The Challenge of Affinity Prediction: Scoring Functions for Structure‐Based Virtual Screening

Abstract

The purpose of virtual screening is to provide hits with novel chemical structure as a starting point for lead optimization after careful assessment of their optimization potential [1].Structure-based virtual screening works on the premise of shape and interaction complementarity of the ligand with its putative target binding site, for which experimental 3D structures or validated homology models are available. Using docking algorithms combined in various ways with other filters (e.g., drug-likeness, physiochemical properties, binding site constraints, pharmacophore models, and molecular similarity), compounds are selected that have a high likelihood to interact with the target. To this end, the first step is to generate a proposed binding conformation (pose) using validated docking engines. These poses need to be sorted to identify the conformation that is closest to an experimental binding pose. Subsequently, the free energy of binding (DG ) for a particular conformation of a single compound interacting with the protein has to be estimated to rank the ligands and select those with the highest chance for biological activity. Approaches providing such estimates of affinity are, thus, of utmost importance in virtual screening. Normally, this pose ranking and affinity estimation is carried out with the help of scoring functions. Scoring functions are approximate mathematical models for predicting the strength of the noncovalent interaction between two binding partners. Such functions also guide the conformational and orientational search of the ligand in the binding cavity during the docking process.This chapter focuses on different aspects of scoring functions for affinity prediction and pose ranking. First, we will outline the physicochemical background of current scoring approaches. As different scoring functions were recently reviewed [2-4] and their performance was extensively compared [5-8], we will provide only a brief overview of the different scoring function classes and focus on unique features and applications. Hence, we outline novel strategies and computational tools with an impact on improving the success rate in scoring protein-ligand interactions.

Structure-Based Discovery of Small Molecules Binding to RNA

“…The main difference between the different conformational search algorithms is how the degrees of freedom of the ligand are treated while the scoring functions differ based on which underlying theory they are built on. Docking is routinely used for ligand discovery for protein targets [28] and there are several comprehensive textbooks and review articles about docking in general (e.g., [26,29,30]). Therefore, we will only give a concise overview and focus instead on what is specific for RNA-ligand docking compared to the more established protein-ligand docking.…”

“…2). There is a wide variety of different techniques available which differ in speed and accuracy [26,30]. In the context of virtual screening, typically only fast scoring functions are applied.…”

“…they contain terms for van der Waals (vdW) and electrostatic interactions [26,30]. They can also include terms for the intramolecular (strain) energy of the ligand.…”

“…Empirical scoring functions include several terms describing properties that are known to be important for forming a noncovalent complex whereas each term consists of a descriptor multiplied by a weighting coefficient [26,30]. The descriptors typically take into account polar interactions such as hydrogen bonds and ionic interactions, apolar interactions such as hydrophobic and aromatic interactions and loss of ligand flexibility.…”

“…Knowledge-based scoring functions attempt to capture the knowledge about receptor-ligand binding that is implicitly stored in large structural databases [26,30]. The underlying assumption is that close intermolecular interactions between certain atom types or functional groups that occur more frequently than expected by random are likely to be energetically favourable and should therefore enhance the binding affinity.…”