Structure-based virtual screening of FGFR inhibitors: cross-decoys and induced-fit effect

Receptor rearrangement upon ligand binding (induced-fit) constitutes a complicating factor in structure-based virtual screening, as protein flexibility is only partially included in many high-throughput docking programs. The effect of protein structure in these cases is rarely discussed.

Our objective was to analyze this influence on three aspects of automated ligand docking: (i) the successful reproduction of binding modes; (ii) the performance in binding site detection for a series of initial decoys positioned on the protein surface; and (iii) the extent to which the protein conformation biases the enrichment factors and the diversity in scaffold retrieval of a Virtual Screening experiment.

A fibroblast growth factor receptor (FGFR), for which several structures complexed with different inhibitors are publicly available, was selected as a study case. Besides its biological relevance, FGFR is an interesting target because of the structural changes occurring on ligand binding in receptor tyrosine kinases. Three common scoring functions (AUTODOCK, ChemScore, and GoldScore), under different parameter settings, were employed to dock a set of inhibitors of FGFR into these structures.

We show how the choice of one particular protein x-ray structure restricts the docking process to the detection of those compounds that belong to the same chemical series or are similar to the chemotype of the corresponding co-crystallized ligand.