Screening of benzamidine-based thrombin inhibitors via a linear interaction energy in continuum electrostatics model

Abstract

A series of 27 benzamidine inhibitors covering a wide range of biological activity and chemical diversity was analysed to derive a Linear Interaction Energy in Continuum Electrostatics (LIECE) model for analysing the thrombin inhibitory activity. The main interactions occurring at the thrombin binding site and the preferred binding conformations of inhibitors were explicitly biased by including into the LIECE model 10 compounds extracted from X-ray solved thrombin-inhibitor complexes available from the Protein Data Bank (PDB). Supported by a robust statistics (r(2) = 0.698; q(2) = 0.662), the LIECE model was successful in predicting the inhibitory activity for about 76% of compounds (r (ext) (2) > or = 0.600) from a larger external test set encompassing 88 known thrombin inhibitors and, more importantly, in retrieving, at high sensitivity and with better performance than docking and shape-based methods, active compounds from a thrombin combinatorial library of 10240 mimetic chemical products. The herein proposed LIECE model has the potential for successfully driving the design of novel thrombin inhibitors with benzamidine and/or benzamidine-like chemical structure.

Estimation of the Binding Free Energy by Linear Interaction Energy Models

“…The approach was also successfully applied to a series of virtual screening experiments which finally lead to the discovery of cellpermeable β-secretase inhibitors with inhibitory activity in the low-micromolar range, as well as a potent and selective nanomolar inhibitor of the tyrosine kinase erythropoietin producing human heptocellular carcinoma receptor B4 (EphB4) [46][47][48][49][50]. The two cited targets were used for deriving a two-parameter model [13]: (12) and a three-parameter model [25]:…”

“…In this regard, the large number of computational experiments performed with LIECE remarked the extreme sensibility of this technique even to small conformational changes [50], as well as its obvious tendency to give better performance in interpolating rather than in extrapolating new external data [53]. It is worthy saying that LIECE model tends also to be affected by the presence of false positives (due to misleading interaction energy values) which, however, can be easily filtered out by a preliminary selection of docking conformations by means of filtering criteria based on energy efficiency or, alternatively, monitoring the occurrence of specific ligand-protein interactions in the obtained ligand binding modes [51].…”

“…A recent application of a LIECE model was also developed and validated on a series of benzamidine-based inhibitors of human thrombin [50]. In this example, a large number of observations (n obs = 27), primarily extracted from X-ray complexes and further complemented with other benzamidine thrombin inhibitors, was successfully used to derive a model with satisfactory ability in prediction (over an almost three times larger external test set, n test = 88).…”

“…b) Histogram indicating the number of actives discovered at different early percentages of the screened combinatorial library. This figure is taken from Nicolotti et al[50].…”