PozeSCAF cuts simulation runtime by over 50%, accelerating drug discovery

PozeSCAF Discovery Solutions focuses on the preclinical stage of drug discovery. Through AxDrug, its proprietary end-to-end platform, PozeSCAF integrates AI and computational chemistry methodologies to accelerate hit identification, hit-to-lead, and lead optimization processes. Its mission is to harness AI for faster and smarter drug discovery, bringing life-saving medicines to patients quicker.

Time to market is critical in drug discovery, where faster pipelines mean faster development of medications. Bhanukishore Kallepalli, chief data scientist at PozeSCAF Discovery Solutions, explains, “It can take 12−15 years to get a drug into production and available after the discovery phase, so accelerating discovery is essential.”

PozeSCAF relies on the open-source GROningen MAchine for Chemical Simulations (GROMACS) software for molecular simulations. However, an older version of GROMACS created scalability and performance challenges, driving up compute costs. Molecular dynamics simulations took more than 30 hours to complete, slowing the discovery and development of new drug candidates. “We needed a more optimized pipeline to run more simulations in less time,” Kallepalli says.

To boost performance and lower costs, PozeSCAF turned to AWS. “We had already worked with AWS in other areas of the business, and we knew they could help us with an optimized platform for GROMACS,” says Kallepalli. The project began with defining performance goals and creating test cases to measure runtime and cost. From there, two priorities stood out: choosing the right Amazon Elastic Compute Cloud (Amazon EC2) instances for GROMACS workloads and upgrading to the latest version of the software.

Working closely with PozeSCAF, the AWS team conducted benchmark tests of different Amazon EC2 instances, including Amazon EC2 G4dn, G6, and G6e GPU instances and Amazon EC2 HPC7a and C8g CPU instances. “The benchmarking exercise revealed that G6e.8xl offers the best performance, but we sometimes use G6.xl instances for further cost optimization, depending on our needs,” says Kallepalli.

PozeSCAF also fine-tuned GROMACS parameters using GPU acceleration for maximum efficiency. “Upgrading the software helped improve performance even more,” explains Kallepalli. As part of its solution, PozeSCAF uses a Slurm cluster on AWS to run large-scale compound screening with AWS ParallelCluster and AWS Batch. 

By optimizing molecular dynamics workloads on AWS, PozeSCAF reduced simulation runtime by more than 50 percent—from 30 hours to under 15. Its teams now run about 2.5 times more simulations in the same timeframe, cutting new hit discovery time by an estimated 5 percent. In a preclinical phase of 3–5 years, this saves approximately 2–3 months. “Our goal is to cut the preclinical phase to 1–1.5 years, and AWS has been key in optimizing our compute processes, which has helped us accelerate our drug discovery and development pipeline,” Kallepalli says.

Additionally, the more efficient pipeline has reduced compute costs by 25−30 percent, cutting overall project costs by about 10 percent. “With these savings, we can offer our clients better prices in addition to shorter discovery timelines, giving our services a clear competitive edge,” Kallepalli notes.

PozeSCAF has started exploring AWS generative AI technology, using large-language models through Amazon Bedrock to build knowledge graphs from project data and integrate agentic AI into its workflows. “When we identify potential molecules, these knowledge graphs may help us flag possible side effects or other issues early,” says Kallepalli. “In some cases, we could make these predictions before clinical trials begin. Overall, it’s been great working with the AWS team to use the latest technologies to advance innovation. We’re excited to expand our use of AWS services as we continue to grow.”