IMPRECO: A Tool for Improving the Prediction of Protein Complexes Mario Cannataro, Pietro Hiram Guzzi, and Pierangelo Veltri Bioinformatics Laboratory, Experimental Medicine Department, University Magna Graecia, Catanzaro, Italy {cannataro,hguzzi,veltri}@unicz.it Abstract. Proteins interact among them and different interactions form a very huge number of possible combinations representable as protein to protein interaction (PPI) networks that are mapped into graph struc- tures. The interest in analyzing PPI networks is related to the possibil- ity of predicting PPI properties, starting from a set of known proteins interacting among each other. For example, predicting the configuration of a subset of nodes in a graph (representing a PPI network), allows to study the generation of protein complexes. Nevertheless, due to the huge number of possible configurations of protein interactions, automatic based computation tools are required. Available prediction tools are able to analyze and predict possible combinations of proteins in a PPI net- work which have biological meanings. Once obtained, the protein inter- actions are analyzed with respect to biological meanings representing quality measures. Nevertheless, such tools strictly depend on input con- figuration and require biological validation. In this paper we propose a new prediction tool based on integration of different prediction results obtained from available tools. The proposed integration approach has been implemented in an on line available tool, IMPRECO standing for IMproving PREdiction of COmplexes. IMPRECO has been tested on publicly available datasets, with satisfiable results. 1 Introduction The interactions of proteins within a cell are very huge and frequent. They inter- act composing a very broad network of interactions, also known as interactome. If two or more proteins interact for a long time forming a stable association, their interaction is known as protein complex. Interactomics study focuses currently: (i) on the determination of all possible interactions and (ii) on the identifica- tion of a meaningful subset of interactions. Due to the high number of proteins within a cell, manual analysis of proteins interactions is unfeasible, so the need to investigate interactions with computational methods arises [1]. We focus on interactomics as the study of Protein-Protein Interaction (PPI) as biochemical reaction among proteins, as well as the study of protein complexes. The most natural way to model PPIs network is by using graphs [2], where proteins are represented as nodes and interactions as edges linking them. The M. Bubak et al. (Eds.): ICCS 2008, Part III, LNCS 5103, pp. 148–157, 2008. c  Springer-Verlag Berlin Heidelberg 2008