Objectives: The objective of this study is to present innovative methodologies that significantly enhance the speed and accuracy of clinical trial simulations. By integrating Just-in-Time (JIT) [1] compilation, parallelization of large-scale virtual patient populations, and advanced strategies for solving ordinary differential equations (ODEs), we aim to simplify simulation setup and enable rapid analysis of results.
Methods: Just-in-Time Compilation: models, including ODE systems and initial conditions, are described in portable formats such as SBML[2]. Utilizing JIT compilation, these models are converted into efficient native machine instructions, optimizing execution speed by leveraging runtime-specific data and modern processor features like SIMD[3] extensions. The JIT implementation is specialized for the solving of QSP derived ODE systems such that it is more efficient than a generic, agnostic JIT. Leveraging QSP-specific knowledge opens opportunities for further optimizations which a generic JIT implementation would not be able to achieve.
Parallelization and cloud computing: Simulations involve large virtual patient populations, categorized as "Embarrassingly Parallel" problems. By leveraging AWS cloud services[4], we scale computational resources to match demand, enabling simultaneous processing of patient simulations and reducing overall computation time.
Efficient ODE Solving: We employ adaptive, implicit solvers from the Sundials suite[5], specifically the BDF method, to handle stiff[6] ODE systems efficiently. These solvers adjust time-step sizes and orders dynamically, balancing accuracy and computational cost.
Results: Using different benchmarks, we show how implementing the above methodologies leads to faster, more scalable simulations. First, the advantage of JIT compilation and efficient ODE solvers is demonstrated using models from the BioModels database[7]: the average solving time per patient in JInko is compared with other softwares. Second, large scale simulations involving millions of virtual patients over thousands of virtual CPUs are run to quantify the scalability of the cloud-based infrastructure.
Conclusions: The integration of JIT compilation, scalable cloud computing, and advanced ODE solving techniques significantly enhance the performance of clinical trial simulations. These innovations facilitate rapid, accurate analyses, thereby accelerating the drug development process and enabling more efficient clinical trial designs.
Citations: [1] Lattner, Chris, and Vikram Adve. "The llvm compiler framework and infrastructure tutorial." International Workshop on Languages and Compilers for Parallel Computing. Berlin, Heidelberg: Springer Berlin Heidelberg, 2004. [2] Hucka, Michael, et al. "The systems biology markup language (SBML): a medium for representation and exchange of biochemical network models." Bioinformatics 19.4 (2003): 524-531. [3] Ibrahim, Amr Hussam, et al. "An analysis of x86-64 instruction set for optimization of system softwares." Planning perspectives (2011): 152. [4] Amazon, E. C. "Amazon web services." Available in: https://aws. amazon. com/es/ec2/(November 2012) 39 (2015). [5] Hindmarsh, Alan C., et al. "SUNDIALS: Suite of nonlinear and differential/algebraic equation solvers." ACM Transactions on Mathematical Software (TOMS) 31.3 (2005): 363-396. [6] Hairer, Ernst, et al. "Stiff problems—One-step methods." Solving Ordinary Differential Equations II: Stiff and Differential-Algebraic Problems (1991): 1-254. [7] Malik-Sheriff, Rahuman S., et al. "BioModels—15 years of sharing computational models in life science." Nucleic acids research 48.D1 (2020): D407-D415.
