The University of Miami’s Systems Drug Discovery Lab routinely runs large-scale, physics-based simulations to study protein–ligand binding and support early-stage therapeutic modeling for research. In a proof-of-concept study, the Lab incorporated ALAFIA’s AIVAS Supercomputer, powered by Ampere’s 192-core AmpereOne processor, reducing simulation times from over 24 hours to just a few. This enabled more efficient exploration of complex molecular interactions—such as cooperative binding mechanisms involved in ternary complex formation for targeted protein degraders.
The animation below, rendered on the ALAFIA AIVAS system, visualizes the type of complex molecular interactions studied, showing a protein pocket flexing around a drug candidate:
Challenges in research computing
As part of its high-throughput research pipeline, the Lab frequently executes computationally intensive workloads, including long molecular dynamics simulations and ternary complex predictions for PROteolysis Targeting Chimeras (PROTACs). While the University provides robust central computing infrastructure, incorporating a dedicated, compute-dense system like ALAFIA’s made it easier to launch these demanding simulations on demand—supporting rapid iteration and streamlining exploratory modeling workflows for targets ranging from cancer to infectious diseases within a research setting.
Dedicated high-performance compute
By tapping Ampere’s high-core-count, power-efficient architecture (AmpereOne), ALAFIA’s AIVAS Supercomputer enabled the University of Miami team to accelerate simulations without reconfiguring workflows or relying on shared compute queues. Researchers now run highly parallelized molecular dynamics workloads more readily, including simulations focused on binding interactions, conformational flexibility, and ternary complex prediction for targeted protein degraders. The system supports seamless prototyping and iterative analysis, making it easier to conduct proof-of-concept studies across multiple targets and chemical series.
In addition, the system’s computational capabilities extend beyond molecular dynamics. The team noted the ease of running other demanding research tasks, such as using large pre-trained Language Models (LLMs) locally on the AIVAS hardware for potential applications like research data analysis.

The PROTAC Builder represents the starting point of the accelerated workflow. Developed in-house at the University of Miami, it allows rapid design of potential HIV drug candidates before they undergo computationally intensive molecular dynamics simulations for research, drastically sped up by the ALAFIA AIVAS supercomputer. (Data/Image courtesy of the Schürer Lab, University of Miami)

Nineteen MD runs on the 192-core AmpereOne box all beat the lab’s prior baseline (red dashed line). Most topped 250 ns/day, enabling same-day research iterations. (Data/Image courtesy of the Schürer Lab, University of Miami)

Wall-clock time for 50 ns PROTAC ternary runs. Each simulation finishes in under 6 hours on the 192-core AmpereOne box, allowing multiple research runs per day. (Data/Image courtesy of the Schürer Lab, University of Miami)
Research Benefits
The lab achieved significant acceleration, described as up to a 10x reduction in runtime for key simulations in specific instances, enabling broader and deeper analysis of protein–ligand interactions and ternary complexes for its research. Access to a dedicated high-performance system simplified modeling workflows and allowed researchers to extend simulation windows potentially from tens towards hundreds of nanoseconds for certain studies, offering richer structural insights. The platform also demonstrated benefits beyond the primary simulations, accelerating Python-based analytics used in downstream data interpretation of research results, while maintaining energy efficiency—ultimately supporting faster, more flexible predictive modeling in the lab’s early-stage translational research.
The ability to run longer or more frequent simulations allows for detailed analysis, such as tracking DockQ scores over time to assess the stability of interactions within the simulated molecular complex:
About the Research Group
The University of Miami Frost Institute for Data Science and Computing’s Systems Drug Discovery Lab develops computational methods to support early-stage therapeutic research across cancer, infectious diseases, and other conditions. The Lab integrates physics-based modeling, AI-driven prediction, and large-scale data analysis to explore protein–ligand binding, ternary complex formation, and other key processes in drug discovery. Its interdisciplinary approach combines molecular simulations, cheminformatics, and machine learning to enable high-throughput, structure-informed strategies for identifying and refining novel therapeutic candidates for research exploration.
Perspectives

Tracking interaction stability (DockQ score) over the course of the simulation provides crucial insights for research. (Data/Image courtesy of the Schürer Lab, University of Miami)
“ALAFIA’s AIVAS Supercomputer, powered by Ampere’s processor, made it significantly easier for us to run our routine molecular simulations more efficiently and with fewer barriers within our research lab,” said Dr. Stephan Schürer, Professor, Department of Pharmacology, Miller School of Medicine, University of Miami. “This system gave us additional flexibility to explore protein–ligand binding and complex formation, particularly in our studies of targeted protein degraders.”
“Our AIVAS Supercomputer was purpose-built to harness Ampere’s compute-dense, power-efficient architecture,” said Camilo Buscaron, co-founder and CEO of ALAFIA. “Ampere-powered systems like ALAFIA represent an important step toward democratizing access to high-performance computing—removing the need for complex clusters or costly cloud setups. That kind of accessibility doesn’t just simplify workflows—it accelerates innovation. In the University of Miami’s case, it enables advanced simulations that drive early-stage drug discovery research forward.”
Editor’s Note: This article was revised on May 1, 2024, based on direct input and clarifications from the University of Miami research team to ensure accuracy and reflect the specific scope and context of their research-focused computational work.