Introduction: One of the main computational challenges of QSP modeling is the possibility to perform thousands of simulations with large-scale dynamic models. Those simulations reflect therapeutic regimes, dosage calibration, virtual patients (VP) trials, etc. Secure cloud environment together with software, which supports parallel simulations setup, can drastically speed up such simulations and improve applicability of VP-based approach.
Objectives: The goal of the study is to propose a software and hardware solution which can address one of the main computational challenges of a QSP project. Namely, provide the users with a secure and easily extendable environment to both develop the models and run computationally demanding simulations in parallel. A cloud-based environment with the modeling server and requested on-demand remote computational nodes is proposed to perform efficient parallel virtual patients simulations. This work is further development of computational environment architecture presented by the authors on ACoP13.
Methods: The application of the proposed infrastructure is demonstrated on the use-case of virtual patients simulations for covid-19 model. The infrastructure contains the modeling server and requested on-demand computational nodes. The user can perform simulations on the server and extend its capacity by provisioning worker nodes at any time. The software part of the environment relies on Kubernetes cluster manager API and the ability of Julia language and SciML packages [1] to distribute workloads between worker nodes. K8sClusterManagers package is used to call Kubernetes from Julia and start Julia worker nodes on the provisioned containers. The key software component of the environment is an open-source package HetaSimulator [2], which provides an interface to set up and run parallel multi-conditional simulations.
Results: The Covid-19 model with a number of simulation conditions and generated parameters sets illustrate the applicability of the presented infrastructure. The dedicated server resources are utilized to simulate small virtual populations while on-demand resources are provisioned for large-scale virtual trials. This example illustrates flexibility of the infrastructure both in terms of cost savings and speeding up computationally demanding simulations.
Conclusions: The proposed architecture allows the users to both develop the models and run computationally demanding simulations (as VP generation) in parallel in the single easy to use environment. It supports unlimited resource extension by adding new compute instances on demand. The security of the proposed infrastructure is guaranteed by cloud provider’s certification. The infrastructure is OS-independent and allows the users to develop the models on the dedicated workstation in the preferred environment (e.g. Windows) and run simulations on the most performant compute instances (Unix). The solution also allows customer’s access to the environment and participation in the model development process and simulations.
Thank you for your message! We will get back to you shortly.
