[PMC free article] [PubMed] [Google Scholar] 31. as one of the first actions in the characterization of an antibody to determine its breadth and potency, the NEP server can be used to predict antibody-epitope information at no additional experimental costs. NEP can be accessed on the internet at http://exon.niaid.nih.gov/nep. INTRODUCTION The determination of epitopes targeted by antibodies is useful for understanding virus escape (1), antibody optimization (2,3) and epitope-based design of vaccines (4). Structure determination (by, e.g. X-ray crystallography) of antibodyCantigen complexes can provide epitope information at the atomic level (5), but in many instances, atomic-level complex structures can be challenging to obtain. Additional experimental methods for epitope delineation are also available, although they are characterized with lower accuracy and typically require substantial experimental effort (5C7). Computational methods for epitope prediction have traditionally aimed at predicting antigen residues that could be a PR-104 part of any antibody epitope, and are thus not antibody specific (8C11). More recently, computational methods for antibody-specific epitope prediction (the prediction of the epitope targeted by an antibody of interest) have been developed (7,12C15). Specifically, we and PR-104 others have focused on combining antibodyCantigen neutralization data with antigen sequence information in order to predict residues that may be part of the epitope for antibodies of interest (7,12,13). Antibody neutralization assays, which measure the reduction of viral infectivity mediated by antibody, are often performed as one of the first actions PR-104 in the characterization of an antibody to determine its breadth and potency. Previously, we developed a neutralization-based epitope prediction method that is applicable to antigens that exhibit substantial sequence diversity, such as human immunodeficiency virus 1 (HIV-1) and influenza (7). The algorithm, named NEP for neutralization-based epitope prediction, is based on the premise that sequence variation PR-104 of epitope residues is usually more likely to have an effect on antibody neutralization than variation of non-epitope residues. For each antigen residue position, NEP estimates the association PR-104 between sequence variation and changes in antibody neutralization for a given set of diverse viral strains. A structure of the unbound antigen, if available, can be used for further improvement in the prediction accuracy. NEP has been validated on a set of HIV-1 antibodies targeting a number of different epitopes around the virus: both for retrospective epitope prediction [for 19 antibodies with known complex structures, with a true positive (TP) rate of 0.403 at a 0.05 false positive (FP) rate level] and for prospective epitope prediction (for HIV-1 antibody 8ANC195, with a previously uncharacterized epitope) (7). Comparable methods for neutralization-based antibody-epitope prediction were also described recently (12,13). In this paper, we describe the implementation of the NEP algorithm as a web-based server. The NEP server allows the user to predict the epitope for an antibody by using antigen sequence alignment for diverse viral strains, antibodyCantigen neutralization data over the same set of strains and (optionally) a structure of the unbound antigen. The results can be downloaded or viewed interactively in a web browser via the JSmol Applet. NEP is the first publicly available server for antibody-epitope prediction using antigen structure and neutralization data of diverse viral strains. MATERIALS AND METHODS Epitope-prediction algorithm For each residue position in an antigen, the NEP algorithm computes a mutual information score (16) between amino acid variation CD5 at that position and changes in sensitivity to virus neutralization. Two method variants were implemented in this server, based on our previously published study (7). Neutralization + sequence: each antigen residue is usually ranked by the normalized mutual information between amino acid types and neutralization IC50 values. The score for residue is usually defined as follows: where is usually a variable that covers the possible amino acid types at position (the 20 natural amino acid types and a gap in the sequence alignment). is usually a binary variable defined by a user-specified IC50 cutoff value that divides strains into a resistant and a.