Background
BioSimulytics, a spin-out from University College Dublin founded in November 2019, has developed an innovative software solution that uses neural network technology and high-performance computing to predict crystal structures of new drug molecules. This breakthrough technology addresses a critical challenge in pharmaceutical development, potentially reducing drug development timelines by up to 50%.
In collaborating with the HPC team at AWS and fourTheorem we are rapidly scaling the appropriate architecture to deliver full pipeline computations and accelerate ourplatform build. fourTheorem’s past experience in handling these kinds of complexities has been invaluable to us.
The Challenge
As BioSimulytics moved toward commercialising their research, they faced several key challenges:
- Computational Scalability: Their drug molecule simulations required significant computing power that needed to scale on demand.
- Cost Management: Balancing high-performance computing needs with budget constraints.
- Technical Expertise Gap: Needed to develop a robust architecture and technical strategy.
- Workflow Optimisation: Required streamlined processes for their complex computational pipeline.
The Solution
Working with fourTheorem, BioSimulytics implemented a refined architecture centred around AWS ParallelCluster to enhance their computational capabilities:
- Elastic High-Performance Computing: AWS ParallelCluster provided the foundation for running complex Density Functional Theory (DFT) calculations using CP2K software.
- Optimised Resource Management: Implemented separate compute queues for different workloads (CP2K computations, model fitting, and crystal state predictions).
- Cost-Effective Infrastructure: All cluster resources were designed to be disposable, with clusters created or destroyed as needed from a workstation node.
- Shared Storage Solution: Utilised FSx for Lustre connected to S3 for
efficient data management across the computational workflow.
The Outcome
Accelerated Research
Enabled faster processing of complex molecular simulations.
Reduced Time to Market
Streamlined the drug development process with more efficient computational workflows.
Cost Optimisation
Pay-only-for-what-you-use model with disposable cluster resources.
Enhanced Collaboration
Improved environment for researchers and developers to work together.