1.0 INTRODUCTION

This literature review is based of the chemistry research project Computational study of protein-ligand interactions in cancer

2.0 COMPUTATIONAL CHEMISTRY

This literature review will be specifically focusing on the area of computational chemistry, alternatively known as theoretical chemistry or molecular modelling.1 Computational chemistry has been defined as the branch, or field, of chemistry that uses computer simulations to assist in solving chemical problems.2 Computational chemistry has its roots in the development of quantum mechanics, in the sense that the foundation was laid with the development of quantum mechanics in the early part of the twentieth century.1 The field we know today as “computational chemistry” is a product of last thirty five1 years of technological advancement, in today’s digital computer and technology centric society.

An important achievement in the field would be Schrödinger’s wave equation and Dirac[A] who said that, using quantum chemistry calculations, it is possible to calculate the energy of molecules.4 Although this method was very time consuming, as it was developed before modern computers. Now that we have computers fast enough, we can do these calculations for small molecules, however not entire proteins.

The computer being the ‘instrument’ of the computational chemist, chemists in the field have taken advantage of the technological advancement to develop and apply new theoretical methods at an astounding pace.1 The field uses theoretical chemistry methods, which are incorporated into efficient computer programs, that calculate the structure and properties of molecules solids. However, the computation chemistry is used more frequently to make chemical predictions, such as the predication of new drug targets or new reactions, which are investigated later experimentally.2 This literature review will be exploring the computational prediction method of docking.

3.0 DOCKING

Docking is defined as a computational method that attempts to predict noncovalent binding of receptor (such as proteins, carbohydrates, nucleic acids, or lipids[B]), and a small molecule (ligand) efficiently. This starts with their unbound structures, structures obtained from molecular dynamics simulations[C], or homology modelling.6Simply, molecular docking is used for computational structures that attempts to predict the structure of the intermolecular complex formed between two or more molecules7: a receptor and a ligand8, or in this review: a protein and a ligand.

Protein–ligand docking aims to predict and rank the structures from the association between a given ligand and a target protein of a known 3D structure. 7,8 Protein–ligand docking occupies a distinct place in the field of docking, due to its applications in medicine.7

After its development in the 1980’s7, docking remains a field of vital research, due to its ability to screen virtual libraries9 of drug-like molecules, in order to obtain leads for further drug development6, drug design8, in addition to being a primary component in drug discovery programs7,8, and protein-function prediction.9

The first stage of docking is pose generation. Pose generation is the prediction of the position, orientation, and conformation of a molecule as docked to the target’s binding site[D]. The second stage, scoring, usually consists in estimating how strongly the docked pose of a ligand binds to the target (the strength is calculated by measures of binding affinity or free energy of binding).9 Prediction of the binding energy is performed by evaluating the most important physical-chemical singularities involved in ligand-receptor binding, including intermolecular interactions, desolvation[E] and entropic effects. Ferreira et al. summarised that, the greater the number of physical-chemical parameters evaluated, the higher the accuracy of the scoring function.11However this statement isn’t necessarily true. Gabel et al. states that machine-learning based scoring functions are insensitive to docking poses, and just describe atomic element counts.12

While there are many comparatively strong and accurate algorithms for pose generation currently available, the inaccuracies in the prediction of binding affinity by scoring functions, continue to be the main restricting factor for the reliability of docking. Despite the thorough research over more than two decades9, the exact prediction of the binding affinities for larger sets of protein-ligand complexes, is still one of the most significant problems in computational chemistry.

3.1 HISTORY OF DOCKING

The earliest reported docking methods were based on the ‘lock-and-key’ assumption proposed by Fischer13, stating that both the ligand and the receptor can be treated as rigid bodies and their tendency to react with another chemical species to form a chemical compound[F], is directly proportionate to a geometric fit between their shapes.14 This really early method is very limited, due to how limited computers were at the time.  Zsoldos et al. states that, say for an average sized ligand with 6 rotatable bonds, would have a total number of bonds[G] poses at 1020. This number alone is so huge, that a “brute force” evaluation of all said posses with a fast scoring function, processing 2000 poses, would take three billion years on a single CPU. Additionally, even using the largest current super computer available in 2007, at the time the article was written, it would still take twenty thousand years to dock a single ligand.

Realistically, treating the ligand and protein as rigid is an impractical assumption, as most of the time said assumption would be incorrect. However, searching every possible combination of the both the ligand and the protein would take, as stated befo