Researchers from the University of Massachusetts system have partnered with Symmetric Computing, Inc. to accelerate the search for potential therapeutic agents for the treatment and possible prevention of Alzheimer’s disease (AD).
The Core Scientific Research
The protein monoamine oxidase B (MAO-B) is believed to play a critical role in the disease in two ways. First, the normal function of MAO-B, the disposal of excess neurotransmitter molecules, produces hydrogen peroxide which is toxic to neurons. Second, it appears that MAO-B and products of MAO-B action disrupt the normal mechanism responsible for processing the APP (amyloid precursor protein) into its mature form, leading to the formation of the plaques and tangles associated with Alzheimer’s.
The goal of the research is to find a drug that inhibits the activity of MAO-B by interacting with its allosteric (control) binding site. If discovered, such a drug might delay the onset and progress of the disease. It would also be an important treatment for Parkinson’s Disease,
However, few drug candidates targeting MAO-B have made it through clinical trials due to toxicity or lack of effectiveness.
Enter Virtual Screening
Virtual drug screening has the potential to accelerate the development of new treatments. Using molecular docking, molecular dynamics and other algorithms, researchers can quickly screen for new drug candidates. This saves the enormous expense and time that would have been required to make the same conclusions about those candidates in the lab and in clinical trials.
Most importantly, virtual drug screening helps researchers significantly narrow the scope of potential new drug candidates that hold promise for further investigation. However, developments in this field require massive compute and storage capacity, and high-performance infrastructures to speed computational workflows.
The Symmetric Computing team believes an approach that combines molecular docking with molecular dynamics can identify novel allosteric binding sites on the MAO-B protein and produce promising new lead compounds for drug development for Alzheimer’s, Parkinson’s and other diseases.
The Compute Platform
The computational platform is an integrated solution composed of a hardware and software component. The hardware component is the Ada departmental supercomputer, a large shared memory GPU cluster designed for high throughput single precision floating point calculations. The software component is a pipeline of software applications that implement a virtual drug discovery process. The application software programs used for the pipeline include:
- AutoDock Vina, a molecular docking program
- NAMD, a molecular dynamics program
- TensorFlow, a machine learning toolkit
Each of these software packages is optimized to run on the Ada supercomputer. The pipeline incorporates a custom-designed drug discovery database and associated visualization tools. The database includes:
- A small molecule database with 3D structures of 500 million compounds.
- A protein database consisting of all known human proteins along with all available 3D structures. In addition, there are proteins from medically relevant microbes.
- A database of known drugs both deployed and experimental.
The Ada supercomputer is composed of a head node and 3 to 6 compute nodes. The head node contains two AMD EPYC™ processors, 2 TB of memory, and three 100 Gb/s dual-port InfiniBand® adaptors. Each compute node includes one AMD EPYC processor, 64 GB of memory, one InfiniBand adaptor, and four Radeon Instinct™ MI25 Accelerators, which combine AMD GPU technologies and the ROCm software platform.
The head node memory is shared with the compute nodes. Each GPU compute node has a floating-point performance of 100 TFLOPS half precision and 50 TFLOPS single precision.
The combined Ada hardware / software compute platform is called the Virtual Drug Discovery Platform or VDDP.
Getting to Results
A novel allosteric binding site was discovered and characterized for MAO-B using the molecular docking and molecular dynamics tools. Targeting this novel site, the large-scale screening of small molecule catalogs is being conducted for drug discovery.
In addition to elucidating novel binding sites, the molecular docking and molecular dynamics programs allow the research team to perform off-target screening to eliminate molecules that may interact (non-specifically) with other proteins, which would cause a deleterious effect in vivo. This can reduce the number of late stage failures of drug candidates undergoing clinical trials due to undiscovered toxicity.
To learn more about Symmetric Computing solutions & services, visit:
To learn how to accelerate your scientific workloads, visit: https://www.amd.com/en/products/epyc-server-customers