01965nam a2200205 a 450000100080000000500110000800800410001910000160006024501420007626001670021830000120038550000390039752012140043665000200165065000130167065000280168365000130171165300180172470000170174210089602020-01-17       2005    bl uuuu     u00u1 u    #d1 aHIGA, R. H.  aPrediction of protein-protein interaction sitesbcomparing neural networks and support vector machines approaches.h[electronic resource]  aIn: X-MEETING; INTERNATIONAL CONFERENCE OF THE AB3C, 1., 2005, [Proceedings...]. [S.l.]: Associação Brasileira de Bioinformática e Biologia Computacionalc2005  ap. 113.  aX-meeting 2005. Presented Posters.  aProtein-protein interaction plays a central role in a number of biological process such as cellular signaling, immune systems and enzyme catalysis. The knowledge of the residues that contribute to the specificity and affinity of protein interactions have important implications for applications such as drug design and analysis of metabolic networks. In silico studies of protein-protein interaction pursue two main objectives: the identification of the network of protein interactions and the identification of the residues involved in these interactions [7]. Recently, a number of studies pursuing the identification of protein-protein interaction sites have been proposed [3][6][7][8][9]. They differ in several aspects such as the information considered to do the prediction, the approach used by the predictors and the database considered to perform the training and testing of the predictors. These differences make difficult to establish a comparison among them. In this study, we use the database of protein-protein complexes used by Fariselli et al. [3] to compare the performance of the two prevalent approaches used to make predictors: Neural Networks -NN [4] and Support Vector Machines - SVM [2].  aNeural networks  aProteins  aSupport vector machines  aProteina  aRedes neurais1 aTOZZI, C. L.