Dakka, Jumana. Enabling ensemble-based methods for computational drug campaigns at scale on high performance computing clusters. Retrieved from https://doi.org/doi:10.7282/t3-ywcx-3x05
DescriptionFree energy calculations that use molecular dynamics (MD) simulations are emerg- ing as an important tool for studying important problems like computational drug de- sign. Recent evidence suggests that free energy calculations, specifically binding affinity calculations, i.e., calculations that quantify the strength of interactions between drug molecules and target proteins, can achieve useful predictive accuracy (< 1 kcal/mol) to impact clinical decision making in computational drug design. However, free en- ergy calculations must provide results rapidly and without loss of accuracy. The dual challenge of scaling thousands of concurrent simulations and adaptive selection of fa- vorable simulations based upon feedback from statistical errors and uncertainty need to be tacked. To address these challenges requires advances in algorithms, efficient utilization of supercomputing resources, and software tools that facilitate the scalable and automated computation of varied free energy calculations. This thesis evaluates the requirements of large-scale and adaptive ensemble-based approaches in order to build a software tool designed to enable applications like computational drug campaigns.
In this thesis, we introduce a software tool called the High-Throughput Binding Affinity Calculator (HTBAC), the primary contributions of which are: (i) its ability to apply recent advances in workflow system building blocks to binding affinity calculations, (ii) the ability to execute simulations independent of the simulation software package or supercomputing resources, (iii) enable features of scalability and adaptivity in order to improve resource utilization and scientific results.