library versus library experiment was also reported. Although still in its developmental phase, ribosome display has the clear advantage of being an in vitro technique and is therefore not affected by the limitations of transformation experiments. Finally Brian Kay (University of Wisconsin, USA) reported how selection of peptide ligands from repertoires displayed on filamentous bacteriophage can be used to infer the natural ligands of any protein domain.

Peptide and protein chips

Three talks by Jens Schneider Mergener (JERINI, Germany), Dolores Cahill (MPI Molecular Genetics, Germany) and Ian Humphery Smith (Utrecht University, the Netherlands) presented a clear overview of this rapidly developing field. Pep Spot (high density peptides synthesised on solid supports) is possibly the most established technique and a variety of applications of "biological relevance" were reported. The assembly of large protein or antibody arrays still poses a variety of technical problems (collection of clones, expression of the protein repertoire in a soluble form, attachment to a solid support without disturbing protein structure, background etc) but there is confidence that these will soon be solved.

Mass Spectrometry

Peter Roepstorff (University of Southern Denmark, Denmark) in his lecture highlighted some recent developments and applications of mass spectrometry (MS) in proteome analysis. This technique is now accepted as the method of choice for the identification of low abundance proteins and characterisation of post-translational modifications. Developments in MS over the past few years now allow the characterisation of subpicomole quantities of gel-separated proteins. The molecular weight and sequence information derived from MS experiments can be used to interrogate large protein and nucleotide databases for the identification of the rapidly growing number of known proteins or alternatively, for the cloning of novel proteins.
Benedetta Mattei (University of Rome La Sapienza, Italy) illustrated the combination of surface plasmon resonance (SPR) biosensors and MS as a tool to couple sensitive affinity capture and characterisation of binding events with the ability to identify and characterise interacting molecules. The biosensor can confirm and quantify specific binding events to a target and it is possible to identify interacting proteins eluted from the chip or even from tryptic digests performed directly on sensor surfaces.
Carol Robinson (Oxford Centre for Molecular Sciences, UK) gave several examples on how electrospray ionization mass spectrometry (ESI-MS) can be used to study protein interactions driven by non-covalent forces, including the study of the 70S ribosomal particle and the complex of the spliceosome. The gentleness of the electrospray ionization process allows intact protein complexes to be directly detected by mass spectrometry, allowing questions about non-covalent assembly to be addressed by the direct observation of gas phase complexes, their assembly in real time and their disassembly by perturbation of solution or instrument conditions. For the study of protein interactions, ESI-MS can be a valid complement to other biophysical methods, such as NMR and X-ray crystallography.
Finally Giulio Superti Furga (CellZome, Germany) reported on an ongoing project which aims to identify all of the yeast protein complexes by genomic tagging and purification of more than 6000 gene products.

Bioinformatics

The session covered two major topics: the annotation and architecture of biological databases (Mike Sternberg and Thure Etzold) and the possibility of deriving information on protein-protein interactions using computational methods (AlfonsoValencia, Rita Casadio, Jong Park and Manuela Helmer Citterich).
Michael Sternberg (ICRF, London) opened the session with a talk about "Structural and functional annotation of genomes". Different methods, PSI-BLAST and 3D-PSSM, were discussed. These can be used on translated ORFs in comparative genomics, screening, drug design and pathway reconstruction. Thure Etzold (EMBL-EBI Cambridge, UK), is involved in the development of the SRS system, and discussed the possibility of integrating biological databases.
Rita Casadio (University of Bologna, Italy) described a new neural network approach to identify protein residues involved in protein-protein interaction. An overview about the mapping of the different folds involved in the formation of protein complexes was given by Jong Park (MRC Dunn, Cambridge, UK). Manuela Helmer Citterich (University of Rome Tor Vergata, Italy) described the SPOT method and the MINT database, dedicated to protein interaction.
Alfonso Valencia (CNB-CSIC Madrid, Spain) closed the session with a talk about the prediction of protein interaction from sequence information using a new method based on the comparison of phylogenetic trees. Some time was also devoted to methods for the extraction of biological information from the literature.

Abstracts ...

How to compare experimental data on protein interactions ?
Pierre Legrain

Large-scale protein interaction maps are, with gene expression profiles, among the first examples of datasets generated without specific knowledge on functions of genes. These are technology-driven experiments rather than hypothesis-driven experiments. They are valuable tools for protein function prediction, despite the occurrence of typical artefacts. These approaches are still in their early stages.
The matrix approach uses the same collection of proteins as both bait and prey. The library screening approach identifies for each interacting prey protein the domain of interaction with a given bait. The rate of false positive interactions is difficult to evaluate and is largely dependent on the criteria applied for the significance of the interactions, such as the reproducibility of results. Moreover, the two matrix exhaustive studies of the yeast proteome have failed to recapitulate as many as 90% of interactions previously described in the literature, suggesting a very high level of false negatives. Evaluation of false positives and reproducibility requires access to primary data. Thus, bioinformatics tools might also contribute to identifying false positives.
Bioinformatics clustering of protein interactions represents a powerful annotation tool which will become more and more useful as the interaction data accumulate. However, one major hurdle in these bioinformatics prediction algorithms is clearly the lack of independently validated methods. In order to be used successfully for appropriate functional annotation of protein clusters, the data needs to be stored in elaborate structures that allow each individual scientist to test his/her own hypothesis against complex heterogeneous primary data and then to design further experimental settings to validate the functional assignment.

Mapping protein-protein interactions with combinatorial peptides
Brian Kay

Once the sequence of a genome is in hand, understanding the function of its encoded proteins becomes a task of paramount importance. Much like the biochemists who first outlined different biochemical pathways, many genomic scientists are engaged in determining which proteins interact with which proteins, thereby establishing a protein interaction network. These interactions have evolved specificity, affinity and cellular function over billions of years; however, in the laboratory it is possible to isolate peptides from combinatorial libraries that bind to the same proteins with similar specificity and affinity and primary structures that resemble those of the natural interacting proteins. We have termed this phenomenon 'convergent evolution'. Thus, a fruitful approach for mapping protein-protein interactions is to isolate peptide ligands via phage-display to a target protein and identify candidate interacting proteins in a sequenced genome by computer analysis. We have applied this method to dissecting molecular interactions in the protein machinery involved in receptor-mediated endocytosis.

Mapping Protein-Protein-Interactions with Synthetic Peptide and Protein Domain Arrays
Jens Schneider-Mergener

Synthetic peptide and protein domain arrays prepared using the SPOT technology are increasingly applied to map the interactions between antibodies/antigens, receptors/ligands and proteins/nucleic acids1,2. The SPOT technology involves different aspects, such as array preparation4, types of molecules selected for the binding studies3,5 and the different types of binding assays performed on the arrays1.

References
1Frank, R. and Schneider-Mergener, J. (2001). SPOT-synthesis: scope and applications. Introduction to: Peptide arrays on membrane supports - synthesis and applications. Springer lab manual (Koch/Mahler eds.), in press.
2Reineke, U., Volkmer-Engert, R. and Schneider-Mergener, J. (2001). Applications of peptide arrays prepared by the Spot technology. Curr. Opin. Biotech. 12, 59-64.
3Töpert, F., Pires, R., Landgraf, C., Oschkinat, H. and Schneider-Mergener, J. (2001). Synthesis of an array comprising 837 variants of the hYAP WW protein domain. Angew. Chemie Int. Ed. 40, 897-900.
4Wenschuh, H., Volkmer-Engert, R., Schmidt, M., Schulz, M., Schneider-Mergener, J. and Reineke, U. (2000). Coherent membrane supports for parallel micro-synthesis of bioactive peptides. Biopolymers 55, 188-206.
5Reineke, U., Kramer, A. and Schneider-Mergener, J. (1999). Knowledge- and library-based mapping of discontinuous protein-protein-interactions by spot synthesis. Curr. Top. Microbiol. Immunol. 243, 23-36.

Generation and Applications of High Density Protein Arrays
Dolores Cahill.

A full understanding of the expression profile of a tissue or organism requires the screening of many genetic and/or protein samples in parallel as rapidly as possible. In our laboratory, we have automated and miniaturised various steps to enable a high-throughput and highly parallel approach to large-scale cDNA and protein analysis; specifically, the generation and picking of cDNA expression libraries, and arraying of clones into microtitre-plates. A technique known as oligonucleotide fingerprinting has been developed to characterise cDNA libraries, which allows the generation of non-redundant UNIgene sets. To apply this technology to generate protein arrays, we have clonally expressed proteins from arrayed cDNA expression libraries, producing high density protein arrays on filters and chips. These protein arrays have been screened with antibodies, which detected specific protein products. This approach makes translated gene products directly amenable to high-throughput experimentation, allowing a link between expressed protein and sequence. We have obtained initial results in characterising antibody specificity and profiling autoimmune sera on protein arrays, as well as developing antibody arrays.

Modification-Specific Proteomics: the next level in proteome analysis
Peter Roepstorff

The advances in DNA sequencing and the rapidly increasing amount of genome sequence data becoming available have changed the scope of protein analysis. Databases now contain the sequences of more than 450,000 proteins and this number is rapidly increasing. Consequently, the amino acid sequence of a protein of interest is likely to be available in a database. Sequencing of complete genomes raises a number of questions:

- "Which of the genes are expressed in the organism?" or, if the organism is a higher eukaryote, "which genes are expressed in which cell types?" The complete 2D-PAGE map of the proteins expressed in a given cell type has recently been termed the "proteome"1. The proteins in relevant gel spots are identified using mass spectrometric peptide mass mapping or sequencing after in-gel proteolytic digestion of the proteins.
- "Does a given protein contain post-translational modifications and, if so, which and where?" Numerous such modifications have been reported in proteins, the most common being truncation, glycosylation, phosphorylation and acylation. Genome and cDNA sequencing, however, do not give information about the presence of these modifications: they must be studied at the protein level and mass spectrometry is the key analytical tool for such studies. The information generated in the protein identification step can sometimes be used directly to assign the type of post-translational modification. Additional mass spectrometric experiments can be performed to fully characterize a protein; however, specific detection of selected modifications would be advantageous.
- The final question concerns the function of the protein - with what does the protein interact and how? This is a much more complex problem which requires a number of biochemical and chemical analytical procedures to solve it, among which mass spectrometry plays an important role.

We have attempted to develop fast, sensitive and highly specific tools based on combining PAGE, mass spectrometry and sequence information in databases2. Recently, we have initiated a new approach which we have termed: 'Modification Specific Proteomics'3. It is based on specific detection of post-translational modifications in the 2D-gel, by specific 'pull-out' of modified proteins/peptides, or by selective detection of the specific type of modification in the mass spectrometer. Examples from our recent research include mass spectrometric identification of the proteins in the 2D-gel, determination of secondary modifications in the identified proteins and our initial attempts to perform modification specific proteomics4-6. We plan to extend our mass spectrometric analysis to studies of protein interaction.

References
(1) Wilkins, M.R. et al. (1996) Bio/Technology, 14, 61-65.
(2) Jensen O.N., Larsen M.L., and Roepstorff, P. (1998) PROTEINS: Structure, Function, and Genetics Suppl. 2, 74-89.
(3) Jensen ON, (2000) Proteomics: A trends guide, 36-42.
(4) Larsen M.R. and Roepstorff P. (2000) Fresenius' J Anal Chem. 366, 677-690.
(5) Larsen MR, Soerensen GL, Fey SJ, Larsen PM, Roepstorff P. (2001) Proteomics 1, 223-238.
(6) Larsen MR, Larsen PM, Fey SJ, Roepstorff P. (2001) Electrophoresis 22, 566-575.

The combination of Surface Plasmon Resonance (SPR) biosensors and mass spectrometry
Benedetta Mattei

Surface plasmon resonance (SPR) biosensors are important tools for ligand screening due to their versatility. They allow to measure interactions in real time, require very little material and usually little or any chemical modification of the interactants. Once a sample is identified as containing a possible ligand, the problem arises to identify the new ligand. The identification of proteins at the femtomole level is made possible by the use of sensitive mass spectrometers and advanced database searching algorithms. Applications of biosensor technology coupled with mass spectrometry have been developed, allowing to characterize proteins eluted from sensor surfaces and to identify proteins from tryptic digestions performed directly on the sensor chip.
A combination of SPR and MALDI-TOF mass spectrometry was used to affinity purify peptides of the enzyme polygalacturonase (PG) that are recognized by its inhibitor PGIP, from a peptide mixture obtained by limited proteolysis of the native enzyme.
One peptide was identified: it comprises residues 181-244 and includes the three catalytic residues D191, D212 and D213. Site-directed mutagenesis and SPR data have shown that these residues are not involved in the interaction with PGIP, while the mutation of residues in the same peptide that are located at the entrance of the active site cleft causes a significant decrease in the affinity for PGIP. From these data we can hypotesize a mechanism of inhibition based on a network of contacts that are close to the active site but not buried inside the cleft, thus providing steric hindrance to substrate entry. The influence of post-translational modifications on the binding has also been studied by performing enzymatic deglycosylation of PG and characterizing by mass spectrometry the protein eluted from the sensor surface.

Dynamic protein complexes: insights from mass spectrometry
Carol V. Robinson

The recent sequencing of the human genome has revealed that the human cell contains only twice the number of genes as, for example, cells in the worm or fly1. This implies that the regulation of gene products and their interactions accounts for the increased biological complexity of higher organisms. Consequently, in order to exploit the wealth of information provided by genome sequencing it is essential to be able to study both stable and transient macromolecular complexes. While the yeast two-hybrid system and cross linking combined with mass spectrometry (MS) are exceptionally powerful approaches to defining stable complexes within the proteome, MS has the additional potential to describe transient, dynamic complexes through two major developments. These are (i) the ability to probe molecular dynamics through the coupling of MS with hydrogen/deuterium exchange technologies, and (ii) the control of the conditions within the mass spectrometer such that non-covalent interactions between proteins and cofactors can be examined.
Over the last decade, hydrogen/deuterium exchange in conjunction with MS has developed to an extent where it can probe the exchange behaviour of regions of secondary structure in macromolecular complexes2 and even individual residues in smaller proteins3. In parallel with these developments, major advances have been made in the ability to study non-covalent complexes in the gas phase. Specifically, the coupling of time-of-flight methods with electrospray and the refinement of this process to a nanoflow technique have enabled the study of simple dimeric complexes4, homo-5 and hetero-oligomeric complexes6 and even whole particles7,8. Consequently, the ability to probe both non-covalent interactions and hydrogen/deuterium exchange enables definition of not only the stoichiometry of interacting subunits but also their conformational dynamics. The fact that complexes can be observed in the mass spectrometer enables their stability and folding to be probed in the presence of a wide range of ligands and cofactors as well as in response to thermal and chemical denaturation.

Chaperone-assisted protein folding
Central to the success of protein folding in vivo is the prevention of aggregation, a role ascribed to molecular chaperones. The most widely studied of all molecular chaperones is the Escherichia coli chaperonin GroEL and its co-chaperonin GroES. GroES forms a single heptameric ring of seven identical subunits (70 kDa). The 14 GroEL subunits (57 kDa) are joined through non-covalent forces to form a double toroidal structure with a molecular mass of 800 kDa. Using a quadruple time-of-flight (Q-ToF) mass spectrometer and a carefully controlled balance of pressures, conditions were found whereby the GroEL and GroES chaperone assemblies remained intact. For the GroES heptamer a population of monomeric subunits was always observed, consistent with the micromolar Kd measured for this oligomeric complex. The GroEL 14-mer was found to be remarkably stable, but acceleration and collision-induced dissociation of this complex within the collision cell of a Q-ToF mass spectrometer revealed the topology of the interacting subunits5.
This protocol of maintaining many low energy collisions to absorb the excess translational energy of the ions or alternatively inducing their dissociation by high energy collisions has recently been applied to probe the subunit arrangement in a newly described molecular chaperone, MtGimC. MtGimC is an archaeal homologue related to the eukaryotic chaperonin cofactor GimC/prefoldin, involved in the folding of actin and tubulin9. The complex was characterized by first defining the molecular weight of the intact complex10. This had been analysed previously by size exclusion chromatography but the precision afforded by the MS method enabled an unequivocal determination of the stoichiometry. This corresponded to a well-defined hexamer of two a and four b subunits10. Dissociation of the complex within the gas phase was used to probe the quaternary arrangement and two central subunits, both a, and four peripheral b subunits, consistent with these measurements, were proposed. In an extension to this study, a thermally controlled nanoflow device was constructed to monitor the thermal stability of this heat shock complex. The results demonstrated that a significant proportion of the MtGimC hexamer remains intact under low-salt conditions even at 70°C. In addition, it was possible to monitor in real-time the assembly of the MtGimC hexamer from its component subunits. A mixture of the two subunits in a 1:2 ratio of a:b subunits was placed in the nanoflow capillary and, after the dead time of the experiment, spectra were recorded continuously. The mass spectra showed the absence of any intermediates, demonstrating that the assembly process is highly cooperative, leading exclusively to the hexamer.
Despite the relative size and complexity of ribosomes, which in E. coli comprises three large RNA molecules and 55 different proteins, these macromolecules have also been shown to remain intact in the gas phase8. Spectra recorded of the70S particle in the presence of Mg2+ showed that ions from the intact ribosome have m/z values in excess of 20,0008. Through controlled dissociation of this particle in the gas phase it was possible to remove subsets of proteins both individually and as complexes of up to six proteins. Further dissociation into smaller macromolecular complexes and then individual proteins could be induced by subjecting the particles to increasingly energetic gas-phase collisions. The ease with which proteins dissociated from the intact species was found to be related to their known interactions in the ribosome particle. The fact that the 2.3 MDa particle can traverse a mass spectrometer remaining intact until mass measurement enables the sensitivity of the ribosome to a number of external conditions to be examined. For example, lowering the Mg2+ concentration in solution led to dissociation into its component 30S and 50S subunits. The dynamic properties of individual proteins (L10 and L11) within the whole particle has also been addressed. The results suggest that these two proteins are tightly packed within the ribosome structure2. The foundations are now in place to gain insight into the structure of the ribosome during various stages of its dynamic function as well in repsonse to the many theraputic agents that are known to target the ribosome.
References
1) International Human Genome Sequencing Consortium, I. H. G. S. Nature 2001, 409, 860-921.
2)Benjamin, D. R.; Robinson, C. V.; Hendrick, J. P.; Hartl, F. U.; Dobson, C. M. Proc. Natl. Acad. Sci. USA 1998, 93, 7391-7395.
3)Tito, P.; Nettleton, E. J.; Robinson, C. V. J. Mol. Biol. 2000, 303, 267-278.
4)Vis, H.; Heinemann, U.; Dobson, C. M.; Robinson, C. V. J. Am. Chem. Soc. 1998, 120, 6427-6428.
5)Rostom, A. A.; Robinson, C. V. J. Am. Chem. Soc. 1999, 121, 4718-4719.
6)Rostom, A. A.; Sunde, M.; Richardson, S. J.; Schreiber, G.; Jarvis, S.; Bateman, R.; Dobson, C. M.; Robinson, C. V. Proteins Struct. Func. and Genetics 1998, Suppl. 2, 3-11.
7)Tito, M. A.; Tars, K.; Valegard, K.; Hadju, J.; Robinson, C. V. J. Am. Chem. Soc. 2000, 122, 350-351.
8)Rostom, A. A.; Fucini, P.; Benjamin, D. R.; Juenemann, R.; Nierhaus, K. H.; Hartl, F. U.; Dobson, C. M.; Robinson, C. V. Proc. Natl. Acad. Sci. USA 2000, 97, 5185-5190.
9)Leroux, M. R.; Fändrich, M.; Klunker, D.; Siegers, K.; Lupas, A. N.; Brown, J. R.; Schiebel, E.; Dobson, C. M.; Hartl, F. U. EMBO J. 1999, 18, 6730-6743.
10)Fändrich, M.; Tito, M. A.; Leroux, M. R.; Rostom, A. A.; Hartl, F. U.; Dobson, C. M.; Robinson, C. V. Proc. Natl. Acad. Sci. USA 2000, 97, 14151-14155.

iSPOT and MINT: a method and a database dedicated to molecular interactions
Manuela Helmer-Citterich

We are interested in the basic principles of molecular recognition. We have developed the SPOT method for the inference of protein domain specificity and a new database of Molecular INTeractions (MINT).
iSPOT (iSpecificity Prediction Of Target) is a web tool developed to infer the protein-protein interactions between families of peptide recognition modules. The SPOT procedure (Brannetti et al, 2000) utilizes information extracted, for each protein domain family, from position-specific contacts derived from all the available domain/peptide complexes of known structure. The framework of domain/peptide contacts defined on the structure of the complexes is used to build a residue/residue interaction database derived from ligands obtained by panning peptide libraries displayed on filamentous phage.
The method is being optimised with a genetic algorithm and will soon be available on the web. It has been applied to SH3 and PDZ domains and to MHC class I molecules. iSPOT will offer the possibility to answer the following questions: which protein (or peptide) is a possible ligand for a given SH3 (or PDZ or MHC class I molecule)? Which is the best possible SH3 (or PDZ or MHC class I) interacting domain for a given protein/peptide sequence? What residues should one mutate in a domain to lower/increase its affinity for a given peptide ligand?
MINT is a relational database built to collect and integrate protein interaction data in a unique database accessible via a user-friendly web interface. MINT now contains experimentally determined protein-protein interaction data. In the near future, MINT will be enriched with protein-DNA and protein-RNA interaction data. It will also allow the collection of peptide lists selected from a molecular repertoire like those resulting from phage display experiments. We plan to add information about interactions inferred by computational predictive methods.
Curators manually submit the interactions. MINT is an SQL database and the web server is written in an HTML-embedded language named PHP (hypertext preprocessor, derived from PERL).

Mapping protein-protein interaction structurally and globally
Jong Park

In the postgenomic era, one of the most interesting and important challenges is to understand protein interactions on a large scale. The physical interactions between protein domains are fundamental to the workings of a cell: in multidomain polypeptide chains, in multisubunit proteins and in transient complexes between proteins that also exist independently. To study the large-scale patterns and evolution of interactions between protein domains, we view interactions between protein domains in terms of the interactions between structural families of evolutionarily related domains. This allows us to classify 8151 interactions between individual domains in the Protein Data Bank and the yeast Saccharomyces cerevisiae in terms of 664 types of interactions between protein families. At least 51 interactions do not occur in the Protein Data Bank and can be derived only from the yeast data. The map of interactions between protein families has the form of a scale-free network, meaning that most protein families interact with only one or two other families, while a few families are extremely versatile in their interactions and are connected to many families. We observe that almost half of all known families engage in interactions with domains from their own family. We also see that the repertoires of interactions of domains within and between polypeptide chains overlap mostly for two specific types of protein families: enzymes and same-family interactions. This suggests that different types of protein interaction repertoires exist for structural, functional and regulatory reasons.

Prediction of protein interactions from sequence information
Alfonso Valencia

A considerable number of computational methods have been recently developed for the prediction of protein interaction partners based on different aspects of genomic information. Two new methods also address the problem of predicting interaction partners. These are based on the study of corresponding multiple sequence alignments, without the explicit requirement of full genomic information.
The first method1 is based on the study of correlated mutations2 between possible interaction partners. We have previously demonstrated that correlated mutations can be used for the detection of regions of interaction3. Predictions generated with this type of method have been successfully tested in two different experimental systems4-5.
The second method is based on the study of the relation between the phylogenetic trees of the corresponding protein family6. The complementarity of the phylogenetic trees is probably a consequence of the co-evolution of the proteins. Some of the best known cases are the interactions between hormones and their receptors. The proposed method uses the degree of correlation between protein family trees as an indicator of the relation between protein pairs.

References
1.- Pazos & Valencia (2001) Submitted
2.- Gobel et al., (1994) Proteins.
3.- Pazos et al., (1997) J. Mol. Biol.
4.- Gdssler et al (1998) Proc. Natl. Acad. Sci. USA
5.- Azuma et al (1999) J. of Mol. Biol
6.- Pazos & Valencia (2001) Prot. Eng.

Prediction of protein-protein interaction sites in heterocomplexes with neural networks
Rita Casadio

We study the problem of extracting from the three-dimensional structures of protein complexes features relevant for predicting protein-protein interaction sites. Our approach is based on information about properties of evolutionary conservation and surface disposition. The predictor we implement is a neural network system, which uses a cross-validation procedure and allows the correct detection of 73% of the residues involved in protein interactions. Our analysis confirms that the physical properties of interacting surfaces are difficult to distinguish from those of the whole protein surface. However, neural networks trained with a reduced representation of the interacting patch and sequence profile are sufficient to generalise over the different features of the contact patches and to predict whether a residue in the protein surface is or is not in contact. The predictor can significantly complement results from functional genomics and proteomics.

List of participants ...

Some candid snaps from the meeting ...