AFMoC enhances predictivity of 3D QSAR: a case study with DOXP-reductoisomerase.

We present structure-activity relationships for 43 inhibitors of 1-deoxyxylulose-5-phosphate (DOXP)-reductoisomerase, derived from protein-based docking, ligand-based 3D QSAR, and a combination of both approaches as realized by AFMoC (adaptation of fields for molecular comparison). DOXP-reductoisomerase (DXR) is a key enzyme of the non-mevalonate pathway for isoprenoid building blocks. This target has been characterized as having potential in the treatment of malaria with fosmidomycin, an established DXR inhibitor, presently in clinical trials. As part of an effort to optimize the properties of fosmidomycin, analogues have been synthesized and tested to gain further insights into the primary determinants of structural affinity. These data have been used to create a predictive model for DXR inhibition applying data taken from several DXR X-ray structures. These structures still leave the active fosmidomycin conformation and detailed reaction mechanism undetermined. This together with the small inhibitor data set provides a major challenge for presently available docking programs and 3D QSAR tools. To overcome these difficulties we have applied the AFMoC protocol. AFMoC makes more efficient use of available modeling data by tailoring DrugScore knowledge-based potentials specifically toward a given protein using inhibitor potency data. While 3D QSAR methods achieved valid models which lack predictivity, AFMoC was found to provide superior performance, based both on cross-validation runs as well as for inhibitors not considered in the training set. In particular, AFMoC's ability to gradually transform between generally applicable unadapted interaction fields to case specifically adapted ones proved to be of major importance. Using 50% tailored fields was found to permit the precise prediction of binding affinities for related ligands without losing the capability to estimate the affinities of structurally distinct inhibitors.