Design of active analogues of a 15-residue peptide using D-optimal design, QSAR and a combinatorial search algorithm

Abstract

This report describes the rational design of novel analogues of a 15-residue antibacterial peptide CAMEL0. A constrained D-optimal design was carried out to derive a training set of 60 analogues. Partial least squares (PLS) models describing quantitative structure-activity relationships (QSARs) were initially derived for the peptides using two published and one novel parameter set. The novel "Design parameters' were based on key structural features identified in hypothetical models of the mechanisms by which peptides interact with cell membranes. In an extension of the PLS method, influence statistics were used to decrease the weighting of compounds having a large effect on model predictions. A combinatorial search algorithm was developed which used PLS models as predictors to select a test set of 39 peptides with high predicted potencies. Within this set, the most potent analogue CAMEL135, which contained seven point mutations from CAMEL0, was identified. For a panel of 24 bacteria, the mean MIC value of CAMEL135 was approximately half of that for CAMEL0. For the parameter sets tested, covariance functions derived from Z-scales gave highest Q2-values for the training set, whilst the model using the the 'Design parameters' gave least error when predicting the activity of the test set. The predictive ability of a third published set of peptide parameters was found to compare favourably with that of the parameters used in the design. Analysis of the PLS models indicates that hydrophobicity and amphipathicity are the most important features influencing activity for this class of compound.