How Proteros accelerates drug discovery by using AWS ParallelCluster

Proteros is a leader in structure-based drug discovery solutions, and supports pharmaceutical, biotechnological, and academic clients with advanced technologies like Cryogenic Electron Microscopy (Cryo-EM) and Protein Crystallography (PX).

In this blog post, we’ll explore how Proteros implemented an HPC solution that scales with their scientific ambitions. We’ll talk about how they started with a secure AWS Landing Zone as their foundation. Then we’ll show how they systematically built on this with cryoSPARC and AWS ParallelCluster, and a new innovative hybrid storage approach. Finally, we’ll examine the key architectural decisions that enabled them achieving high scalability, while delivering substantial cost efficiencies compared with conventional HPC solutions ¾ all without compromising performance.

The challenges

Cryo-EM and PX are essential for visualizing protein structures at near-atomic resolution. Cryo-EM is a Nobel Prize-winning technique that revealed the critical SARS-CoV-2 spike protein structure. It captures biomolecules in their near-native states through flash-frozen samples. PX, on the other hand, provides detailed insights through crystallized structures.

Together, these technologies form the backbone of modern therapeutic discovery. But each of them generates gigantic amounts of data ¾ often terabytes per sample. This demands massive computational resources, too.

Processing such amounts of data raised lots of challenges for Proteros:

1. The scientists needed tools and infrastructure for analyzing datasets of growing complexity and size.
2. The processing pipelines demand high-performance GPU acceleration coupled with interactive 3D visualization capabilities.
3. The computational demands fluctuated dramatically across different drug discovery pipelines. Highly parallel jobs were effective at using all the available on-premises resources, but those same resources remained heavily underutilized during less parallel or GPU-intensive phases, which resulted in an overall inefficient resource allocation.

In response, Proteros saw a need to modernize their compute environment to match their growing ambitions. They embraced this opportunity by adopting a more flexible and scalable cloud-based HPC solution ¾ one capable of supporting highly dynamic workloads, enabling scientific agility, and maintaining strict security and regulatory compliance.

Building a foundation

To build a solid foundation, Proteros started by setting up a secure AWS Landing Zone using AWS Control Tower. This equipped them with a strict compliance framework and advanced security controls, ensuring a safe handling of sensitive pharmaceutical research data within a growing AWS environment for years to come.

One important aspect of this foundation was an implementation of hybrid connectivity between AWS and Proteros’ on-premises data center. This enabled an easier transition towards cloud-based processing for all their scientists. This involved some key AWS services:

• Amazon Route 53 Resolver endpoints and forwarding rules ¾ To enable DNS resolution across both environments
• AWS Transit Gateway with redundant AWS Site-to-Site VPN connections / tunnels for private connectivity

AWS ParallelCluster

To match their HPC needs, AWS ParallelCluster provided a perfect framework. ParallelCluster is an open-source HPC cluster management toolkit that helps with deploying and managing all HPC-related building blocks using infrastructure-as-code. This includes compute resources Amazon Elastic Compute Cloud (Amazon EC2), job scheduling (Slurm), and filesystems (like Amazon FSx for Lustre).

The Proteros solution involved multiple Slurm queues with different Amazon EC2 instance types tailored for specific steps of Proteros’ processing pipelines.

With ParallelCluster, Proteros gained:

• Elastic scaling ¾ automatically spinning up compute resources during peak processing periods and shutting down during idle times.
• Elimination of bottlenecks that previously constrained scientific progress when multiple research projects required simultaneous processing.
• Flexibility to test with different instance types for optimal performance and cost balance.

Hybrid storage: optimizing for performance and cost

One key innovation in their solution was a new hybrid storage approach. Instead of relying exclusively on  Lustre as many conventional HPC solutions do, they developed an approach using various AWS services based on access patterns and performance requirements.

Amazon FSx for Lustre is used for specific data that requires the highest possible throughput for complex GPU processing, while Mountpoint for Amazon S3 was deployed for all other data that tolerated lower levels of performance.

Cryo-EM workflows generally access data only once during its initial motion-correction task and then convert it into smaller files for downstream processing, thus making Mountpoint for Amazon S3 ideal for a low-cost and durable storage and access layer.

In contrast, PX workflows access raw data multiple times during processing. By using Mountpoint for Amazon S3, we took advantage of local caching (making use of free disk space). Since PX data is relatively small, caching provides highly effective access with minimal performance losses.

This approach reduced their cloud storage costs by approximately 60-70% compared with using Amazon FSx for Lustre exclusively, while they maintained a high-performance bar where it mattered most.

Multi-user access and identity integration

To support Proteros’ collaborative drug discovery approach, Proteros implemented a multi-user architecture by integrating their Amazon EC2 instances with their existing Active Directory and Domain Controllers. This enabled synchronization of user accounts and group policies, allowing them to enforce strict permission boundaries and granular access controls based on each user’s role. Additionally, they used Amazon DCV to provide scientists with secure, high-performance remote visualization capabilities.

This approach ensured that every scientist could access AWS ParallelCluster and all its related components using their corporate credentials, delivering a unified operational experience.

The solution architecture

The architecture consists of several key components:

• AWS Landing Zone: Implemented via AWS Control Tower, providing all required governance, security and network controls.
• Networking: Hybrid connectivity between AWS and on-premises data center via AWS Site-to-Site VPN, AWS Transit Gateway and Amazon Route 53.
• AWS ParallelCluster: Multiple queues with different instance types for different computational needs.
• Storage: Hybrid approach with an intelligent combination of Amazon FSx for Lustre and Mountpoint for Amazon S3.
• CI/CD: Automated pipelines and a deployment process that matched Proteros’ release and staging concept for both development and production workloads.
• IAM: Integration with existing identity provider, re-using existing users and group policies.

Figure 1:  One of several workload accounts running AWS ParallelCluster, with centralized hybrid connectivity

Results and benefits

The results achieved with the depicted solution enable continuous, sustained growth for Proteros:

• Scalability: On-demand compute resources eliminated any processing bottlenecks and enabled limitless flexibility for business expansion.
• Cost efficiency: The hybrid storage approach and right-sized compute resources drastically reduced projected costs.
• Operational efficiency: Automation through CI/CD and golden image pipelines minimized operational overheads.
• Improved security: The secure landing zone and identity integration provided advanced security controls.

Conclusion

By using AWS ParallelCluster and a hybrid storage approach with automated CI/CD pipelines, Proteros achieved measurable performance gains and cost efficiencies for its Cryo-EM and PX processing capabilities.

Their solution delivers a lot of value for different types and sizes of pharmaceutical and biotechnological companies, as well as academic institutions engaged in structural biology, drug discovery, and chemistry research. Anyone in this sector requiring HPC at scale while still maintaining cost efficiency and security compliance is well advised to take a look.