Otsuka Pharmaceutical business conducts research and development of pharmaceuticals in the areas of cardiovascular/renal, infectious diseases, ophthalmology, and dermatology, in addition to its top priority areas of psychiatric disorders, neurological diseases, and oncology/immunology. The company which has a global research base in addition to Tokushima opened a new research facility as "Osaka Drug Discovery Research Center" in Minoh City Osaka Prefecture in August 2022 with the aim of realizing highly innovative drug discovery using cutting-edge technology.

In recent years, the utilization of big data obtained from laboratory instruments, real-world data obtained from clinical sites, and open data provided by governments and other organizations has become indispensable in drug discovery research and development. The company is developing a research environment using high-performance computing (HPC) and machine learning for its traditional on-premises environment, yet challenges with hardware procurement and lack of flexibility have slowed progress.

"In the digitalization of drug discovery research, we need to be able to respond to the ideas of researchers proactively. To achieve this, it is essential we use a highly agile cloud and in-house development environments." said Mr. Tomoya Inaba who is the Manager of Scientific Partnering in Research Management Department.
"On the other hand, we handle highly confidential information, so ensuring security was a prerequisite for promoting the cloud utilization" said Inaba.

According to this background, we began considering the promotion of cloud utilization in the research department, focusing on two key issues: security and in-house development.
"Security is a top priority from the standpoint of handling confidential information. Although there were some initial concerns about placing data outside of the company, we were fully convinced that it was possible to create a highly secure environment by verifying various configurations ourselves and gaining a better understanding of the functions. In addition, to maintain an application that contributes to research work, it was essential that we work through a trial-and-error process by in-house production" said Inaba.

Otsuka Pharmaceutical has focused on drug discovery research using cryo-electron microscopy to analyze the three-dimensional structure of proteins, as it began to shift its research environment to the cloud starting in the Fall of 2020. To promote and scale research based on Cryo-EM-derived data as quickly as possible, the team launched a project to build an HPC environment for analysis on the cloud.

AWS was selected as the cloud service and AWS ParallelCluster was selected as the cluster management tool to automatically scale compute nodes for building its HPC environment. Mr. Riki Kajiwara, from the Research Management Department explained the reason for the selection. He said "AWS ParallelCluster was a very useful service that allows us to flexibly build a computing environment. As this service provides a managed environment for job scheduler and parallel calculations, we didn't have to build it ourselves. Also, it was another deciding point that software primarily used in the Cryo-EM data analysis workload, such as Relion was already proved to run on AWS”.

The development of the HPC AWS ParallelCluster environment started in February 2021 and was fully operational by August of the same year. "The process of setting up AWS ParallelCluster and installing the analysis software took only a one week and we were able to start PoC* immediately.” said Mr. Kajiwara

Cost optimization of the HPC environment was achieved by using a combination the latest Amazon Elastic Compute Cloud (Amazon EC2) instances equipped with CPU/GPU and the utilization of Amazon EC2 Spot instances. Further cost and performance optimization was achieved by using additional AWS services such as Amazon Snowcone to transfer data to Amazon S3, and Amazon FSx for Lustre, a high-performance distributed file system, to process up to 10 TB of cryo-EM data at a time.

The company also uses AWS in the development of machine learning models for drug discovery. Amazon SageMaker Studio as an integrated development environment for machine learning, is employed for the training and building of Python based models using Anaconda.
"We also used Amazon SageMaker Studio to evaluate a machine learning model for predicting protein conformation called AlphaFold2, and were able to quickly perform GPU-based validation. Amazon SageMaker Studio is useful because it enables us to easily port existing training models run on-premises to cloud, and train only when needed, using the plentiful resources of AWS at low cost," said Mr. Kajiwara.

In addition, we are building a secure operation system by utilizing AWS services such as Amazon Inspector, which diagnoses vulnerabilities on Amazon EC2 instances, to ensure security to protect confidential data for drug discovery research.

Building HPC environments in AWS is creating a paradigm shift, reducing analysis time and eliminating wait times for researchers.
"By processing and analyzing Cryo-EM data in the cloud we were able to reduce structure determination from 7 days on-premises, to 2-3 days. Analysis cost concerns were resolved when we discovered that we were able to significantly reduce those costs by refining the way we use the system," said Mr. Kajiwara.

Procurement of the on-premises computing environment used to take nearly 6 months, however by using AWS, researchers are able to provision and perform large-scale analyses on demand. Additionally, analysis is no longer tied to the location of the laboratory. By utilizing NICE-DCV, a high-performance remote display software in AWS, scientists are able to remotely analyze and visualize structural data in real-time.
“AWS is well received because it can respond quickly to the needs of researchers by providing on-demand computing resources whenever they want to analyze. In the analysis of Cryo-EM data, researchers are gradually realizing the benefits of reduced analysis time and costs as they use the AWS HPC environment," said Mr. Inaba.

In the future, we plan to expand HPC use cases to molecular dynamics calculations, bioinformatics, and run many more workloads. In addition, for dramatically increasing data, we plan to develop an archiving and backup environment using AWS storage services so that researchers can access the data they want at any time. In the near future the company intends to expand the foundation for utilizing Amazon SageMaker and to build MLOps to cover everything from data pre-processing to deployment and operation of predictive models.
"We will continue to provide new technologies based on AWS services to research departments while capturing the needs of the field, thereby helping to improve their competitiveness in drug discovery research," said Mr. Inaba