more similar AUC values)
more similar AUC values). Given the differences observed Cabergoline between nave and expert BNs and the better performance of expert BNs under both Ab+ and Ab datasets; two expert BNs (tailored for the Ab+, Ab datasets) that combine all seven measures were selected as default predictors (Figure 2A for schematic representation of the BNs architectures), and used in all subsequent analyses. == Goat polyclonal to IgG (H+L)(FITC) Figure 2. either missing or not reliable (e.g. evolutionary information). == INTRODUCTION == In order to fulfill their function, proteins must interact with one another and with other biomolecules, thus offering an enormous potential for the discovery of novel therapeutic agents able to act either as antagonist or agonist of proteinprotein interactions (PPIs). Crystallographic studies have shown that proteins interact through large, typically 150300 nm2(1,2) [60 nm2is the minimum area required to make a water-tight seal around a critical set of energetically favorable interactions (3)], and relatively featurelessness surfaces. Given these large interfacial areas, one school of thought considers that small-molecule inhibitors require binding-pockets or clefts at the proteinprotein interface, in order to attain the required affinities (4). However, as discussed by Wells and McLendon (5), this and a number of other objections to target the disruption of proteinprotein interfaces have recently been challenged by new data [reviewed by Yin and Hamilton (6) and Wells and McLendon (5)]. Many of these successes have been aided by the Cabergoline realization, following Cabergoline the pioneering work of Clackson and Wells (7), that the binding energy for many proteinprotein associations can be ascribed to a small and complementary set of interfacial residuesahot spotof binding energy surrounded by weaker interactions providing specificity. Protein interaction interfaces include many intermolecular contacts, involving 1030 side chains on average from each protein. However, a typical hot spot accounts for less that half of the contact surface (5). The experimental detection of residues located in hot spots can be achieved by Alanine scanning mutagenesis (8), Alanine shaving (9) and residue grafting (9). These techniques are very time consuming, labor-intensive and involve a high economic cost. Therefore, there Cabergoline is a great interest in computational tools that can predict critical residues with high accuracy and thus, be used to aid and complement experimental efforts. Several computational tools have been described in the past. These can be categorized in three groups depending on the information used for the prediction. Thus, methods that estimated the energetic contribution of each individual residue to the global binding energy, so called computational Alanine scanning (1012), or by free-energy decomposition (13), have been proposed. Other methods exploit the structural features that are characteristic of hot spots such as solvent accessibility (1416), atomic contacts (17), structural conservation (18), restricted mobility (19), and location in the interaction patch (20). Finally, a third type of method accounts for evolutionary information such as sequence conservation (15,2123), sequence environment and evolutionary profile (24), and pattern mining (25). While theses attributes are informative, it has been found that individually they cannot unambiguously define hot spots (26). In this article, we Cabergoline present a novel probabilistic method, Presaging Critical Residues in Protein interfaces (PCRPi), that combines these three main sources of information, namely energetic, structural and evolutionary determinants by using Bayesian Networks (BNs) (27,28). Many applications in Bioinformatics and Computational Biology use BNs (2935) offering several clear advantages over alternative modeling approaches as it has been described elsewhere (36). We have developed and extensively benchmarked several BNs in a dataset of experimentally confirmed hot spots residues. Results show that PCRPi delivers robust and reliable predictions with a high.