MML deployed Azure HPC + AI resources on more than 18,000 Azure HBv2 VM cores for about 120 ASD use cases optimally configured for high-performance computing with efficient node-to-node interconnectivity, AMD EPYC™ processors, and NVIDIA InfiniBand-enabled SKUs. As a Microsoft Partner Network member, MML found its perfect match both in Azure and in AMD as an implementation vendor for MML’s critical high VM density. Above all, MML is proud that it has enhanced its core capabilities, particularly with respect to simulating ASD.
Revolutionizing drug development
While drug formulation has advanced to cater for remarkable new treatments, humans have often had trouble metabolizing chemically complicated drugs. As the molecules of active pharmaceutical ingredients (API) become increasingly complex, they can compromise bioavailability and the fraction of the API that eventually reaches the therapeutic site. In response, drug manufacturers use a formulation technique called amorphous solid dispersion (ASD), which involves breaking up API crystals and blending them with organic polymers to improve the solubility, dissolution rate, and bioavailability of drug delivery systems.
“Having run exhaustive benchmarks prior to committing to Azure virtual machines and CPU capacity along with Azure HPC + AI in order to understand modelling restrictions owing to system size, we are absolutely thrilled to be able to use Azure architecture, cloud resources, and consumption models for our simulations.”
Dr. Georgios Antipas, CEO, Molecular Modelling Laboratory
Although a wonder of modern pharmaceuticals, ASD has traditionally been a complicated, lab-based process that can add months to drug development cycles. For example, ASD are metastable, which puts their physical stability at risk, and if the concentration of API exceeds solubility limits, the ASD may gradually degrade or break down altogether. The prediction of ASD stability has long been considered too complex for algorithm modelling and high-throughput computational screening, and so ASD formulation has been largely limited to manual trial and error. This has resulted in limited experimentation and missed opportunities for new discoveries and intellectual property.
Led by a team of researchers highly experienced in computational materials science and electron microscopy, Swiss-based Molecular Modelling Laboratory (MML) has played a critical role in ASD pre-formulation screening, pivoting its drug research and development to small organic and biomolecular polymers. Through these molecular simulations, MML has been able to process a very large amount of possible API and polymer combinations, studying how they behave in various dispersions and conditions. This approach can reveal new and possibly unforeseen ASD combinations, enhance drug safety, and help reduce drug development time to delivery and cost.
Thus far, MML has achieved spectacular results with its ASD simulations and molecular modelling, frequently managing to predict API solubility limits close to the equivalent of a full-scale lab experiment and helping its customers fine-tune drug candidate structures. When scaling up the solution, MML discovered that the calculations and simulations were too complex and CPU-intensive to run on in-house infrastructure.
“We needed to put together a solution based on the best possible implementation of underlying physics and the most cost-efficient high-performance computing execution,” explains Dr. Georgios Antipas, CEO of Molecular Modelling Laboratory.
MML turned to Microsoft Azure HPC + AI and Azure Virtual Machines to deploy virtual machines (VMs) powered by AMD EPYC™ processors to scale up its capacity for modelling simulations and drive down delivery time. “From a computational perspective, we’re in ever-increasing demand for processing power and efficient data storage resources to optimize solution time to delivery and cost, which led us straight to Azure HPC + AI and VMs,” says Antipas.
Higher efficiency, lower costs with Azure HPC + AI
MML deployed Azure HPC + AI resources on more than 18,000 Azure HBv2 VM cores for about 120 ASD use cases optimally configured for high-performance computing with efficient node-to-node interconnectivity, AMD EPYC™ processors, and NVIDIA InfiniBand-enabled SKUs. MML uses the VMs to optimize high-throughput drug screening and API solubility limit detection without prior experimentation, aiming to eventually alleviate traditional development hurdles. Companies screening potential drug candidates could use MML simulations as a pre-formulation decision support tool or even select optimal excipient blends for specific APIs—all made possible thanks to the use of the massive scale out architecture offered by Azure.
Drugs may behave differently under different conditions, including temperature, humidity, altitude, and pressure. Pharmaceutical manufacturers must identify the capacity of each potential medium for sending the drugs anywhere, from the Arctic to the tropics. Through the massive computing resources MML gained with Azure HPC + AI, it unlocked the ability to rapidly model ASD structures against a vast array of conditions.
“It was only when we first got our hands on Azure Virtual Machines and deployed VMs with AMD EPYC™ processors that we managed to streamline our ASD research and development and bring it down to a very digestible cost, which is important for us and our customers,” says Antipas. Adds Dr. Nikolaos Ntallis, Research Scientist at MML, “We now use Azure HPC + AI to run simulations at one-tenth the cost compared to earlier iterations of our modelling platform.”
High VM density, high-performance processors
As a Microsoft Partner Network member, MML found its perfect match both in Azure and in AMD as an implementation vendor for MML’s critical high VM density. “High VM density means a large number of CPU cores on the same motherboard. It’s particularly important for us because that’s what drives down delivery times for calculating energetics and performing large-scale molecular dynamic simulations or drug degradation and chemical reactivity modelling via techniques that can address the mesoscale,” says Antipas. “These supercells that we’re using are so large that we need to break them up across multiple virtual machines.”
Although currently using HBv2 VMs, the company is evaluating Azure HBv3-series VMs, which feature third-generation AMD EPYC™ processors that provide top-of-the-line bandwidth and enhanced performance with 120 cores per server. “Having run exhaustive benchmarks prior to committing to Azure VMs and CPU capacity along with Azure HPC + AI in order to understand modelling restrictions owing to system size, we are absolutely thrilled to be able to use Azure architecture, cloud resources, and consumption models for our simulations,” says Antipas.
MML’s projects range from classic molecular modelling to first-principles (atomistic) modelling, addressing not only drug formulation but also stability and reactivity in organic and inorganic compound classes. These can require anywhere from three to four nodes all the way up to tens of thousands of cores per use case. To prove that its simulations could be done without disruption, the company conducted a pilot implementation in which it predicted drug chemical reactivity running a large range of elementary chemical reactions, utilizing in excess of 20,000 cores in parallel. “This sort of pilot implementation had tremendous potential because, in addition to drug degradation, it’s also relevant to adsorption, combustion, the construction of system-specific reactive force fields, the prediction of organic polymorphs, and a variety of other chemical challenges,” says Antipas. “That implementation proved to us and to our customers that we can have a foundation on Azure HPC + AI bundling up all sorts of technologies, which we can then mold into more segment-oriented solutions.”
From startup to industry heavyweight
MML’s adoption of Azure HPC + AI helped it transition from a small startup making major breakthroughs to an established company working with some of the top pharmaceutical manufacturers in the world in a very short time—without needing significant hardware investments. The company recognizes that without its new computing environment, it couldn’t take on these sorts of large projects. Now, it can flexibly scale and build a cloud roadmap and infrastructure fit for any future. “Having node-to-node interconnectivity with Azure HPC + AI and VMs maximizes our compute capacity and processing power, enabling us to do more and ensure reduced time to delivery, the cornerstone of our business offering,” says Ntallis.
Things are only looking up for MML as it dreams big for what’s ahead. “Despite our ability to simulate large atomic systems, even a one-million atom super cell is particularly small in both mesoscopic and, of course, continuum terms compared to sample quantities measured in lab experiments,” says Antipas. “So, our aim within the next couple of years is to be able to simulate rare events in mesoscopic-sized systems under thermodynamic bias. I project that we will do this via Monte Carlo–based methods and obviously on very high VM density and Azure HPC + AI.”
Above all, MML is proud that it has enhanced its core capabilities, particularly with respect to simulating ASD. “It is a great challenge to reconstruct physically relevant atomic systems and predict properties that can be measured in the lab straight from an atomic system,” Antipas explains. “And, operationally, it is particularly important that we execute this in a cost-efficient manner—we’ve done exactly that with Azure.”
Learn more about Molecular Modelling Laboratory on LinkedIn.
“Having node-to-node interconnectivity with Azure HPC + AI and VMs maximizes our compute capacity and processing power, enabling us to do more and ensure reduced time to delivery, the cornerstone of our business.”
Dr. Nikolaos Ntallis, Research Scientist, Molecular Modelling Laboratory
Follow Microsoft