The Fourth Georgia Tech-University of Georgia International Conference in Bioinformatics: in silico Biology, Biological Networks, from Genomics to Epidemiology (November 13-16, 2003, Atlanta, Georgia, USA)

Three previous bioinformatics meetings organized by Georgia Tech were held in Atlanta in 1997, 1999 and 2001 have attracted outstanding researchers from 14 countries around the globe and established this conference as a major academic forum for intensive and open exchange of new ideas. This year we again have invited many rising junior scientists to deliver plenary lectures on their discoveries made in silico, with many of these accomplishments driven by the explosion of genomic and proteomic data. The focus of the 2003 conference, in comparison with the previous ones, has shifted even further to the most complicated aspects of genomics and proteomics manifested by the impressive advances in creating consistent models of networks of cellular processes (http://opal. biology.gatech.edu/GeneMark/conference/). Other changes include the University of Georgia joining Georgia Tech as a co-organizer of the meeting and a move of the conference site to the brand new Georgia Tech Conference Center & Hotel opened in the Midtown area extension of Georgia Tech campus near the Atlanta Arts Center and the city’s historic area. Several papers presented at the conference are included in this current Special Issue of Bioinformatics, with their publication serving a parallel avenue of presentation for conference participants. Seven papers out of seventeen submitted manuscripts successfully passed peer review and were accepted for publication. The paper by Rocco and Critchlow describes a method for scanning the internet to identify web sites with particular functionalities and developing wrappers for automatic queries of those sites. They tested it on sites that perform BLAST searches and showed reasonable success at automatic classification and query submission. An important aspect of the work is the ability of the system to learn the appropriate queries. This is an important and very challenging problem and this paper offers interesting algorithmic ideas. Qian, Lin, Luscombe, Yu and Gerstein address the important and challenging computational problem of defining relationships between transcription factors (TFs) and their target genes. They apply support vector machines to microarray expression data to generate novel predictions from known interactions. Interestingly, their approach does not use DNA sequence information. Existing databases of experimentally confirmed TF-target pairs only contain positive examples, so the authors are careful to construct a balanced set of both positive and negative training data. The resulting database ofin silico predictions of TF-target pairs is a valuable complement to the information that can be derived from chromatin-immunoprecipitation microarray (‘ChIP on a chip’) experiments. Lukashin, Lukashev and Fuchs attack the problems of inference of topology of gene regulation networks that manifests itself in capricious patterns of gene expression in microarray experiments. They verify that genetic networks have many properties, such as connectivity degree distribution, error and attack tolerance and network redundancy, not unlike those of protein–protein interaction networks established a few years earlier. Krause, von Mering and Bork describe a careful large-scale omputational analysis of yeast protein interactome, aiming at identifying yet undiscovered protein complexes and evaluating reliability of known pieces of information. They define a new similarity measure for grouping protein-interaction data and end up with a minimally redundant collection of protein complexes produced by an unsupervised clustering of raw experimental data. The paper by Han and Ju describes a program for rapidly displaying graphs with thousands of edges and nodes, as is common for protein-interaction networks. The method allows a variety of viewing options, such as focusing on specific subgraphs and collapsing parts of the graph to reduce its size. The algorithm is efficient enough to make it practical on the very large graph problems required for whole proteome analysis. They also provide tools for comparing two or more graphs to highlight their similarities and differences. Joel Bader describes an efficient, greedy algorithm that automatically extracts biologically relevant networks from protein–protein interaction data, starting from selected seed proteins (hence the interesting name of the algorithm, SEEDY). The interaction data are notoriously noisy and Bader attempts to use confidence criteria developed in his previous work to enhance the reliability of the produced networks. There is also an interesting attempt to analyze simultaneously protein–protein and genetic interaction data. Gomez, Noble and Rzhetsky address the formidable problem of predicting protein–protein interactions on genome scale by developing an attraction–repulsion learning model. Simply put, evolutionarily conserved features of interacting protein pairs contribute to attraction, whereas the features of pairs that do not interact contribute to repulsion. The authors show that the model works well with known interactions and that including repulsion (not only attraction)