Each year, 500,000 babies are born with a genetic disorder called sickle cell disease, a chronic illness that causes patients’ red blood cells to be abnormally shaped and to stick to the walls of blood vessels. The disorder can cause blockages, debilitating pain, and even damage to the body’s organs. Though careful monitoring can help patients manage sickle cell disease, a cure remains beyond modern medicine. With the help of supercomputers, however, researchers at Brown University are trying to change that.

A team led by Brown’s George Karniadakis devised a multiscale model of sickle cell disease that captures the disorder’s molecular origins inside red blood cells. Using the Titan supercomputer—the leadership-class machine of the Oak Ridge Leadership Computing Facility, a US Department of Energy (DOE) Office of ScienceUser Facility at DOE’s Oak Ridge National Laboratory—the team modeled the disease from the ground up, devising a method that could help researchers assess and evaluate treatments to combat it. The technique, called the mesoscopic adaptive resolution scheme (MARS), selectively sacrifices detail to capture new information about sickle cell disease.

Scaling the technique to thousands of processors on the Titan supercomputer, the team bridged time and length scales separated by three orders of magnitude. “This method allows us to actually look at this process from the hemoglobin molecule all the way to the red blood cell,” Karniadakis said. “We’re trying to connect the microscopic to the macroscopic.”

Creative constraints

In 1949, American chemist Linus Pauling traced the cause of sickle cell disease to its genetic origin—a mutated protein called hemoglobin. In healthy red blood cells, hemoglobin carries oxygen to the body’s tissues and is dispersed evenly throughout the cell. But in sickle red blood cells, hemoglobin polymerizes, or binds together to form long chains, in the absence of oxygen.

The hemoglobin chains, or fibers, grow like tentacles at different rates and in different directions in a process that spans milliseconds to seconds and nanometers to micrometers. Eventually the fibers push into the cell membrane, distorting the cell into a “C” shape not unlike the classic harvesting tool for which the disorder is named. “Ideally, we would like to track every molecule at every stage at every time,” Karniadakis said. “The problem is if you do that, you’re just going to have a few molecules of hemoglobin for a very short amount of time—even if you’re using a really powerful supercomputer.”

To get around these limitations, Karniadakis’s team employs a dynamic technique called dissipative particle dynamics that models systems as a mixture of individual and collective particles. This method has the advantage of applying scarce compute resources where they are needed most. In the case of polymerizing hemoglobin, that place is at the end of a fiber, where new molecules are added to the chain.

As the fibers mature, the teams’ MARS scheme automatically dials down the resolution, cutting out the computational cost of fine interaction while retaining essential biophysics related to the fibers’ mechanics. “Unless we have this adaptive way of coarse-graining on the fly, we wouldn’t be able to look at the whole process,” Karniadakis said.

Deploying MARS simulations on the Titan supercomputer, the researchers demonstrated the link between polymer fiber growth and the various red blood cell shapes observed in sickle cell disease patients. Only around 20 percent of sickle red blood cells resemble the familiar crescent moon shape. The remainder exhibit variations. Using the simulation data, Karniadakis’s team was able to produce a polymerization profile for each of the major cell shapes associated with the disease.

Central to this achievement was the use of Titan’s GPUs, which allowed the team to compute millions of particles for each time step of the simulation.

Source: Brown University