This allows for billions of pieces of data to be analyzed, which is hugely beneficial for pursuing the possibilities of drug discovery.
AWS high performance computing (HPC) instances are used in platforms for research and development. Virtual screening is to find optimal combinations of proteins that cause illness and chemical compounds that are effective against the illness. AWS makes it possible to complete calculations that once took thirty days in a single day by completing all calculations at once with the use of several thousand cores.
Eisai is a pharmaceutical company that is mainly involved with prescription medicine, such as dementia drugs and anticancer drugs, and is also active in producing over-the-counter drugs. AWS has allowed them to greatly increase their productivity in research and development by deploying optimized instances for individual research and allowing flexible use of ICT resources for drug discovery.
Eisai works towards developing, manufacturing, and marketing pharmaceuticals under their corporate philosophy of being a human healthcare (hhc) provider aiming to provide healthcare that puts patients and their families first. Eisai has raised a policy to use ICT in order to further innovate, and has established hhc Data Creation Center (hDAC) as a core organization. "Our vision is to utilize advanced data-science technology including AI to accelerate drug discovery. In our Data Science Laboratory, we are aiming to quickly create value to meet patient needs through centralized management and analysis of all kinds of data relating to drug discovery, such as genomes, chemical compounds, electronic health records, and medical prescriptions," said Ken Aoshima, head of the hDAC Data Science Laboratory.
Drug discovery usually starts with exploratory research to find proteins that cause illnesses, and then creates new chemical compounds that bind to the proteins to stop them from causing harm. New drug candidates are evaluated for their physiochemical and biological properties during drug discovery and preclinical development phase, and arrive on the market after undergoing clinical development. However, there are over 100,000 different types of proteins that make up the human body and tens of millions of compound candidates that can bind to proteins related to illnesses, and therefore an environment is needed in which large-scale data can be generated, collected, and analyzed. Eisai set up an environment that allows for analysis on-premise, but in order to keep costs down and increase the speed of drug discovery, they now think that it is more effective to use cloud computing, which can provide resources to meet their needs.
Eisai established a pilot environment in the US region of AWS in 2010 to try cloud utilization, but it was not fully introduced at that time due to the time required to transfer analytic data overseas through WAN lines. A few years later in April of 2016, they again started to set up an AWS environment, and in October of the same year they began to utilize an HPC platform that included Amazon EC2, Amazon ECS, and Amazon S3.
"We used Virtual Private Cloud (VPC) to ensure security, and we also used AWS Snowball for the initial data migration from on-premise to AWS, and we were able to securely transfer around 50 TB of existing raw, analytical data. When it came to data that was created after the transfer, we were able to upload that data in real time by using AWS Direct Connect. When we considered using AWS in 2010, we calculated that it had taken one and a half to two weeks to transfer data from the Tsukuba Research Laboratories to the U.S., so AWS has shown a great improvement since then." said Yuji Miura, Senior Researcher at the hDAC Data Science Laboratory.
Chemical compound data is one of the pharmaceutical industry's best kept secret, so using cloud computing to handle such data was one of the most progressive decisions in the industry. hDAC Data Science Laboratory Director Takashi Seno had this to say;
"In order to utilize big data, there was no way we could keep being particular about our on-premise environment. I had upper management understand the situation by logically explaining to them that AWS ensures flexibility and security for analysis, and that there is no risk of information leaking. ”
The HPC platform built on AWS is currently used for exploratory research for drug discovery, primarily virtual research for computational verification of combinations of proteins and chemical compounds. In order to shorten processing at our on-premise environment where resources were limited, over ten million compound candidates were narrowed down to around ten thousand before we even started. However, using AWS' resources allowed us to increase the likelihood of identifying a candidate compound by simulating combinations of about 10 million compounds and about 200 proteins by using all possible combinations thereof.
Eisai uses a cluster environment consisting of CfnCluster provided by AWS, and uses a several thousand core resources that are charged per time.
"AWS has enabled us to undertake analysis that goes beyond conventional thinking. Our on-premise environment was limited to using several hundred cores, but AWS can use several thousand cores at once, meaning that calculations which once took thirty days can now be completed in just one day. Furthermore, large-scale calculations at an on-premise environment require coordination with other researchers, but with AWS it is possible to work by establishing an HPC environment to suit whatever the objective may be. It would cost over one million dollars if such resources were to be provided on-premise, but, on AWS, it is possible to procure only the resources that are needed for each instance. Spot Instances are also used depending on the objective, thus achieving a better balance between cost and speed. " said Kazuya Nagaoka, hDAC Data Science Laboratory Manager.
Shortening the time of performing calculations is directly connected to shortening the time until the next calculation can be made, which in turn is connected to reducing development costs and improving development productivity. Another benefit is that you can launch instances that are optimized for your application, operating system, and amount of memory needed for calculations, or switch to faster instances. Keishi Akada, Senior Researcher of the hDAC Data Science Laboratory, had this to say, “When we switched the brainwave calculations running on GPU instances from Amazon EC2 P2 instances to more powerful P3 instances which were just released in Japan at that time, the actual calculations became almost seven times as fast, and machine-learning calculation results that used to take over a week were done in three to four days. We can now perform three calculations in the time it used to perform one. ”
Technical support from AWS has received positive feedback, such as setting up the system environment for the Proof of Concept (PoC) verification stage and responding to requests to increase service quotas after deployment. "They were flexible to respond for our company's needs to discover innovative new drugs. “ (Keishi Akada)
Looking towards the future, there will be a gradual shift to AWS from HPC environments using a hybrid of on-premise environments and AWS, and the time and costs related to drug discovery and development will be further reduced, while migrating all phases of drug discovery applications to AWS such as genomic data analysis and image analysis via machine learning. Furthermore, it is clear that we will make contributions to personalized medicine through running data lakes that store real-world data, such as electronic medical records and medical prescriptions, in the cloud and conducting factor analysis.
"Rather than using an external cloud, we now can experience pharmaceutical research and development in a secure environment, almost as if AWS were a part of our company, Eisai. Looking towards the future, we will continue to expand the use of this cloud technology in a variety of analyses for drug discovery to meet the needs of medical conditions for which no cure has been found. " said Ken Aoshima eagerly.
Ken Aoshima
Takashi Seno
Kazuya Nagaoka
Keishi Akada
Yuji Miura
For more information about how the AWS can help to high performance computing (HPC), could you please refer to the High Performance Computing in AWS in AWS information page.