Gábor Iván

When the Web meets the cell

MOTIVATION Enormous and constantly increasing quantity of biological information is represented in metabolic and in protein interaction network databases. Most of these data are freely accessible through large public depositories. The robust analysis of these resources needs novel technologies, being developed today. RESULTS Here we demonstrate a technique, originating from the PageRank computation for the World Wide Web, for analyzing large interaction networks. The method is fast, scalable and robust, and its capabilities are demonstrated on metabolic network data of the tuberculosis bacterium and the proteomics analysis of the blood of melanoma patients. AVAILABILITY The Perl script for computing the personalized PageRank in protein networks is available for non-profit research applications (together with sample input files) at the address: http://uratim.com/pp.zip.

Equal Opportunity for Low-Degree Network Nodes: A PageRank-Based Method for Protein Target Identification in Metabolic Graphs

Biological network data, such as metabolic-, signaling- or physical interaction graphs of proteins are increasingly available in public repositories for important species. Tools for the quantitative analysis of these networks are being developed today. Protein network-based drug target identification methods usually return protein hubs with large degrees in the networks as potentially important targets. Some known, important protein targets, however, are not hubs at all, and perturbing protein hubs in these networks may have several unwanted physiological effects, due to their interaction with numerous partners. Here, we show a novel method applicable in networks with directed edges (such as metabolic networks) that compensates for the low degree (non-hub) vertices in the network, and identifies important nodes, regardless of their hub properties. Our method computes the PageRank for the nodes of the network, and divides the PageRank by the in-degree (i.e., the number of incoming edges) of the node. This quotient is the same in all nodes in an undirected graph (even for large- and low-degree nodes, that is, for hubs and non-hubs as well), but may differ significantly from node to node in directed graphs. We suggest to assign importance to non-hub nodes with large PageRank/in-degree quotient. Consequently, our method gives high scores to nodes with large PageRank, relative to their degrees: therefore non-hub important nodes can easily be identified in large networks. We demonstrate that these relatively high PageRank scores have biological relevance: the method correctly finds numerous already validated drug targets in distinct organisms (Mycobacterium tuberculosis, Plasmodium falciparum and MRSA Staphylococcus aureus), and consequently, it may suggest new possible protein targets as well. Additionally, our scoring method was not chosen arbitrarily: its value for all nodes of all undirected graphs is constant; therefore its high value captures importance in the directed edge structure of the graph.

A hybrid clustering of protein binding sites.

The Protein Data Bank contains the description of approximately 27 000 protein–ligand binding sites. Most of the ligands at these sites are biologically active small molecules, affecting the biological function of the protein. The classification of their binding sites may lead to relevant results in drug discovery and design. Clusters of similar binding sites were created here by a hybrid, sequence and spatial structure‐based approach, using the OPTICS clustering algorithm. A dissimilarity measure was defined: a distance function on the amino acid sequences of the binding sites. All the binding sites were clustered in the Protein Data Bank according to this distance function, and it was found that the clusters characterized well the Enzyme Commission numbers of the entries. The results, carefully color coded by the Enzyme Commission numbers of the proteins, containing the 20 967 binding sites clustered, are available as html files in three parts at http://pitgroup.org/seqclust/.

Being a binding site: Characterizing residue composition of binding sites on proteins

The Protein Data Bank contains the description of more than 45,000 three-dimensional protein and nucleic-acid structures today. Started to exist as the computer-readable depository of crystallographic data complementing printed articles, the proper interpretation of the content of the individual files in the PDB still frequently needs the detailed information found in the citing publication. This fact implies that the fully automatic processing of the whole PDB is a very hard task. We first cleaned and re-structured the PDB data, then analyzed the residue composition of the binding sites in the whole PDB for frequency and for hidden association rules. Main results of the paper: (i) the cleaning and repairing algorithm (ii) redundancy elimination from the data (iii) application of association rule mining to the cleaned non-redundant data set. We have found numerous significant relations of the residue-composition of the ligand binding sites on protein surfaces, summarized in two figures. One of the classical data-mining methods for exploring implication-rules, the association-rule mining, is capable to find previously unknown residue-set preferences of bind ligands on protein surfaces. Since protein-ligand binding is a key step in enzymatic mechanisms and in drug discovery, these uncovered preferences in the study of more than 19,500 binding sites may help in identifying new binding protein-ligand pairs.

On dimension reduction of clustering results in structural bioinformatics

OPTICS is a density-based clustering algorithm that performs well in a wide variety of applications. For a set of input objects, the algorithm creates a reachability plot that can either be used to produce cluster membership assignments, or interpreted itself as an expressive two-dimensional representation of the clustering structure of the input set, even if the input set is embedded in higher dimensions. The focus of this work is a visualization method that can be applied for comparing two, independent hierarchical clusterings by assigning colors to all entries of the input database. We give two applications related to macromolecular structural properties: the first is a sequence-based clustering of the SwissProt database that is evaluated using NCBI taxonomy identifiers, and the second application involves clustering locations of specific atoms in the serine protease enzyme family-and the clusters are evaluated using SCOP structural classifications.

Four spatial points that define enzyme families

The catalytic properties of enzymes, containing the Asp-His-Ser triads are deeply investigated for a long time. Serine endopeptidases, cutinases, acetylcholinesterases, cellulases, among other enzymes, contain these triads. We found that solely the geometric properties of just four points in the spatial structure of these enzymes are characteristic to their family.

Cysteine and tryptophan anomalies found when scanning all the binding sites in the Protein Data Bank

The Protein Data Bank (PDB) is one of the richest sources of structural biological information in the World. It started to exist as the computer-readable depository of crystallographic data complementing printed papers. The proper interpretation of the content of the individual files in the PDB still needs the detailed information found in the citing publication. An advanced graph theoretical method is presented here for automatically repairing, re-organising and re-structuring PDB data yielding the identification of all the protein-ligand complexes and all the binding sites in the PDB. As an application, we identified strong cysteine and tryptophan irregularities in the data.

An optimized distance function for comparison of protein binding sites

An important field of application of string processing algorithms is the comparison of protein or nucleotide sequences. In this paper we present an algorithm capable of determining the dissimilarity (distance) of protein sequences originating from protein binding sites found in the RS-PDB database that is a repaired and cleaned version of the publicly available Protein Data Bank (PDB). The special way of construction of these protein sequences enabled us to optimize the algorithm, achieving runtimes several times faster than the unoptimized approach. One example the algorithm proposed in this paper can be useful for is searching conserved sequences in protein chains.