Reduces monthly data processing costs by 30%
20x faster training speed for machine learning models
Improves computational performance with Google Cloud SQL to analyze novel protein structures faster
Developed scalable data infrastructure to support company growth
Reduces time spent on support and data maintenance to focus on building new data models
As a leader in AI-driven drug discovery, Ordaōs relies on its cloud computing capabilities to design, process, and analyze data for millions of protein structures. It needed a cloud partner that could support larger-scale data sets as its operations grew. The company moved its cloud computing and storage to Google Cloud Storage and Google Kubernetes Engine to achieve increased flexibility and easier scalability as it took on new, larger AI projects.
The drug discovery process is equal parts precise and serendipitous. One of the most famous tales of drug discovery is the story of penicillin: Dr. Alexander Fleming returned from a holiday to a molding petri dish, and the mold that accidentally formed in it would lead him to the world’s first antibiotic. Nearly 100 years later, the drug discovery process is more intentional, but it still relies on a dash of luck.
Ordaōs is a human-enabled, machine-driven drug design company on a journey to overcome the industry’s dependency on chance with an AI-based approach to intelligent drug design. Rather than synthesizing thousands or millions of in-lab tests, the company uses generative AI to design mini-proteins, achieving better results 30x faster than traditional discovery.
“Using AI to make intelligent choices in the discovery process helps us avoid scaling problems that traditional labs run into,” said David Longo, CEO and Co-founder of Ordaōs. “Our major advantage with an AI, cloud-first approach is that we’re able to traverse a massive design space in a more methodical way.” To handle their growing AI needs and create novel therapies for more patients faster, Ordaōs sought a reliable cloud storage and processing provider with scalable solutions.
Managing and processing terabytes of data is a monumental task, but for Ordaōs, doing so effectively is a core business offering. With its previous cloud provider, the company experienced outages that impacted its ability to work on projects. “With our old cloud provider, it felt like something else would break every month,” Longo said. “And when we needed support from their team, we didn’t have it.”
The company began its search for alternative cloud solutions and found close alignment between its own roadmap and Google Cloud AI. It also saw value in other partnerships available through Google Cloud, including AlphaFold, an AI program that predicts protein structures. “From the very beginning, Google Cloud wanted to enable our success and our value generation,” Longo said. “Along with migrating our data, the team connected us with new integrations and partners, including the teams at DeepMind and AlphaFold. Those connections are really impactful for a small startup.”
The organization was also looking for more ways to optimize the performance of its SQL databases running on Kubernetes. “Kubernetes is our fundamental orchestration technology. With Kubernetes we can launch new apps and models in minutes while maintaining complete control over resource utilization and interaction,” Longo shared. “We were running fine on our previous provider, but when we tested Google Kubernetes Engine (GKE) and Cloud SQL, everything worked more smoothly. It’s more manageable and there are more features that support the work we’re doing.” With GKE, Ordaōs has access to a scalable, automated Kubernetes service. Combined with Cloud SQL, which provides the company with complete database management and provides a foundation for upcoming generative AI projects, Ordaōs is able to set itself up to grow. Plus, with improved cloud uptime, Ordaōs can trust its data is available when it's needed and focus on improving its testing process. These performance improvements have contributed to Ordaōs’ ability to train its machine learning models more than 20x faster than what was possible with its previous provider.
Confident that Google Cloud could support its future projects, Ordaōs began its migration. To support its small team during the move, the company worked with Google Cloud partner Persistent. “The Persistent team was invaluable to us,” Longo said. “We effectively had an additional three full-time employees to handle the hands-on-keyboard work needed to migrate our workloads.”
Since the migration, Ordaōs has spent less time fixing and more time creating. “We’ve been able to accomplish more at a larger scale, run a higher volume of experiments faster, and generally improve operation speed,” Longo added. This acceleration, along with finding and removing unused resources during the migration, has saved the Ordaōs team 30% in monthly data operations costs.
In addition to processing its own data, Ordaōs uses GKE to ingest and experiment with third-party data, including files from AlphaFold. “We found that the fastest way to load AlphaFold’s enormous, multi-terabyte database was actually from memory because GKE is just that fast,” Longo said. Datasets of this size typically require databases to process since disk space is limited, but Ordaōs has been able to trust GKE to load and process at scale. “We keep finding optimization techniques that I can only attribute to Google Cloud.”
While the company’s migration was largely driven by its current data needs, it also wanted to find a cloud provider to help the team scale up and down to meet its future requirements. Ordaōs connected its GKE clusters to Memorystore for Redis, an in-memory service that reduces application latency, to support event-driven scaling via KEDA.
This means that the team can automatically scale its computing infrastructure up or down as projects come in, which reduces the engineering resources required for lab teams to start working.
In the long term, Ordaōs sees new opportunities to iterate on its infrastructure with Google Cloud. “We continue to use more Google Cloud solutions in this period of rapid iterations and mass data gathering,” Longo added. “With structured cloud data storage we have access to knowledge graphs and big schema ingestions at a scale that allows us to validate our in silico (digital) designs, drag those into Google Cloud, iterate on them with AI, and put them back into the lab in record time. I’m excited to see how our processes will continue to grow in the future.”
During the migration to Google Cloud, Ordaōs also moved into Johnson & Johnson’s JLABS facility to work on even more cutting-edge projects. In this growth period, the company continues to explore new ways to collaborate with Google Cloud to best make use of its experiment-related data.
Ordaōs is a human-enabled, machine-driven drug design company. Their miniPRO™ proteins help drug hunters deliver treatments that are safer and more effective than traditional discovery methods.
Industries: Healthcare & Life Sciences
Location: United States
Products: Google Cloud, Google Kubernetes Engine, AlloyDB, Cloud GPUs, Cloud SQL, Cloud Storage, Compute Engine, Memorystore
About Google Cloud Partner- Persistent
Persistent is a global cloud consulting and services partner that focuses on enterprise modernization and data engineering.