This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
Experimental techniques for the determination of three-dimensional structure proteins crystallographic and magnetic resonance protocols have contributed for the deposition of over 12,000 protein structures in the Protein Data Bank. Although the number of available experimental protocols is large and improving rapidly, the determination of the structure of all detected protein-molecule interactions experimentally at high resolution is still an impossible task. Hence, reliable computational methods are of increasing importance. Protein docking involves the calculation of the three-dimensional structure of a protein-molecule complex. The molecule can be another protein, a small peptide or other small molecule (e. g. ligand). Ligand docking is nowadays of great importance in the drug discovery area, with great scientific and commercial interest. The main goal of protein docking is to predict how a pair of molecules interact, predicting accurate ligand poses and evaluating the main existing interactions. It should be able to adequately search the conformational available space and calculate the free energy of each conformation to identify the minimum energy conformation.
Goals and Steps
Protein docking requires the structures of the elements that form the complex and aims to predict correctly the binding site on the target, the orientation of the ligand and the conformation of both. At the end, a rank of possible docking poses based on estimated binding affinities or estimated free energies of binding is given.
To successfully predict a target/ligand complex three steps are needed: (1) have accurate structures of the molecules involved in the interaction, (2) location of the binding site, and (3) determination of the binding mode and evaluation.
According to Gray, the best docking targets are single-domain small proteins with known monomer structures, with experimentally-determined micromolar or better binding affinity, and minimal backbone conformational change after binding. The docking problem becomes more complicated when one of the structures undergoes significant conformational changes upon binding , for proteins whose structure was solves by homology modeling or for molecules with high degrees of freedom. However there have been reported successful docking results with modeled targets.
The second step depends on the algorithm behind the docking software. Some of the used algorithms will be described further on. The hypothesis behind docking predictions is that the structure of a complex is the lowest free energy state that is accessible to the system. In Nature a protein-molecule complex change their conformations to become more compatible to one another, shifting two equilibriums progressively from less compatible to most compatible conformations for both, located at the local minimum of their potential energy surfaces. However ligands do not always adopt their lowest potential energy conformations when binding to their protein targets. Combining these two facts, the results can be influenced by the previous knowledge of the system. If a ligand has to explore a large area of the protein surface to find an adequate docking location, there is a lower probability of find the energy minimum than in the case of docking to a well-defined binding site on the protein. If a putative interaction region has been experimentally determined, this information can be used as useful input to guide the docking algorithm. Several new techniques to locate putative binding sites based on physicochemical properties or evolutionary conservation have been developed in recent years and are reviewed elsewhere. However, a good docking algorithm has to be able to predict realistically the docking site and distinguish it from nonspecific and/or energetically unfavorable ones even when performing a blind docking calculation.
The third step is the determination of the binding mode and it mainly depends on the atoms surrounding the docking site and the distance between suitable interacting pairs, as well as the specific conformation and orientation of the molecules of the complex. The resulting conformation is ranked according to its evaluation by the used scoring function.
The speed and accuracy of the docking results depends on the used docking approach. Two major docking approaches are used by the available docking softwares.
Shape Complementarity/Matching Methods
This is the most common docking technique. The molecules are described in terms of descriptors, which may include structural complementarity terms (solvent-accessible area, overall shape and geometric constraints) and binding complementarity terms (hydrogen binding interactions, hydrophobic contacts and van der Waals interactions). Taking these terms into account, a given molecule is docked into the protein target by matching features. A combination of different descriptors is found to be able to enrich the number of near-native solutions in the set of best ranked docking solutions. This is a fast and robust technique that has been used successfully to screen large compound databases. Its main disadvantage is based on the incapacity of modeling accurately large protein motions and dynamic changes in the conformations.
The second approach simulates the real molecular recognition mechanism, a more complicated and detailed process. According to this method, the two molecules from the complex are distanced by a physical distance and the ligand explores its conformational space and finds its docking site after a finite number of moves. These moves can be translations, rotations, torsion angle rotations or others, and each have a different contribution to the final total energy of the system. The advantages of this approach include a better incorporation of ligand flexibility and a physically closer approach to what happens in reality. However, as the ligand has to explore a large energy landscape, this approach takes longer to evaluate the best docking site. Grid-based techniques and fast optimization methods are being developed to overcome this disadvantage.
Mechanics of Docking
The success of a docking software depends on two components: (1) the search algorithm, and (2) the scoring function. The combination of these two components will dictate the overall results of the docking task.
All possible rotational and translational orientations, distortions, backbone and side chain flexibility and various degrees of freedom make it impossible to perform an exhautive sampling. To lower the possibilities, most docking programs account only for ligand flexibility (e.g. representing it as a ensemble of structures), maintaining the target rigid. Others attempt to insert some target flexibility by using rotamer libraries, or some degree of side-chain flexibility by using soft interfaces and scaling sterical interactions, or a further side-chain refinement stage.
Some of the most used search algorithms are described below.
Systematic or stochastic torsional searches about rotatable bonds
Rigid body methods
This searching method is based on a simplified rigid body representation of the protein onto a regular 3D Cartesian grid. Then it distinguishes grid cells according to whether the two molecules are near or intersect the protein surface, or are deeply buried into the protein core and the degree of overlap is scored. This method generates a large number of docked conformations with favorable surface complementarity. The disadvantages of this searching method are that it maintains the target protein rigid and it cannot find binding modes with a high degree of accuracy due to its inherent simplification of the complex. However, most rigid-body procedures result in good docked conformation if the used structure of the target protein used is obtained by experimental data.
Molecular dynamics simulations
In this approach the protein is kept rigid while the ligand explores freely the conformational space, obtaining a ensemble of states accessible to the complex. The generated conformations are docked and a determined number of minimization steps are performed, followed by an overall ranking. This is a computational complex method, although it does not need a specialized scoring function and it provides a useful tool to generate ligand conformations. In principle, it allows for full atomic flexibility or flexibility restricted to relevant parts of the complex during the docking task.
These searching algorithms perform global conformational searches particularly well. Based on the language of natural genetics and biological evolution, their goal is to "evolve" previous conformations into new low energy conformations. Each spatial arrangement of the pair is represented as a "gene" with a particular energy and the entire "genome" is a representation of the complete energy landscape which will be explored. Similar to biological evolution, random pairs of individuals are "mated" using a process of crossover and there is also the possibility of a random mutation in the offspring. During each iteration, high-scoring features in the current "generation" are preserved in the next cycle. This approach permits exploring of large conformational spaces. The main disadvantages include requiring the target protein to remain fixed during the docking task and multiple runs to obtain reliable results, which makes it a poor candidate to perform large databases screening. Limiting the conformational space to explore and the explorations of conformational changes at sites of interest can largely increase the performance of the docking task using this algorithm.
In docking, the goal of a scoring function is to serve as a mathematical method to predict the strength of the non-covalent interaction between the two molecules. Usually, this value is represented as the binding affinity, and indicates how favorable the binding interaction is. An ideal scoring function should be able to recognize favorable native contacts and discriminate non-native contacts with lower scores, and rank a set of molecules, predicting the correct modes of binding. These scoring functions can be parameterized (trained) against a set of experimental data for combinations of binding affinities, buried surface areas, desolvatation and electrostatic interation energies and hydrophobicity scores of molecular species similar to the species in study. There are four classes of scoring functions, which are described below. Choosing a scoring function should always be based on the resolution of the search method.
Most scoring functions are physics-based molecular mechanics force fields that estimate the nonbonded interaction energy of the docking pose. Affinities are estimated based on the total internal energy, which is estimated taking into account the strength of intramolecular van der Waals and electrostatic interactions and the desolvation energy. It is know that the free energy of binding is higly dependent on the system and it is often dominated by desolvation or electrostatic contributions. Other software also take into account the torsional free energy and the unbound system's energy as penalizing terms. At the end, a low (negative) energy indicates a stable complex, with a likely binding interaction.
Empirical scoring functions define simple functional forms for interactions between the two molecules of the complex. Some examples include the number atoms in contact between ligand and receptor, change in the solvent accessible surface area, number of hydrogen bonds, conformational entropy, and hydrophobic and hydrophilic contacts. These provide a fast method to rank potential inhibitory candidates.
Knowledge-based scoring functions are based on statistical analysis on intermolecular interactions and interactions distances extracted from large databases of protein-ligand complexes (e.g. PDB). This method is based on the assumption that there are intramolecular interactions between certain atoms that occur more frequently, which will be energetically favorable. If detected these interaction will contributed more to a favorable binding affinity.
Hybrid scoring functions combined one or more features from the ones described above.
There has is always a focus on the scoring function when developing a new docking program. Newly developed scoring functions are evaluated based on their ability to reproduce known ligand-binding patters for well-studied receptors. Despite the development of new and improved scoring functions, there is still a difficulty in identifying the best docking solutions from a list of false positives or decoys.
Disadvantages of Molecular Docking
Docking calculations can be hampered by a number of reasons: (1) the ligand binds to deep specific pockets of the protein structure; (2) does not consider the presence of solvent, which can be crucial to allow hydrogen bond interactions to occur; (3) if there is an attachment of the ligand to a solid surface (e.g. resin) via a spacer arm; (4) ligands with high flexibility; (5) weak interactions between the ligand and the protein; (6) large-scale motions of the peptide backbone. However, new optimizations and extensions are being developed into existing programs to overcome these drawbacks.
Autodock (version 4.0.1) was the program package that was used for the docking task in this work. It is used for automated docking of small molecules (e.g. peptides, enzyme inihibitors and drugs) to macromolecules (e.g. proteins, antibodies, DNA and RNA). It is a very complete software package, allowing a robust and accurate procedure and a reasonable computational demand. AutoDock which allows the use of ligand with fixed and flexible degrees of freedom.
The searching function used by AutoDock is the Lemarkian Genetic Algorithm (LGA), throughly described by Morris et al. LGA is a hybrid searching algorithm that combines the advantages of the global search of the common genetic algorithms and the advantages of a local search method to perform energy minimization, enhancing the performance relative to genetic algorithms. The local search does not require gradient information about the local energy landscape, facilitating torsional space search and allowing to handle more degrees of freedom.
The AutoDock scoring function (described by Huey et al is a semi-empirical free energy force field scoring function that evaluates conformations and calculates the ligand-receptor binding affinity. The force field was parameterized using a large set of complexes with known inhibition constants (Ki), structure and binding energies. It evaluates enthalpic contributions (e.g. repulsion, hydrogen bonding) using a molecular mechanics approach and evaluates de changes in solvation and conformational mobility through an empirical approach.
At the end of the docking task, Autodock returns a set of the top ranked answers according to the input system and parameters. Each is accompanied by the information regarding the estimated Ki and estimated free energy of binding, which is decomposed into (1) final intramolecular energy (van der Waals, hydrogen bond, desolvation and electrostatic energy), (2) final total internal energy, (3) torsional free energy, and (4) unbound system's energy and estimated as: (1)+(2)+(3)-(4).
Due to its technical characteristics, automated docking with AutoDock is not widely used to screen a large number of compounds. However, Park et al performed a benchmarking which showed the potentialities of this software for database screening, with a overall better average docking time and performance than other tested docking software.
The vast conformational sampling, degrees of freedom, complicated steric and chemical complementarity still offer a challenge for the computational approach to molecular docking. The inclusion of all possible conformational changes during docking searches is still impossible, and it would be of particular importance where only homology modeled structures are available. Slight modeling inaccuracies can result in false negatives, weak binding or even wrong docking poses. Better insights into the nature of protein folding and binding, protein dynamics and biomolecular energetics will allow the development of better docking algorithms.