Harnessing complexity to advance epilepsy research: Learning the language of spike-wave discharges

Jesse A. Pfammatter was awarded the Epilepsia Open® Prize 2020 Basic Science Prize for An automated, machine learning-based detection algorithm for spike-wave discharges (SWDs) in a mouse model of absence epilepsy. Jesse A. Pfammatter, Rama K. Maganti, Mathew V. Jones. Volume 4, issue 1. EPI4-0070-2018.R3

Watch a recording of the June 17 Journal Prize Symposium on YouTube

Clinical Spike-wave discharges (SWDs) are EEG patterns seen particularly during absence seizures, a nonconvulsive type of seizure. Though they were first identified more than 100 years ago, many aspects of SWDs still are being discovered.

Despite the time and effort, the standard of practice both clinically and in research is to manually identify spike-wave discharges, versus using software. Jesse Pfammatter, a post-doc in Mathew V. Jones’ lab at the University of Wisconsin, didn't subscribe to this standard for a couple of reasons. First, Pfammatter hadn’t been an epilepsy researcher for very long. Second, he was self-professedly “terrible” at manual identification.

Pfammatter also had noticed that the practice seemed quite subjective. “Every human is different in their scoring of these (EEG records),” said Pfammatter. “Not only are they different from other humans’ scoring, but people change their minds about a discharge depending on its context. If it’s near other events, people tend to mark it as epileptiform, whereas if it’s by itself, they don’t.”

The group’s research also found that expert human scorers could change their minds about a specific EEG waveform upon repeated viewing. “Sometimes, a single expert would disagree with themselves quite often, depending on what waveform they were looking at,” said Jones. “This suggests that human scoring of EEG waveforms is not entirely reliable.”

There’s useful information in variability

Rather than give in to the idea that SWDs are merely difficult to categorize, Pfammatter saw the complexity as potentially useful. “We put a lot of things in categories that don’t necessarily fit in categories,” he said. What if the subjective nature of the discharges could be harnessed?

Pfammatter and colleagues developed an automated detection algorithm that uses machine learning to highlight the variability in SWDs, rather than ignore it. “The software uses a probabilistic scale, not a categorical scale,” explained Pfammatter. “Our research assumes there’s useful information in variability.”

The algorithm’s probability scores correlated with the variability seen in human scoring. The group also found that higher-probability events more strongly related to the physiology of the mice, particularly in terms of sleep-wake transitions, which are associated with many epilepsies.

Next, the group plans to apply the algorithm to experimental situations in which SWD variability provides information about physiology. “For example,” said Pfammatter, “We know that treatments are effective across a spectrum—it’s isn’t just yes, they stop seizures or no, they don’t. We’re hoping we can test some of these signals to see if they change in the presence of certain therapeutics. That could give us insight into how treatments are working. And maybe we can start to understand why a medication works in one animal but not in another.”

The lab also works with models of post-traumatic epilepsy. “After a trauma, there’s some probability that someone will develop epilepsy, in some unknown amount of time,” Pfammatter said. “We’re finding subtle EEG markers that might indicate which individuals might go on to develop epilepsy. It’s a higher-resolution understanding.”

From entomology to epilepsy

Back to that first point, that Pfammatter hasn’t been an epilepsy researcher for very long. Pfammatter grew up in a northern suburb of Chicago and earned a PhD in entomology at the University of Wisconsin. He studied bark beetle ecology and took extensive graduate coursework in statistics and biometrics.

Toward the end of his graduate career, he met associate professor of neurology Matthew V. Jones at a bar frequented by graduate students and faculty members. The two began talking and learning about one another’s research. Though they were studying different topics, both were immersed in multivariate data. “His research was really interesting to me,” Pfammatter said. “There are similarities between community ecology and neuroscience, and the statistical analysis tool set is similar.”

Pfammatter and Jones eventually agreed that they might work well as a team. “We were working on different subjects, but we had a lot in common; it’s worked out fabulously,” said Jones. “There was a learning curve, but it’s all gone very well. I’m really proud of Jesse.”

Beyond the lab, Pfammatter loves outdoor adventures and likes to paint mountain scenes, but lately there hasn’t been time. He and wife Rachel recently finished home renovations and have just embarked on their next big adventure—the birth of their son.