We investigate the problem of sequential variable selection in a logistic regression application, when the predictor data consist of a high-dimensional array of pairs of predictive measures. In addition to the paired nature of the predictor data, an order structure may be exploited, such that the second component from any pair can only be included in the model, conditional on selection of the primary component. The objective of the analysis is to identify which pairs contain predictive information for the outcome of interest as well as the assessment of the added-value of the secondary pair component for those pairs selected. A sequential adaptation of reversible-jump variable selection is developed to account for the paired nature of the predictor data. We present application in a novel form of high-dimensional high-resolution proteomic data, where the predictor data is expressed in isotopic density clusters, which can be described by both intensity and shape of expression. A post-hoc exploratory data analysis is developed to verify the fit of the reversible-jump model implementation. Simulations are used to assess the performance of the methodology in a controlled setting.