Prediction of immunogenicity of therapeutic proteins: validity of computational tools.

Most protein therapeutics have the potential to induce undesirable immune responses in patients. Many patients develop anti-therapeutic antibodies, which can affect the safety and efficacy of the therapeutic protein, particularly if the response is neutralizing. There are a variety of factors that influence the immunogenicity of protein therapeutics and, in particular, the presence of B- and T-cell epitopes is considered to be of importance. In silico tools to identify the location of both B- and T-cell epitopes and to assess the potential for immunogenicity have been developed, and such tools provide an alternative to more complex in vitro or in vivo immunogenicity assays. This article reviews computational epitope prediction methods and also the use of manually curated databases containing experimentally derived epitope data. However, due to the complexities of the molecular interactions involved in epitope recognition by the immune system, the heterogeneity of key proteins in human populations and the adaptive nature of the immune response, in silico methods have not yet achieved a level of accuracy that enables them to be used as stand-alone tools for predicting clinical immunogenicity. Computational methods, particularly with regard to T-cell epitopes, only consider a limited number of events in the process of epitope formation and therefore routinely over-predict the number of epitopes within a molecule. Epitope databases such as the Immune Epitope Database (IEDB) and the proprietary T Cell Epitope Database (TCED) have reached a size and level of organization that increases their utility; however, they are not exhaustive. These methods have greatest utility as an adjunct to in vitro assays where they can be used either to reduce the amount and complexity of the in vitro screening, or they can be used as tools to analyze the sequence of the identified epitope in order to locate amino acids critical for its properties.