Using Graph Builder to Improve Data Quality in Fiber Photometry Data Sets
Fiber photometry is a cutting-edge technique that captures real-time in-vivo brain activity in laboratory animals. Often aligned with video-tracked behavioral data, fiber photometry allows scientists to directly pair observed behavior and brain signaling. While fiber photometry is useful for novel behavioral experiments, each experimental step (data collection, processing, analysis) introduces the possibility of error. As such, ensuring research is reproducible is critical, but the size of the data sets (up to 0.25 gigabytes) makes quality control a cumbersome task.
JMP Graph Builder provides an excellent interface for dynamically and efficiently identifying quality control issues. In a project using fiber photometry to understand how norepinephrine signaling accompanies fear behavior in rodents, we used JMP to detect misalignment between the Ethovision XT-tracked behavior data and the fiber photometry data, to provide easy identification of behavioral tracking disruptions and to ensure that expected patterns were present. These quality control checks allowed for timely understanding of, intervention to, and correction to the data set, thus promoting research integrity. Additionally, the quality control plots gave scientists a novel and insightful way to understand their experiments.
Graph Builder provided data quality checks for a sound analysis via JMP’s modeling platform. The scientists were also able to use JMP scripts and were motivated to learn how to use JMP themselves from this process. In this talk, we showcase the problem of quality control for large, complicated data sets, as exemplified through fiber photometry, and how JMP’s graphing capabilities allowed us to ensure quality data and reproducible research.
Hello, everyone. My name is Kate Konrad. Today, I'm going to be telling you about my project Using JMP Graph Builder for Improving Data Quality in Longitudinal Fiber Photometry Data sets. I'm a statistical consultant at DLH, and we work with scientists at the National Institute of Health on this project. My collaborators were working with fiber photometry. This is a really cool technique where you can watch nerves fire.
The scientists are able to add a fluorescent dye to these animals' brains. As those brains fire and different neurotransmitter signals come out, a camera will watch them fluoresce. It's a very new technique. It's very cool. It's on the cutting edge of neurobehavioral research, and it lets us understand the brain in a new and different way. However, with these data sets, they're processed in a couple of different stages.
There's a couple of different softwares that happen between the camera in the mouse's brain and when I get it to analyze it. This isn't a bad thing, but it just means that there's a higher possibility for error in the various different algorithms and classifiers that happen between the camera and my data set. We want to be able to QC it to make sure that everything in that process happened as we expected.
These data sets are also large, which makes it harder to catch errors if they're in there. With this experiment, it was done with EthoVision XT software on 41 different mice, and we have one recording for every 0.04 seconds for 10 minutes. We also record these animals three different times, so each of them has three different 10-minute segments.
While we record the behavioral measures, this includes things like the animals moving around the space, whether or not they're really stretched out or whether they're scrunchy, whether they're not moving, if they're moving really fast. While we record those, we're also recording the neurotransmitter measures, and this is the fiber photometry part. We're watching the norepinephrine and the dopamine signals in these animals' brains as they move. We're watching them fire.
That's one thing that we want to make sure. We want to make sure that we have their movement and their neurotransmitter signals aligned so that we know what the body is doing while the brain is firing. Our goal with this project, we wanted to identify any misalignment in this data. We wanted to ensure that the patterns that we expected to be present were in fact present, and through this, to promote research integrity.
I'm going to show you three different QC examples of how we took this fiber photometry data set, used JMP Graph Builder, and together got these high-quality data sets for reliable research. Let's start with our first QC. We're going to go to JMP Graph Builder. I'm going to introduce you to this data set a little bit. You see we have our animal ID different for each animal. Their time, we have two different measures of their brain signals of this dopamine signal in this case.
We have their X center and their Y center here. It's just their location in the space. Elongation, so whether they're scrunchy or whether they're stretched out, two measures of their activity, whether they're highly active or not moving at all, and then our three different exposure dates, which is, these are our three different time points.
We're going to go over to Graph Builder. We're going to pull this out here. Let's see. We're in Graph Builder. We're going to take our X location and our Y location, and we're going to use points to look at just where these animals are in the space as they're moving. Now, I'd like to see this with a different color for each animal, so I can see where everybody is in this space as they go.
We're going to move that over to the color block, and then we're going to separate out for three different exposure times, and we're just going to look at the first one. We see we have our different animals here, and they take up that XY space. They're filling it up. As we click on different animals, you can see how they fill that space with the different colors. Some of them are a little bit hard to see, but others, you can see, generally, they're taking up this space. They're moving through the X and Y pretty thoroughly, except for this one.
This animal, animal N091, really just stayed up in that upper right-hand corner. It wasn't taking up a lot of the space. It was just right there. It looks different. We wanted to know, are there different measures that also look funky with this mouse? Which takes us to our next graph. We're going to go up to Graph Builder again. This time, we're going to look at the elongation measure. We're going to do a different box plot for each animal.
Again, we're going to pull that animal ID to the color. We're also going to pull it down to our X-axis, so we get a different box plot for each of our different animals. What we're going to see is that our animal N091, which is highlighted right now, it's lower. It has a lower measure of elongation, so it's staying scrunchier compared to the other mice.
I'm going to remove those outliers because I'm not too worried about that. I want to look at the main data. I'm going to change a little bit of the styles, but you can see, compared to the other animals, this animal is a little funky.
Here, these are the two graphs that we've done before, and we looked at a third of looking at the percent of time that this animal spent not moving. You can see it's sedentary quite a lot. From these three graphs, we saw that this animal's tracking was misaligned. Something happened with the video, but when I shared this with my collaborators, they were able to fix it. We used Graph Builder to identify this problem in this big data set and get it fixed.
Now, our next example is a little less flashy. It's just confirming the patterns we expect to see are there. Dopamine signals, this is a neurotransmitter in the brain. As we watch it flash, it's known to decrease over time. With these graphs, we have our time on our X and the signal on the Y. You see? That dopamine signal is going down over time, just as we expect.
After this dopamine signal is collected, a correction algorithm is applied to it to correct for that decrease over time. We wanted to make sure that also was working. We see quick simple graph, colored for each animal. This looks good. With this quick easy check, we were able to confirm that this classifier and the signals were being collected and working as we thought. Good for QC, good for research.
Now, our last thing was to confirm a movement classifier. The movement could either be inactive or very active, but it shouldn't be both. These animals shouldn't be sprinting and sedentary at the same time. We're going to go over to Graph Builder again to check and make sure that no data is classified as both highly active and inactive.
We're going to drag those two columns over to our graph, and we're going to switch to points again because I like points after we change it for our animal ID. You can see the big blank spot in that upper right-hand corner. That would be where we'd see if the classifier messed up and said that something was highly active and not active at the same time. Sure enough, it's blank. Looks good. This also confirms that our classifier is doing its job and did not make this mistake.
Our goals have been met. We looked at a complicated fiber photometry data set collected over time, got that longitudinal data, and we used JMP Graph Builder to quickly and easily QC this data set and make sure that we have high-quality, reliable research. If you'd like to learn more about dopamine and norepinephrine in the brain and how that works with these particular mice, please scan the QR code, it'll take you to our paper. I hope you learned something. Thank you.
