Our method’s overall performance is evaluated based on these 59 complexes by using leave-one-out mix validation. == Epitopes and paratopes in our teaching data == Given an antigen-antibody PDB [21] complex, a distance threshold of 4 is used to determine the epitope residues and paratope residues from your contact residues. of interacting residue pairs. As the association itself does not include the neighborhood information on the primary sequence, residues’ cooperativity and relative composition are then used to enhance our method. Evaluation carried out on a benchmark data set demonstrates Cytosine the proposed method produces very good performance in terms of accuracy. After compared with additional two structure-based B-cell epitope prediction methods, results show the proposed method is definitely competitive to, sometimes even better than, the structure-based methods which have much smaller applicability scope. == Conclusions == The proposed method leads to a new way of identifying B-cell epitopes. Besides, this antibody-specified epitope prediction can provide more exact and helpful info for wet-lab experiments. == Background == Secreted antibody takes on a critical part in humoral immune reactions. These antibodies guard the normal cellules or cells from invaders and infected self Cytosine cells by neutralizing them through interacting with NGF the pathogenic providers. Subsequently, the neutralized cells are eliminated by scavenger cells, such as macrophage. During this process, antibody interacting with antigen Cytosine is definitely a fundamental and essential step in immune response. Hence, identifying the set of residues within antigen which are recognized by a specific antibody is definitely pivotal for understanding the mechanism behind antibody-antigen connection. Consequently, this knowledge in antibody-antigen connection will shed fresh light on vaccine design, disease therapy and so on [1]. The small set of residues within antigen sequence that can be identified by antibody is named as epitope [2]. Epitopes can be classified into two types: continuous and discontinuous [3]. A continuous/linear epitope is definitely a stretch of consecutive residues in the primary sequence that can bind to a specific antibody, while a discontinuous/conformational epitope is definitely comprised of stretch of residues that are far away from each other in the primary sequence but are brought to spatial proximity as a result of polypeptide folding. Accordingly, a paratope is the portion of residues within antibody that interact with the related antigen. Due to the importance of identifying epitopes within antigen, many experts possess dedicated themselves to this area. Intensive efforts have been made to forecast epitopes based on physico-chemical properties of antigen interacting with antibody, particularly focus on linear epitope prediction due to its relatively lower difficulty. For example, the hydrophilicity level information of the individual amino acids [4] was used by Parkeret al., and the flexibility of epitope sequences was used by Karpluset al., to forecast linear epitopes [5,6]. The relative accessible surface area of each residue and the three dimensional structure info on antigen sequences were combined together by Kulkarni-Kaleet al.to predict the conformational epitopes [7]; and the exposure area, amino acids statistical significance and spatial information were utilized by Andersenet al.to predict the conformational epitopes as well [8]. Besides, other features, such as polarity [9] and antigenic propensity [10] were also considered to cope with this prediction problem. However, the prediction results are far from satisfied. For example, the performance of the propensity scale based methods are only slightly better than the random projection method [11,12], and it does not improve much even after structural information is usually added [13]. Several reasons can be used to explain this intractable problem. First of all, epitopes highly depend on specific type of antibody that can recognize them, and most of the antigen surface residues may be antigenic when it is exposed to different circumstances. Therefore, epitope prediction based on binary classification may not reveal the biological reality [14]. Unfortunately, all the aforementioned methods only focused on antigens and overlooked the Cytosine antibody-antigen relationships. Second, antigen itself is very complicated, and it can range from a few residues to a very large protein. However, epitope residues only take a small portion of the entire antigen residues, thus it is an anomaly detection problem. Third, although the residues that constitute the epitopes are rare, they should cooperate with each other rather than appear independently [15]. However, all the properties that have been used are residue-independent, and only a few methods consider the effect from the neighborhood residues [16]. To overcome these obstacles for a better understanding of antibody-antigen conversation, we propose a novel method to predict epitopes based on associations between antibody-antigen interactions. The intuitive reasons of identifying epitope by associations are: (i) associations not only address the contextual dependence between antibody and antigen, but also reveal the spatial relation within the contact residues; (ii) epitope prediction is very difficult while paratope identification is much easier, therefore linking antibody and antigen together will bridge over the gap; (iii) many research findings have corroborated that paratope-epitope conversation has a tight complementary relationship [17,18], thus it is plausible to link.