Assessment of Computational Methods for Ligand Binding

Abstract

Most drugs act on biomacromolecules. The Cost of developing new drugs is very high. A method to accurately predict binding affinities would be very useful. We have studied molecular mechanics with generalised Born and surface--area solvation (MM/GBSA) and alchemical free energy perturbation methods (FEP) for use in calculations of ligand binding energies.



For the MM/GBSA method we have tested:

Calculating the non-polar solvation term with the polarized continuum model, a method based on cavity and dispersion terms, and a method based on a linear relation to the solvent-accessible surface area.

Replacing molecular mechanics terms with energies calculated with the semiempirical quantum mechanics AM1, RM1, PM6 Hamiltonians, and adding hydrogen bond and dispersion corrections.

Inclusion of explicit water in the binding site of a protein.

Effect of system truncation on estimated energies.



The results show that for continuum solvation models knowledge of hydration state of binding site is important. The rest of variations of the MM/GBSA method for the tested systems showed only minor improvements.



We have done a large systematic study of calculating relative binding free energies for 10 proteins binding 107 ligands with Bennett acceptance ratio (BAR) method. For the most of systems binding affinities could be calculated within 4 kJ/mol of experimental values.



We have also participated in the SAMPL3 and SAMPL4 blind binding challenges to see how the MM/GBSA and FEP methods perform. The MM/GBSA failed to predict experimental binding affinities, which might be due to poor precision of the method as experimental data had very narrow range of about 9 kJ/mol. In SAMPL4, the BAR method gave the best predicted binding affinities.

Abstract (Swedish)

Popular Abstract in English

Proteins are a big biological molecules that help cells to function. Proteins play important roles in all vital processes of the living cell, transporting molecules, speeding up chemical reactions, supporting structure of other proteins and the cell, helping the cell to move.



Various disease-causing agents (e.g. bacteria and viruses) are constant threat to us. It is therefore an important aim of science to stop them. This could be done by disabling proteins that are essential for the life cycle of those organisms. This is called inhibition. Proteins typically act by forming a complex with a small molecules, e.g. activators or substrates. Molecules that bind to a protein are called ligands. The binding is similar to the docking of a spacecraft to a space station. In fact, this is a good analogy, because the spacecraft can only dock to the station in a specific location. The same applies to a protein and a small molecule -- the ligand binds only to a specific place on the protein. This the process is called ligand binding.



The binding is controlled by a quantity called the free energy, G_{bind}. This is an important quantity as it tells us how strong the complex is. G_{bind} can be measured experimentally or estimated by a computational-chemistry software. There are many different methods to calculate the binding free energy. In this thesis, I have evaluated how some of these methods perform.



One such methods is called MM/GBSA. We have tried to improve this approach by improving some of the approximations involved. Unfortunately, our efforts gave only slight improvements over the standard method, but our improvements could be useful for some specific cases.



We have also investigated more demanding methods for ligand binding, viz. the alchemical free energy perturbation methods. They involve non-physical transformations of the ligands. In theory, these methods are exact, but only when used with exhaustive sampling and a perfect energy function. In practice, this is not possible and therefore the performance of the method depends on the protein studied.



It is important to know how these methods would perform in more realistic situations. With this aim, there are blind binding challenges for researchers to evaluate their methods. In these competitions, the experimental results are unknown to the participants during the calculations, but they are revealed after the submission of all computational results. This can show how general and robust the methods are. We have participated in two such competitions, SAMPL3 and SAMPL4 with varying success.



Drug discovery and development is a very expensive process. It costs around 1 billion $ and takes over 10 years to develop a new drug. Hopefully better computational methods can reduce the cost and time it takes to make new drugs, as well as help to reduce number of animals used for testing the potential drugs.

• author: Mikulskis, Paulius
• supervisor: Ulf Ryde
• opponent: Essex, Jonathan W., University of Southampton
• organization: Computational Chemistry
• publishing date: 2015
• type: Thesis
• publication status: published
• subject: Theoretical Chemistry
• keywords: binding affinities, MM/GBSA, MM/PBSA, LIE, FEP, TI, BAR, protein-ligand complexes, drug design
• pages: 191 pages
• defense location: lecture hall A, Center for Chemistry and Chemical Engineering (Kemicentrum)
• defense date: 2015-04-24 09:15:00
• ISBN: 978-91-7422-395-8
• language: English
• LU publication?: yes
• additional info: The information about affiliations in this record was updated in December 2015. The record was previously connected to the following departments: Theoretical Chemistry (S) (011001039)
• id: 8215e5c8-e3dc-41f6-8c17-c2c4b4cc75e7 (old id 5155224)
• date added to LUP: 2016-04-04 12:18:07
• date last changed: 2018-11-21 21:10:09