This year, the Allen Institute for Brain Science (AIBS) hosted a unique data challenge called MapMySpikes where participants were tasked with mapping “mystery” datasets featuring electrophysiological data from individual cells onto AIBS Patch-seq datasets, which include both electrophysical data and transcriptomic data.

They were challenged to create an algorithm and user-friendly tool to map the data, and the goal was to make connections between cells’ electrical and molecular properties—a current challenge in the neuroscience field.

This was a tricky task, including for the winner, Sam Mestern. “The most challenging aspect was handling the differences between the datasets as a result of different [chemical] solutions, temperatures, and everything. I really didn’t know if it was possible when I started out.”

Sam’s winning solution PatchOTDA (Patch Clamp Optimal Transport Domain Adaptation) won both parts of the data challenge; it had the highest accuracy for mapping the mystery cells onto the AIBS Patch-seq datasets and was voted the most user-friendly tool by our judging panel. Sam is a Ph.D. student at Western University (Ontario, Canada) in Dr. Wataru Inoue’s lab, using electrophysiology (in vitro patch clamp and dynamic clamp) and in silico modeling to study how stress neurons in the hypothalamus perform computations.

To align the electrophysiology datasets, he turned to techniques used in other neuroscience fields. “My solution was most inspired by the transcriptomic field. They do a lot of research related to integrating data from across species and labs. I wanted to look there to see how they’re accomplishing their data integration and that led me into the field of domain adaptation,” he said. “I wondered if this is a problem where we see a shift in the data distribution, so we can probably use optimal transport-based solutions to get things aligned and integrated together.” Domain adaptation is a technique in machine learning focused on adapting a model trained on one dataset to work on others.

I think data challenges like this can bring a lot of fresh ideas and perspectives since they are call-outs to the major neuroscience community—it can bring in a lot of people who are maybe on the outskirts.

Sam Mestern

Ph.D. student at Western University

Upon hearing about the challenge, he was excited as it was relevant to his research and he was already familiar with Allen Institute data. “We use it at all stages of research from theorizing, to hypothesis generation, to validation. We use the Connectivity Atlas and the ISH Atlas when we’re planning injections for circuit tracing. Most of our analysis code for electrophysiology is based off of the Allen Institute’s open electrophysiological analysis tools,” said Mestern.

“Overall, the Allen Institute tools have made my work much more efficient and much more straightforward—they’ve eased a lot of my anxiety because I know I have them to reference when I want to validate something or make sure the tool is working like intended.”

Making the science accessible

Even though Sam is a proponent of open science having used many open datasets, tools, and code himself, it was a new experience being a contributor, as the data challenge involved creating a user-friendly tool. “Learning to package, build, document my code and write it in a way that the public could use and comprehend was a whole new challenge for me, because most of the research code I write here I just use personally,” said Mestern.

He also commented on the benefits of open data challenges like MapMySpikes: “I think data challenges like this can bring a lot of fresh ideas and perspectives since they are call-outs to the major neuroscience community—it can bring in a lot of people who are maybe on the outskirts.”

Sam was invited to give a live webinar on November 20 where he showcased PatchOTDA, which features a Python package and a web-based tool.

Although the data challenge has ended, Sam plans to continue working with his PatchOTDA algorithm. “We’ve always been interested in integrating datasets across species and labs, and we’re really interested in applying this technique I developed to our electrophysiology data and hopefully drive some novel insights.”

The MapMySpikes data challenge was supported by the National Institute of Neurological Disorders and Stroke of the National Institutes of Health under Award Number U24NS133077. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.