Graduate Students’ Class Project Produces New Tool for Neuroscience

As new graduate students in the neuroscience department, Kenneth Latimer and Jacob Yates did a class project in a business class that eventually resulted in a prestigious publication in the journal Science, as well as a new tool for neuroscience.

Their discovery – about how neurons interpret visual cues – began in the class of Carlos Carvalho, an associate professor in the McCombs School of Business and the College of Natural Sciences. The class focused on statistical techniques economists use to tease out trends from stock market data, and it included an assignment to use the techniques to learn something from non-economic data.

Yates proposed that he and Latimer work together to test a long-held assumption about how the brain works, using data on the activity of neurons collected in the lab of professor Alex Huk.

Scientists had assumed that individual neurons interpret which way an object is moving — such as a car on a foggy day — after going through stages of indecisiveness, gradually becoming more certain about what it is the brain sees. In other words, a neuron's electrical activity would arc upward as it became more and more certain.

Latimer and Yates, applying statistical techniques to data gathered by former graduate student researcher Miriam Meister, proved instead that individual neurons act more like a switch, flipping from undecided to decided. When many neurons register their decisions spaced out over time, the end result is roughly the same — a gradual ramping up of activity until a consensus is reached. But the individual neurons do it in an unexpected way.

"This project was driven by the students," says Huk. "They showed a lot of independent initiative and pulled it off."

Yates notes this is not how science is normally done. Instead, a faculty member usually gets funding for a specific project and then recruits graduate students to work on it.

"In this case, the project didn't exist," says Yates. "We created it for a class project and took advantage of resources we already had. So instead of coming from a grant, it will probably end up generating a grant."

Huk and former UT Austin assistant professor Jonathan Pillow, co-mentors of both Latimer and Yates, intend to follow up this work in a collaborative grant.

Besides disproving a long-standing assumption in neuroscience, the researchers have created a computational tool that can help neuroscientists probe ever-deeper mysteries about the brain. To come to their conclusion, the team needed to sift through a huge data set in new ways. Analyzing the full amount of data would have taken weeks on a desktop computer using standard programming, and many researchers might simply try averaging the data from many trials to make the computations quick and easy.

But that simplification might obscure what individual neurons are actually doing. So Latimer, whose undergraduate degree was in computer science, took a different approach. He wrote code that took advantage of the way graphics cards split up computations into many small pieces and run them in parallel. In a sense, he could get supercomputer performance from a personal computer.

"We wanted this to be something we could do in the lab and something that we could share with other labs," says Latimer.

Because the code is open source and can run on relatively affordable desktop computers, Latimer says it has the potential to be widely used. The researchers are now exploring ways to expand their analyses from the activity of single neurons to that of many neurons simultaneously.