Empirical Scoring Functions for Structure-Based Virtual Screening: Applications, Critical Aspects, and Challenges

Abstract

Structure-based virtual screening (VS) is a widely used approach that employs the knowledge of the three-dimensional structure of the target of interest in the design of new lead compounds from large-scale molecular docking experiments. Through the prediction of the binding mode and affinity of a small molecule within the binding site of the target of interest, it is possible to understand important properties related to the binding process. Empirical scoring functions are widely used for pose and affinity prediction. Although pose prediction is performed with satisfactory accuracy, the correct prediction of binding affinity is still a challenging task and crucial for the success of structure-based VS experiments. There are several efforts in distinct fronts to develop even more sophisticated and accurate models for filtering and ranking large libraries of compounds. This paper will cover some recent successful applications and methodological advances, including strategies to explore the ligand entropy and solvent effects, training with sophisticated machine-learning techniques, and the use of quantum mechanics. Particular emphasis will be given to the discussion of critical aspects and further directions for the development of more accurate empirical scoring functions.

From machine learning to deep learning: Advances in scoring functions for protein–ligand docking

“…Empirical SFs adopt similar functional forms used by FF‐based SFs except that some additional terms, such as the counts of the rotatable bonds of the ligands, solvent accessibility surface area (SASA), etc., are added to provide better characterization of protein–ligand interactions. Besides, unlike the identical weights for all terms in an FF‐based SF, each term in an empirical SF has its own weight, which can be obtained by linearly fitting the scoring terms to experimental binding affinities . Empirical SFs account for the largest proportion of all the SFs, and many well‐known SFs belong to this class, such as Autodock Vina, GlideScore, ChemScore, and X‐Score .…”

CBSF: A New Empirical Scoring Function for Docking Parameterized by Weights of Neural Network

“…NNScore (Durrant et al, 2013;Durrant & Mccammon, 2011;Durrant & McCammon, 2010), SFCscore (Sotri er et al, 2008) and RF-Score (Ballester & Mitchell, 2010;Wójcikowski et al, 2017). A brief review of such methods can be found in (Guedes, Pereira, & Dardenne, 2018). Furthermore, some of these scoring functions, such as RF-Score and NNScore, were developed by means of the machine-learning techniques.…”

“…An appealing feature of our scoring functions is their simplicity, as they are not overloaded with supplementary terms present in other empirical scoring functions (counting hydrogen bonds and rotational bonds, taking into account partial charges or electrostatic potentials and so on (Baek, Shin, Chung, & Seok, 2017;Guedes et al, 2018;Jain, 1996;R. Wang, Lai, & Wang, 2002)) and do not contain terms from third-party scoring functions (Pereira, Ca arena, & Dos Santos, 2016;Tanchuk, Tanin, Vovk, & Poda, 2016;C.…”

High-Throughput Docking Using Quantum Mechanical Scoring

“…In high-throughput docking (HTD), where the protein is usually considered rigid or with very few degrees of freedom, and thousands to millions of molecules from a chemical library are screened, the goal is to generate a sub-library enriched with potential ligands, which will be prioritized for further experimental evaluation. In HTD, two different stages can be distinguished: the assessment of the best binding mode(s) of each molecule of the library ("docking stage"), and, on each in silico generated protein-small-molecule complex, the calculation of a score reflective of the likelihood that the molecule will actually bind to the target ("scoring stage") (Cavasotto and Orry, 2007;Guedes et al, 2018). In the docking stage, the docking energy (DE) is used to select, for each molecule, the lowest-energy pose(s) from a large amount of conformations generated, while the docking score (DS) is generally calculated as a fast approximation to the binding free energy ( G binding ), and depends on several factors, such as the energy representation of the system, the model used to represent the aqueous environment and the consideration of explicit water molecules within the active site (Cozzini et al, 2006;Amadasi et al, 2008), and the degree of consideration of receptor flexibility (Cavasotto and Singh, 2008;Spyrakis et al, 2011;Spyrakis and Cavasotto, 2015).…”

“…Only very recently, a QM-based approach was presented displaying a very good performance on discriminating ligands and decoys on a single system (heat shock protein 90, HSP90) (Eyrilmez et al, 2019). In fact, the development of fast yet accurate docking scoring functions still constitutes an area of active research (Cavasotto, 2012;Guedes et al, 2018). Moreover, the blind challenges ran by the Drug Design Data Resource (D3R) for ligand-pose and affinity prediction in 2015 (Gathiaka et al, 2016), 2016 (Gaieb et al, 2018), and 2018 (Gaieb et al, 2019), have shown the importance of method development and benchmarking in pose prediction and binding affinity ranking of ligands.…”