Culture, Society & Family Feature Health & Medicine Learning & Education Mind & Brain Researcher news Student researchers Technology Videos

SMU Guildhall and cancer researchers level up to tap human intuition of video gamers in quest to beat cancer

Massive computational power of online “Minecraft” gaming community bests supercomputers

Video gamers have the power to beat cancer, according to cancer researchers and video game developers at Southern Methodist University, Dallas.

SMU researchers and game developers are partnering with the world’s vast network of gamers in hopes of discovering a new cancer-fighting drug.

Biochemistry professors Pia Vogel and John Wise in the SMU Department of Biological Sciences, and Corey Clark, deputy director of research at SMU Guildhall, are leading the SMU assault on cancer in partnership with fans of the popular best-selling video game “Minecraft.”

Vogel and Wise expect deep inroads in their quest to narrow the search for chemical compounds that improve the effectiveness of chemotherapy drugs.

“Crowdsourcing as well as computational power may help us narrow down our search and give us better chances at selecting a drug that will be successful,” said Vogel. “And gamers can take pride in knowing they’ve helped find answers to an important medical problem.”

Up to now, Wise and Vogel have tapped the high performance computing power of SMU’s Maneframe, one of the most powerful academic supercomputers in the nation. With ManeFrame, Wise and Vogel have sorted through millions of compounds that have the potential to work. Now, the biochemists say, it’s time to take that research to the next level — crowdsourced computing.

A network of gamers can crunch massive amounts of data during routine gameplay by pairing two powerful weapons: the best of human intuition combined with the massive computing power of networked gaming machine processors.

Taking their research to the gaming community will more than double the amount of machine processing power attacking their research problem.

“With the distributed computing of the actual game clients, we can theoretically have much more computing power than even the supercomputer here at SMU,” said Clark, also adjunct research associate professor in the Department of Biological Sciences. SMU Guildhall in March was named No. 1 among the Top 25 Top Graduate Schools for Video Game Design by The Princeton Review.

“If we take a small percentage of the computing power from 25,000 gamers playing our mod we can match ManeFrame’s 120 teraflops of processing power,” Clark said. “Integrating with the ‘Minecraft’ community should allow us to double the computing power of that supercomputer.”

Even more importantly, the gaming community adds another important component — human intuition.

Wise believes there’s a lot of brainpower eager to be tapped in the gaming community. And human brains, when tackling a problem or faced with a challenge, can make creative and intuitive leaps that machines can’t.

“What if we learn things that we never would have learned any other way? And even if it doesn’t work it’s still a good idea and the kids will still get their endorphin kicks playing the game,” Wise said. “It also raises awareness of the research. Gamers will be saying ‘Mom don’t tell me to go to bed, I’m doing scientific research.”

The Vogel and Wise research labs are part of the Center for Drug Discovery, Design and Delivery (CD4) in SMU’s Dedman College. The center’s mission is a novel multi-disciplinary focus for scientific research targeting medically important problems in human health. Their research is funded in part by the National Institutes of Health.

The research question in play
Vogel and Wise have narrowed a group of compounds that show promise for alleviating the problem of chemotherapy failure after repeated use. Each one of those compounds has 50 to 100 — or even more — characteristics that contribute to their efficacy.

“Corey’s contribution will hopefully tell us which dozen perhaps of these 100 characteristics are the important ones,” Vogel said. “Right now of those 100 characteristics, we don’t know which ones are good ones. We want to see if there’s a way with what we learn from Corey’s gaming system to then apply what we learn to millions of other compounds to separate the wheat from the chaff.”

James McCormick — a fifth year Ph.D. student in cellular molecular biology who earned his doctoral degree this spring and is a researcher with the Center for Drug Discovery, Design and Delivery — produced the data set for Clark and Guildhall.

Lauren Ammerman, a first-year Ph.D. student in cellular and molecular biology and also working in the Center for Drug Discovery, Design and Delivery, is taking up the computational part of the project.

Machines can learn from human problem solving
Crowdsourcing video gamers to solve real scientific problems is a growing practice.

Machine learning and algorithms by themselves don’t always find the best solution, Clark said. There are already examples of researchers who for years sought answers with machine learning, then switched to actual human gamers.

Gamers take unstructured data and attack it with human problem-solving skills to quickly find an answer.

“So we’re combining both,” Clark said. “We’re going to have both computers and humans trying to find relationships and clustering the data. Each of those human decisions will also be supplied as training input into a deep neural network that is learning the ‘human heuristic’ — the technique and processes humans are using to make their decisions.”

Gamers already have proven they can solve research problems that have stymied scientists, says Vogel. She cites the video game “Foldit” created by the University of Washington specifically to unlock the structure of an AIDS-related enzyme.

Some other Games With A Purpose, as they’re called, have produced similar results. Humans outperform computers when it comes to tasks in the computational process that are particularly suited to the human intellect.

“With ‘Foldit,’ researchers worked on a problem for 15 years using machine learning techniques and were unable to find a solution,” Clark said. “Once they created the game, 57,000 players found a solution in three weeks.”

Modifying the “Minecraft” game and embedding research data inside
Gamers will access the research problem using the version of “Minecraft” they purchased, then install a “mod” or “plugin” — gamer jargon for modifying game code to expand a game’s possibilities — that incorporates SMUs research problem and was developed in accordance with “Minecraft” terms of service. Players will be fully aware of their role in the research, including ultimately leaderboards that show where players rank toward analyzing the data set in the research problem.

SMU is partnering with leaders in the large “Minecraft” modding community to develop a functioning mod by the end of 2017. The game will be heavily tested before release to the public the second quarter of 2018, Clark said.

The SMU “Minecraft” mod will incorporate a data processing and distributed computing platform from game technology company Balanced Media Technology (BMT), McKinney, Texas. BMT’s HEWMEN software platform executes machine-learning algorithms coupled with human guided interactions. It will integrate Wise and Vogel’s research directly into the SMU “Minecraft” mod.

SMU Guildhall will provide the interface enabling modders to develop their own custom game mechanic that visualizes and interacts with the research problem data within the “Minecraft” game environment. Guildhall research is funded in part by Balanced Media Technology.

“We expect to have over 25,000 people continuously online during our testing period,” Clark said. “That should probably double the computing power of the supercomputer here.”

That many players and that much computing power is a massive resource attacking the research problem, Wise said.

“The SMU computational system has 8,000 computer cores. Even if I had all of ManeFrame to myself, that’s still less computing and brainpower than the gaming community,” he said. “Here we’ve got more than 25,000 different brains at once. So even if 24,000 don’t find an answer, there are maybe 1,000 geniuses playing ‘Minecraft’ that may find a solution. This is the most creative thing I’ve heard in a long time.” — Margaret Allen, SMU

Health & Medicine Videos

Drugs behave as predicted in computer model of key protein, enabling cancer drug discovery

New model allows pharmacological researchers to dock nearly any drug and see how it behaves in P-glycoprotein, a protein in the cell associated with failure of chemotherapy

Drugs important in the battle against cancer responded the way they do in real life and behaved according to predictions when tested in a computer-generated model of one of the cell’s key molecular pumps — the protein P-glycoprotein, or P-gp.

Biologists at Southern Methodist University, Dallas, developed the computer generated model to overcome the problem of relying on only static images for the structure of P-gp, said biologist John G. Wise, lead author on the journal article announcing the advancement.

The new SMU model allows researchers to dock nearly any drug in the P-gp protein and see how it will actually behave in P-gp’s pump, said Wise, an associate professor in SMU’s Department of Biological Sciences.

“The value of this fundamental research is that it generates dynamic mechanisms that let us understand something in biochemistry, in biology,” he said. “And by understanding P-gp in such detail, we can now think of ways to better and more specifically inhibit it.”

P-gp is the cellular pump that protects cells by pumping out toxins. But that’s a problem when P-gp targets chemotherapy drugs as toxic, preventing chemo from killing cancer cells. Scientists are searching for ways to inhibit P-gp’s pumping action.

The SMU researchers tested Tariquidar, a new P-gp inhibitor still in clinical trials. Inhibitors offer hope for stopping P-gp’s rejection of chemotherapeutics by stalling the protein’s pumping action. Pharmacology researchers disagree, however, on where exactly Tariquidar binds in P-gp.

When run through the SMU model, Tariquidar behaved as expected: It wasn’t effectively pumped from the cell and the researchers observed that it prefers to bind high in the protein.

“Now we have more details on how Tariquidar inhibits P-gp, where it inhibits and what it’s actually binding to,” Wise said.

SMU researchers report that their computer model simulation reveals the binding sites of Tariquidar — a P-gp inhibitor —  as the “pump” opens and closes. (Image:  James McCormick)
SMU researchers report that their computer model simulation reveals the binding sites of Tariquidar (orange blob) — a P-gp inhibitor. (Image: James McCormick)

Also using the model, the researchers discovered greater detail than previously known about the behavior of other drugs as well, and how those drugs bind in P-gp to stop its pumping action.

The study was funded in part by the National Institutes of Health. The lab was recently awarded a second NIH grant for the research.

The findings are published in the journal Biochemistry. The article, “Multiple drug transport pathways through human P-glycoprotein,” is published online in advance of print at NIH’s PubMed Central.

A still image of the modeled protein in action will appear on the cover of the October through December issues of Biochemistry.

Testing the virtual P-gp model by virtually docking real drugs
Wise and his colleagues tested one of the workhorse drugs of chemotherapy, daunorubicin, a close cousin of Adriamycin.

An aggressive chemotherapeutic, daunorubicin stops DNA replication in the cell, and is a classic target for P-gp to pump out of a cell, Wise said.

“For a long time, it’s been thought that there are at least a couple of distinct binding sites for drugs,” Wise said. “Sure enough, with our models, we found that daunorubicin, at least, prefers to bind on one side of the P-gp model, while verapamil – a commonly prescribed blood pressure medicine – prefers the other side.”

SMU researchers report that their computer model simulation reveals the binding sites of Tariquidar — a P-gp inhibitor —  as the “pump” opens and closes. (Image:  James McCormick)
SMU researchers report that their computer model simulation reveals the binding sites of Tariquidar (orange blob) — a P-gp inhibitor — as the “pump” opens and closes. (Image: James McCormick)

Not only did the researchers show computationally that there are two different starting points for drugs, they also showed that there are two different pathways to get the drugs through.

“The two different drugs start at different sites and they’re funneled to the outside by being pushed by the protein,” Wise said. “But the actual parts of the protein that are pushing the drugs out are different.”

Wise and his co-authors, SMU biologists Pia Vogel and James McCormick, created the P-gp computer-generated simulation using SMU’s High Performance Computer, ManeFrame.

Molecular model can aid in fight against multi-drug resistance of cancer cells
The capability of watching molecular machinery up close, while doing its job the way it does in real life, may spark new drug discoveries to fight cancer.

“Having an accurate model that actually moves – that shows the dynamics of the thing – is incredibly helpful in developing therapies against a molecular target to inhibit it. The only other ways to do it are blind, and the chances of success using blind methods are very low,” Wise said.

“Scientists have tried for 30 years to find inhibitors of this pump and have done it without knowing the structure and with only little knowledge about the mechanism, screening more or less blindly for compounds that inhibit the thing,” Wise said. “They found drugs that worked in the test tube and that worked in cultured cells, but that didn’t work in the patient. With our model, because we can see the pump moving, we can probably predict better what’s going to make an inhibitor actually work well.”

Vogel and Wise led a team of researchers in using the P-gp model to virtually screen millions of publically available drug-like compounds.

Verapamil (green blob), inhibits the P-gp pump. But until now, the workings of the pump could not be observed so researchers could only speculate where Varapamil “binds” in P-gp. SMU researchers report that their computer model simulation reveals Varapamil’s binding sites while the “pump” opens and closes. (Image: McCormick)
Verapamil (green blob), inhibits the P-gp pump. Until now, the workings of the pump could not be observed so researchers didn’t know exactly where Varapamil “binds” in P-gp. SMU researchers report that their simulation reveals the binding sites. (Image: McCormick)

They discovered three new drug leads that could ultimately inhibit P-gp and offer better odds of survival to prostate cancer patients. The researchers reported those findings this month in the journal Pharmacology Research & Perspectives,

New SMU model simulates molecular machinery in action
Researchers look for drug compounds that can temporarily stop or inhibit the P-gp pump, so that the chemotherapy drugs that enter the cancer cell will stay there and do the job of killing the cancer. Finding the right pump inhibitor requires understanding the pumping action. That’s difficult without seeing the pump at work.

The structures of proteins similar to P-gp have been previously available in a static state through X-ray crystallography. Scientists use X-ray crystallography as a tool that essentially draws the details of biological structures by identifying their atomic and molecular structure through diffraction of X-rays by the atoms themselves.

Scientists often contribute the resulting protein structures to the U.S. Protein Data Bank repository for public use.

Detailed data combined with several trillion calculations produced model
To build the P-gp model, Wise used structures from the repository, showing various stages of transport, to simulate four points of reference. From there, SMU’s ManeFrame supercomputer was fed parameters and characteristics of the protein as well as how it should behave physically, including when kinetic energy was added to bring the protein and its surrounding membrane and water up to body temperature. The animated model resulted from calculating differences between two structures and using targeted molecular dynamics programs to slightly nudge the model to the next step.

“You do that several million times and make several trillion calculations and you arrive at the next structure,” Wise said. “In this way, we can nudge P-gp through a full catalytic transport cycle.”

Finally, using a docking program, the researchers individually introduced daunorubicin and other drugs into the protein, and watched the drugs move through P-gp’s catalytic cycle.

“What happened was — the drugs moved,” Wise said. “And they moved the way they should move, clinically, biochemically, physiologically, to pump the compounds out of the cell.”

Vogel added that, “in some of the zoom-ins of the model, you can actually see the amino acids paddle down the drugs.”

Further challenging and testing the model
The researchers ran a critical control to further test if the model worked.

“We thought maybe anything you put in the protein, relevant or not, would get pumped through. So we put in something that is not a transport substrate of P-gp, something that biochemically would never be transported by P-gp,” Wise said. “We put it in, starting where daunorubicin is effectively pumped out, and very quickly the compound left the protein — but it left the opposite way, back into the cell. This experiment gave us more confidence that what we are seeing in these models is reflecting what happens in the cell.”

Wise admits that until he saw it for himself, even he had doubts the virtual P-gp model would behave like real-life P-gp.

“It’s a crude approximation of a complex, sophisticated human protein, but it’s so much better than the static images available now,” Wise said. “I’ve got to emphasize for all the disbelievers, for the ‘culture of doubters’ out there, that this model works — it moves the drugs through the membrane. That speaks for itself. What P-gp does in the cell, cancerous or normal, it does in our simulations.”