The search for novel leads is a critical step in the drug discovery process. Computational approaches to identify new lead molecules have focused on discovering complete ligands by evaluating the binding affinity of a large number of candidates, a task of considerable complexity. A new computational method is introduced in this work based on the premise that the primary molecular recognition event in the protein binding site may be accomplished by small core fragments that serve as molecular anchors, providing a structurally stable platform that can be subsequently tailored into complete ligands. To fulfill its role, we show that an effective molecular anchor must meet both the thermodynamic requirement of relative energetic stability of a single binding mode and its consistent kinetic accessibility, which may be measured by the structural consensus of multiple docking simulations. From a large number of candidates, this technique is able to identify known core fragments responsible for primary recognition by the FK506 binding protein (FKBP-12), along with a diverse repertoire ofnovel molecular cores. By contrast, absolute energetic criteria for selecting molecular anchors are found to be promiscuous. A relationship between a minimum frustration principle of binding energy landscapes and receptor-specific molecular anchors in their role as "recognition nuclei" is established, thereby unraveling a mechanism of lead discovery and providing a practical route to receptor-biased computational combinatorial chemistry.
Concepts of Lead DiscoveryUnderstanding the principles of molecular recognition is a long-standing problem in molecular biology (1, 2). Methods to discover novel lead molecules and assess their binding affinity and receptor specificity are of considerable utility in receptor structure-based drug design (3-7). Approaches that computationally screen data bases for complete inhibitors (8-15) are required to both assess the structure of the bound ligandprotein complex and reliably estimate the binding free energy. Each candidate is ranked on the basis of the best energetic orientation, evaluated by criteria such as packing density, electrostatic complementarity, molecular mechanics force field energy, and empirical solvation free energy corrections (8-11). These scoring functions, which are approximate for binding affinity, often are unable to distinguish between the structures of native and nonnative ligand-protein complexes. A detailed description of ligand-protein association involves a delicate balance between van der Waals and electrostatic interactions, solvation effects, and conformational entropy, resulting in a highly frustrated energy landscape of molecular recognition with many energetically similar but structurally different local minima, which makes reliable structure prediction difficult (16,17). Hence, even if a thermodynamically complete and accurate energy function suitable for rigorous predictions of binding affinity were available, it would not solve the equally important problem o...