The development of a typical prescription drugcosts up to $2.
6 billion, and takes longer than a decade to get to market65. When a drug gets approved for use in clinicalpractice, its effectiveness differs from patient to patient, sometimes havingno effect at all, not to mention adverse or side effects. More effective waysare needed to develop new drugs as well as to improve management of existingdrugs for specific patients. To accelerate drug development, scientificadvances and associated technologies (genomics, proteomics, metabolomics, andbioinformatics) have been incorporated into the process. To improve therapeuticefficacy, ‘targeted therapy’ has been developed, in which drugs are directedagainst disease-specific molecules.
Society is in need of many more personalizeddrugs, a precision medicine, produced far more rapidly and cheaply than iscurrently the case. For the drugs which do have effects, the emergenceof resistance is usually unavoidable, which presents health care systems withserious challenges. Given the complexity of emergent mutations indisease-related proteins, the central mechanism for drug resistance,researchers must invoke computational tools to help understand the mechanismsof protein function and inhibition, how mutations affect them and how drugs canbe designed to circumvent resistance. One of the central quantities of interest whenassessing a potential drug is the thermodynamics approach (evaluating bindingaffinity or binding free energy). It determines the strength of theinteraction, which is key for successful pharmaceutical drug development andthe efficacy of clinical intervention. The use of computationally basedmolecular modeling and dynamics methodology to calculate the strength ofmacromolecular binding free energies is naturally of major interest in drugdiscovery and development. Furthermore, computational modeling and simulationsare easier and much cheaper to apply on large scale to evaluate many potentialmedicines along with accounting the diversity in any patient genomedifferences. Historically, the industry has avoided engagementwith high performance computing due to its lack of accuracy andreproducibility, and/or long turnaround time.
Most existing protocols forcomputing free energy changes are regarded as unreliable as well ascomputationally expensive. A recent survey by Nature revealed that more than70% of researchers failed to reproduce another scientist’s results, while morethan half were unable to reproduce their own66. This holds irrespective of the field of researchand applies to both experimental and computational methods.
In the case ofexperiments, a variety of reasons ranging from mixed up chemicals, throughfluctuations in the environment, variations in the experimental setup toconfirmational bias can be held responsible for non-reproducible results. Inthe case of molecular simulations, the reasons reside in a combination oftheory and the model including the accuracy of force fields, convergence of thecalculations, efficiency of software and so on. However, for all traditionalmolecular dynamics (MD) based methods, lack of reproducibility is intrinsic andis independent of these other issues67. This is because the prediction of Gibbs freeenergy macroscopic properties requires ensemble averaging over microscopicstates.
Recently however, pharmaceutical companies arebecoming more interested in calculating binding free energies due to advancesin technical and practical implementation. In particular, the FEP+implementation of Free Energy Perturbation (FEP) has shown potential to improvethe ability to predict protein-ligand binding affinities on an industriallyrelevant time scale68. Research is ongoing to understand how broadly applicable the methodis, and how accurate its predictions are when applied to active drug discoveryprograms. There still exists a gap between published success cases and thescale of implementation and robustness needed for industrial drug discovery;such work needs to be performed in a collaborative, non-competitive framework.
Our work in this area, in developing and applying the molecular approach usingGROMACS58, FOLD-X59-60, and Friend with StSNP server61-62 integrated with python automated tool for efficient bindingenergy calculation is simple yet rationally significant to make free energycalculations rapid, accurate, precise and reproducible. Our proposed study will evaluate the potential ofour approach for the improved predictive power for accurate ranking of ligandsby their binding free energies. We will apply the integrated approach on a fewkey drug targets in the association study of diabetes and breast cancer foraccurate, precise and reliable free energy predictions. With our method,comprising of GROMACS58, FOLD-X59-60, and Friend with StSNP server61-62 integrated with the python automated tool, to execute theworkflows of binding affinity calculation will help us complete the wholeexecution in less time, depending on the architecture and hardware available.
The accuracy and speed of the calculation will make it possible to performactionable decisions in clinical or industrial settings. As large and securecomputing resources become more routinely available in Qatar, for examplethrough cloud computing, it will become increasingly easy for research groups inQatar to access approaches like the one used in this study. Consequently, therobust prediction of protein?ligand binding affinities in an industrial settingshould become more routine and offer a long awaited development in the field ofstructure-based drug design and sequence-based personalized medicine.