SandboxAQ News
Massive financial outlays, protracted timetables, and a startling failure rate characterize the conventional drug discovery process. It might take decades to bring a new drug to market, from preliminary research to regulatory approval. Many medication candidates that appeared to be very promising during this time fail to live up to expectations, either because of safety issues or ineffectiveness. Only a small percentage of applicants pass regulatory requirements and clinical testing.
Here comes SandboxAQ, which is assisting scientists in precisely predicting biological events, exploring large chemical regions, and gaining profound insights into molecular interactions. In the end, it shortens the time needed for drug discovery and development by using state-of-the-art computational techniques like active learning, absolute free energy perturbation solution (AQFEP), generative AI, structural analysis, and predictive data analytics. All of this is accomplished on a cloud-native basis.
Drug design is an iterative process known as the Design-Make-Test cycle, which entails creating, synthesizing, and testing compounds. When their computational approaches are failing, many clients turn to SandboxAQ during the design phase. SandboxAQ accelerates this step of the cycle to help medicinal chemists introduce new and powerful molecules. In a neurodegenerative disease project, SandboxAQ extended chemical space from 250,000 to 5.6 million molecules, yielding in a 30-fold increase in hit rate and faster candidate compound discovery.
Cloud-native development for scientific insight
Large-scale computation is a key component of SandboxAQ’s software, and in order to optimise flexibility and scale, they employ a cloud strategy that combines Google Cloud tools and infrastructure.
Large-scale virtual screening campaigns require technologies that can grow economically and with agility. In particular, SandboxAQ engineers must be able to rapidly repeat scientific code, run that code at scale in an economical manner right away, and store and arrange all of the data that is generated.
SandboxAQ used Google Cloud infrastructure to deliver a notable increase in scalability and efficiency. They used tens of thousands of virtual machines (VMs) in parallel by scaling their computing throughput by 100X. By cutting down on idle time by 90%, they also increased utilization. SandboxAQ streamlined its processes, from large-scale batch processing and machine-learning model training to code development and testing, by combining development and deployment on Google Cloud.
What is SandboxAQ?
SandboxAQ is developed and deployed entirely in the cloud. Development is carried out using a cloud-based platform that gives scientists and engineers self-service virtual machines (VMs) with standardized and centrally managed environments and tools, while code and data reside in cloud-based services. This is significant because scientific programming frequently calls for powerful computer hardware. Scientists can use GPUs or 96-core CPUs. As shown below, they can also build new machines with different CPU types or configurations, allowing for low-friction testing and development procedures across diverse resources.
The company’s bench client is used by SanboxAQ scientists and developers to control and access their Bench equipment. They can utilise a variety of managed tools, such as JupyterLab for a recognisable notebook development flow or a browser-based VNC service for fast remote desktop, or they can connect to computers over SSH.
Researchers can use an internal tool powered by Batch, a fully managed service to schedule, queue, and execute batch jobs on Google infrastructure, to dispatch SandboxAQ parameterized sets of computations as jobs once the code is ready to be performed on a wider scale. Changes may be swiftly implemented at scale with the close synchronization between the development and batch runtime environments. Bench machine code is instantly available for batch execution after being posted to GitHub.
The new tools are then automatically available on SandboxAQ scientists’ bench computers as they are reviewed and integrated into main of the company’s monorepo. These scientists can use either on-demand or Spot VMs to launch parallel jobs processing millions of molecules on any type of Google Cloud virtual machine resource in any global zone.
Globally resolved transitive dependency trees in SandboxAQ simplify package and dependency management. By using this technique, Google Batch can easily interface with other tools created by engineers to train many instances of a model simultaneously.
SandoxAQ’s strategy heavily relies on machine learning, therefore having simple access to data is crucial. The Drug Discovery team at SandboxAQ also works with clients who have sensitive data at the same time. Bench and batch workloads read and write data from a single interface that is controlled by IAM, giving the company granular control over various data sources, in order to protect the data of its clients.
In the meantime, it is easy to create tools to monitor these workloads, surface logs to SandboxAQ scientists, and sort through massive volumes of output data with Google Cloud services. Without requiring the scientific team to deal with infrastructure, modifications are made instantly available when new features are tried or issues are discovered. As the code stabilizes, they can then integrate it into production apps downstream in a centralized, uniform manner on Google Cloud.
To put it briefly, SandboxAQ encounters less difficulty while developing new workloads and executing them at scale because to Google Cloud’s unified development, batch computing, and production environment. Customers may migrate from experimentation to production quickly and easily with SandboxAQ’s shared environments for scientific workload research and engineering, which quickly yields the desired results.
SandboxAQ solution in the real world
Drug development initiatives aimed at a variety of difficult-to-treat illnesses are already being significantly impacted by SandboxAQ. For instance, there are cutting-edge partnerships with Riboscience, Sanofi, the Michael J. Fox Foundation, and Professor Stanley Pruisner’s group at the University of California San Francisco (UCSF), to mention a few. This strategy, which is based on Google Cloud SandboxAQ, has a higher success rate than previous techniques, such as high throughput screening, proving SandboxAQ’s revolutionary potential in drug development and expediting patient treatments.
