X

The latest cloud infrastructure announcements, technical solutions, and enterprise cloud insights.

New computer models from UC Davis could help make drugs safer for your heart

Hispanolitcs/Getty Images

Months after being tested under an emergency use authorization to treat COVID-19, the American Heart Association reported the pharmaceutical cocktail of hydroxychloroquine and azithromycin to have the potential to cause drug-induced heart complications for some patients with cardiovascular disease.

While the pressure of a global pandemic put this decision in the spotlight, drug makers and regulators live with this tension all the time, balancing the need to get drugs tested and approved quickly with the possible risks. The threat of drug-induced cardiac health risks, cardiotoxicity, stands among the most common risks to weigh.

Researchers at the University of California (UC) Davis School of Medicine are working to spot such heart risks sooner in the drug development and testing processes, using computer models that assess drugs in context to a person’s entire physiology. Potential factors include protein molecules, cells, tissues, organs, the person’s sex, and any pre-existing heart conditions.

Igor Vorobyov, Assistant Professor in the Department of Physiology and Membrane Biology and the Department of Pharmacology at the UC Davis School of Medicine

“This work is especially critical because even the slightest modification to a drug’s chemistry could alter its effects from therapeutic to lethal,” says Igor Vorobyov, a UC Davis assistant professor in the Department of Physiology and Membrane Biology and the Department of Pharmacology.

Running molecular dynamics simulations takes enormous computing power, far more than the researchers can access from the university’s on-premises resources. Their research involves 500 million energy and force computations on more than 100,000 different atoms for a typical simulation.

Today, working with Oracle for Research, UC Davis research teams run multiscale molecular dynamics simulations on high-performance computing (HPC) from Oracle Cloud. Staff and student researchers can run multiple simulations simultaneously in the cloud from their laptops or desktops.

Colleen E. Clancy, Professor in the Department of Physiology and Membrane Biology and the Department of Pharmacology at the UC Davis School of Medicine

“We can run 50 different simulations at once, which allow us to test all sorts of conditions and ensure that our research is not limited by the speed of our simulations,” says Colleen E. Clancy, a UC Davis professor of Physiology and Membrane Biology and professor of Pharmacology.

From atoms to rhythms

Because cloud computing resources let researchers analyze so many more variables at such a large scale, people like the UC Davis team are challenging some long-held assumptions in drug testing. They’re questioning the idea that all cardiotoxicities, which are typically measured by QT intervals, the rate at which heart muscles contract and relax, are dangerous.

“Grapefruit juice causes QT prolongation, but it’s not exactly a trigger for cardiac arrhythmias,” Vorobyov says. “Yet, the medical industry has historically applied cardiac drug safety tests that rely almost entirely on the basis of hERG blocking and QT prolongation.”

If a drug blocks ion flow through the cardiac potassium channel protein hERG, preventing it from maintaining normal heart functions and leading to prolonged QT intervals, it can create the risk of severe arrhythmias (abnormal heart rhythms) and sudden cardiac death. But Vorobyov argues that truly assessing cardiotoxicity risk requires looking at many more variables and other risk factors.

Vorobyov and his team of researchers begin their study by developing a “ready-to-go recipe” that simulates how a drug interacts with atoms and molecules of human proteins. They then collaborate with Clancy’s lab to connect to functional models developed in her lab, where cardiac cell and tissue simulations are performed. The predictions of the computer models are then compared to clinical data from electrocardiogram results of patients.

Since the molecular dynamics simulations involve millions of individual time steps of a tiny fraction (a millionth of a billionth) of a second, the researchers rely on hundreds of compute cores and Nvidia Tesla P100 and V100 GPUs, which they can quickly provision in a bare metal instance on Oracle Cloud Infrastructure (OCI). Using cloud infrastructure gives researchers access to these kinds of new technologies, in the size and shape they require, without the overhead of procuring and maintaining a changing mix of hardware.

“Oracle’s high-performance computing (HPC) platform not only helps us run atomic-resolution investigations of ion channel functions and ion channel-drug interactions. We’re also able to accurately predict the safety and efficacy of preclinical drugs,” Vorobyov says.

A critical link

While running microsecond-long molecular dynamics simulations of atomic-resolution structures to test drugs for cardiotoxicity, it’s also important to link these atomistic models to millisecond- and second-long simulations of “functional” models, including channels, cells, and tissues. But because of differences between the atomistic and functional time scales, such linkages have been a perplexing problem for researchers to solve.

“Even on the fastest supercomputer in the world, you wouldn’t be able to run a molecular dynamics simulation long enough to connect it to a higher-level functional scale and get any meaningful data,” says Clancy. Supercomputers are typically designed to run one specific task really fast. In contrast, the cloud’s distributed architecture makes it much easier to process multiple tasks simultaneously, and it can scale faster by provisioning more capacity without buying more hardware. Cloud architecture is ideal for the application of multitasking computational workflows.

Because UC Davis researchers need to solve multiple differential equations on many different time scales, they require a high level of precision to process those calculations and a high-performance architecture to run them at each scale. To get the right level of performance to link the molecular and functional scale models, Clancy’s team runs their simulations on an OCI bare metal Compute instance, using 12-core Intel Xeon CPUs.

What’s so powerful about OCI’s HPC platform, Clancy says, is how the team can apply it to so many different kinds of research challenges. They’re already looking ahead to new use cases.

“It’s infinitely expandable and able to support our research across a wide variety of problems,” Clancy says. “In the next phase of this project, we’ll be running simulations to test drugs that target adrenergic receptors, so we can better understand how the brain controls the heart.”

The beat goes on

Parya Aghasafari, Post Doctoral Scholar in the Department of Physiology and Membrane Biology at the UC Davis School of Medicine

UC Davis postdoctoral scholar Parya Aghasafari examines cardiotoxicity from another angle, using artificial intelligence built into OCI Data Science to analyze heart rhythms when testing drugs.

Aghasafari built a deep learning framework to predict heartbeat patterns for “drugged” and “drug-free” heart muscle cells. This framework enables cardiologists to decide whether a drug is safe for human use by testing the drug on engineered heart muscle cells and interpreting their observations to reflect any impact on a human heart. During the simulation, Aghasafari looks for irregular heart rhythms, such as beat-to-beat instability, the most significant characteristic of drug-induced cardiotoxicity.

To train the deep-learning network, Aghasafari uses a preinstalled virtual environment on OCI Data Science, which has a Jupyter Notebook integrated development environment and all the Python programming language and machine learning libraries, including PyTorch, NumPy, Pandas, and scikit-learn. She created a bare metal instance on OCI, provisioning Nvidia Tesla V100 GPUs, and used the Oracle AI all-in-one image for Data Science to run her machine learning model in hundreds of experiments and tests.

“I was able to complete the machine learning simulation in about 700 seconds. That’s twice as fast as when I was running it locally,” Aghasafari says. Aghasafari then takes her completed machine learning simulations and validates them with patient-specific experimental data.

While the UC Davis researchers have focused on determining drug-induced cardiotoxicity, they believe that these models can help drug makers and clinicians tackle all sorts of diseases, from cancer to metabolic disorders to inflammation.

Aghasafari is inspired by the wide potential she sees for the work she and her colleagues are doing. “Our models aren’t just repeatable, they can be applied broadly and translated to different ages, male and female sexes, and animal species,” she says.

References

Be the first to comment

Comments ( 0 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.Captcha