B-cell epitope prediction can enable novel pharmaceutical product development. However, a mechanistically framed consensus\nhas yet to emerge on benchmarking such prediction, thus presenting an opportunity to establish standards of practice that\ncircumvent epistemic inconsistencies of casting the epitope prediction task as a binary-classification problem. As an alternative\nto conventional dichotomous qualitative benchmark data, quantitative dose-response data on antibody-mediated biological effects\nare more meaningful from an information-theoretic perspective in the sense that such effects may be expressed as probabilities\n(e.g., of functional inhibition by antibody) for which the Shannon information entropy (SIE) can be evaluated as a measure of\ninformativeness. Accordingly, half-maximal biological effects (e.g., at median inhibitory concentrations of antibody) correspond\nto maximally informative data while undetectable and maximal biological effects correspond to minimally informative data. This\napplies to benchmarking B-cell epitope prediction for the design of peptide-based immunogens that elicit antipeptide antibodies\nwith functionally relevant cross-reactivity. Presently, the Immune Epitope Database (IEDB) contains relatively few quantitative\ndose-response data on such cross-reactivity. Only a small fraction of these IEDB data is maximally informative, and many more of\nthem are minimally informative (i.e., with zero SIE). Nevertheless, the numerous qualitative data in IEDB suggest how to overcome\nthe paucity of informative benchmark data.
Loading....