Mukarram Tahir

Massachusetts Institute of Technology

A science fair project – and the home-built computing cluster that followed – were Mukarram Tahir’s introductions to research.

As a high school junior, Tahir studied machine learning, in which an algorithm is fed data that trains it to recognize patterns in new data. His science fair project on the subject won prizes, but Tahir was frustrated that he didn’t have enough computer power to adequately train his model.

So he spent the following summer – and around $500 of the cash he won with his project – to build a computing cluster using 15 motherboards, each with an Intel Pentium 4 processor. “By today’s standards, it’s pretty bad,” says Tahir, now a Department of Energy Computational Science Graduate Fellowship (DOE CSGF) recipient. It impressed a professor at Virginia Commonwealth University, near Tahir’s home, enough that he took the high school student into his laboratory.

And recently, when Tahir was commissioned to build a cluster for his Massachusetts Institute of Technology research group, he referred to a paper he’d written in high school on how to set up a basic system. “Probably some of the details are out of date, but it was a pretty helpful guide.”

As a doctoral student, Tahir uses that combination of home-grown and formal training to address materials science problems, with implications for treating serious diseases.

With his advisor, Alfredo Alexander-Katz in the materials science and engineering department, Tahir applies computing to nanoparticles, atoms clumped into globs thousands of times smaller than a hair’s diameter. Alexander-Katz and his colleagues want to coat these nanoparticles, usually made of gold atoms, with compounds that interact with cell membranes, triggering membrane fusion, molecular transport and other processes.

“This is where computation is key. Computation is required to find the right chemistry to generate targeted interactions between the nanoparticle and biological matter” and a cell membrane, Tahir says. Using molecular dynamics (MD) techniques that calculate how atoms and molecules move and interact, he models new nanoparticle-coating combinations, seeking ones that work like the body’s natural proteins. Tahir tunes these nanostructures to get the best results, then passes the formulations to experimental researchers who make and test them.

In one project, Tahir sought a synthetic nanostructure that could attach to cell membranes and trigger fusion, giving scientists control over a critical step in cargo transport, pathogen interaction, reproduction and other processes.

“We don’t have a full understanding of how cells achieve this and we definitely don’t have a synthetic object that can drive this process,” Tahir says. His models found that a nanoparticle coated with a chemical that is attractive to both water and lipids (the fats in cell membranes) can start the merger of two nearby membranes. The model also suggested that when calcium ions are added to water surrounding the cells, the initial connection expands into a fusion pore. Electron microscopy showed that the nanostructures act as the model predicted.

Tahir: Fusion

The results indicate computation can be used to design nanomaterials that function as synthetic alternatives to proteins, Tahir says. With the capacity to drive membrane fusion, for example, scientists could create novel treatments that operate at the cellular level. The nanoparticle itself also could be designed to encapsulate drugs that can’t dissolve in water.

The research demands terrific computing power: MD simulations model nearly everything at the atomic scale, and Tahir must test multiple chemical-particle combinations. His calculations have run on Stampede, a now-retired University of Texas at Austin supercomputer. He’s seeking an allocation on Cori, a Cray XC40 at Lawrence Berkeley National Laboratory’s National Energy Research Scientific Computing Center.

Tahir tackled another nanotechnology application during his 2015 Argonne National Laboratory practicum. Working with DOE CSGF alumnus Stefan Wild, he focused on nanostructures that self-assemble from even smaller chemical building blocks. They turned the problem around by starting with a target structure and identifying parts that would spontaneously merge to create it. It was fundamentally an optimization problem – finding the most efficient solution from myriad options – like those Wild and his colleagues investigate.

“The group I was working with are world-class researchers in optimization,” Tahir says. He delved into the field, learning the mechanics of methods that had largely been a mystery to him. Because the research relied on many small computing jobs, he also developed a scheduler to efficiently parcel them out over an allocation of computing time.

Tahir expects to graduate in late 2018. Until then, he’s still working on how to make minuscule objects do his bidding, turning next to what happens if multiple particles cooperatively interact with a cell membrane.

Image caption: Molecular dynamics simulation of nanoparticle-mediated fusion between lipid vesicles. The net negative charge and amphiphilic (water- and fat-loving) surface chemistry allow the nanoparticles to bring the vesicles into close proximity and induce lipid mixing. Adding calcium to the surrounding water facilitates pore formation, triggering completion of the fusion process. Credit: Mukarram Tahir.