Hi, everyone. My name is Carol Coe. I am a statistician at DLH. At DLH, we do statistical analysis of toxicological studies conducted by our clients. Today, I wanted to show you a dashboard we built to look at microtiter plate data. This work was done in collaboration with my DLH statisticians, Catherine Allen Moyer, Sean Harris, Juan Hwasier, and also, our collaborators from NHS, Helen Coney, and Keith Shockley.

Before I jump into the dashboard, I just wanted to give a quick overview of what the microtiter plate is. Microtiter plates are a standard tool in laboratory experiments. These are plates that contain walls arranged on a grid, and each wall can handle small volume of liquids. Scientists like to use these a lot because they can run several experiments at the same time.

The idea here is that each one of these wall is considered to be an experimental unit. You can put different test articles and different concentrations across these wells. These plates come in different sizes. This is a picture of a smaller one. In my example, I'm going to be using a 384 wall plate. It has 16 rows 24 columns. Once all of these walls are filled, these plates go into a plate reader where a device can measure responses from each one of these walls in the plate.

In my experience working with the scientists, they usually run these plates or run these experiments using several plates to account for the different factors that they're interested in. Examples of these factors might be the type of assay, cell culture, or the number of test articles that they want to experiment on. Usually when I get the data from these experiments, I get them all at the same time. Typically, I'm dealing with several 1000 rows from dozens of plates. Ideally, I usually like to be able to visualize each plate and check for any potential data issues before I do any statistical modeling.

It can be very difficult to see and understand what's happening when you have too much data. My team and I, we've thought about how can we make this, the visualization process easier, and a solution we came up with and that was also, thankfully, also easy to implement in JMP. Was a dashboard made up of, a local data folder and then three graph builder panels so that you can look at different views of the same dataset but from different perspectives. In this presentation, I'm actually not going to show you how to build a dashboard.

Instead, I'm just going to refer you to a link, a tutorial in here that's linked in here. And yeah. Now, I'm actually gonna switch over to JMP. That we can look at this dashboard in more detail. While I'm switching over, I just wanted to say that all the data that I'm using here have been simulated. All of the chemical names or initials that you see, they're all fictional, just for the purposes of this presentation.

The one key thing to do if you're working with plate data is to make sure that you understand the plate design. Because with these types of experiments they're prone to have edge effects or, like, location effects. Understanding where what chemical was placed in which wall actually can give you a lot more insight in your data. Because of that, I added a graph builder tab.

It's tucked away in here that shows you the plate design or the plate map. I'm going to maximize this so that we can see this in more detail. There we go. This is an example of a plate design. Some key things to look for are what type of controls do you have and how many do you have. In this particular example, I have vehicle controls, labeled here as media. They're located in the first two and last two columns of the plate. I also have a negative control on the first row and then two different types of positive controls in rows L and M.

The colors correspond to the different concentrations. We're going from darker to lighter green, and so that means we're going from highest concentrations to lower concentrations. Each one of these initials, you can think of them as a specific chemical or a test article. It looks like each row has a particular test article. The other thing to note with this particular example is that it looks like we have duplicate walls, and so you can see like pairing. For every two walls, the same concentration and test article was used.

If you want to go back to the dashboard, just click on this restore button, and then it'll bring you back to the dashboard. When you're just handed a batch of data and you don't really know much about it yet, one thing that I like to do is I just toggle back and forth between... This is assuming they're from the same assay and, two separate plates. I just toggle back and forth to see, did they actually use the same plate design? It would make sense that they do, but sometimes you never know. This tells me that, yes, the plate design is the same, but the data, of course, is changing because they're running at different times.

Then, also, you can toggle back and forth between your different assays. Here, I can see, K, the layout, there's some similarity in terms of where the vehicle controls are located, like, these yellow columns. But it looks like in this other assay, there is a negative control. Now it's in row b, and then there's no positive controls, and they also use a completely different, set of chemicals.

I'll switch over to this other tab where I have, another heat map, but this time when I'm plotting the colors correspond to the values of my responses. This how I've had them laid out it's basically similar to assay if you're looking at the plate. This can be very useful again because of those edge effects that tend to happen in these types of experiments. What they really like about the dashboard is that it's very easy to isolate problematic observations from looking at them at these different views.

For example, when you look at this heat map, the first thing that I see are, like, this column striping effect that's happening. What you could do is, you could select these wells. What I'm doing is I'm just holding my shift key while I'm selecting these. What you'll see is that in these other panels, they start to highlight too because of all the interconnectedness between these different graphs.

What is actually happening is that these darker wells, these are the points that are above. It looks like something went wrong in here. What I would do is I probably go back to the scientist or the person who did the actual experiment and ask, what happened in here? What can we do about... Do we need to exclude them? Do we now need to take the average of these two duplicate wells, in order to analyze our data further?

The scatter plot here on the right shows you a distribution of your data points. This can be useful if there is some natural chemical groupings in your plate. For this particular example, because I knew that it had, three different types of controls. It's just easy or useful for me to look at them, how are they distributed, when I group them according to chemical groupings. What stands out to me when I look at my vehicle controls is that I have a cluster of observations right here that seem to be lower than the others. Again, when I highlight them, immediately see these are the observations that are coming from the last column in this corner. Again, this is going to be another one of those.

When you talk to the scientists who did this experiment, you can ask them, did something happen in here, or maybe do we need to exclude these observations if they're just an artifact of something else. Another way that we've used this dashboard aside from showing our clients, okay, this is what your data looks like is also, like a way for them to be able to tag or flag observations that they want us to exclude in our analysis.

An example might be, so maybe they go through the dose response curves here on the bottom. There's this one particular data point in here that looks really suspicious because it's higher than the rest. Maybe they looked at it, and then they decided, maybe we should just exclude this observation because there's something weird or funky that happened in here that's not reflective of what they're expecting.

What we would ask them to do is to highlight this observation. Then, if you right-click on that particular point, go under rows, and just do a row label. It doesn't look like it did too much in here. It just added the name because I have name as a label in my data table. But what it does is that if you go back, if you look at your data table and this is that, row that we just added a label to. Now it has this yellow tag on it.

Imagine if you have a lot of these observations that you're tagging and maybe at the end of it, you want to have a listing of, these are the observations that you want to go back to and look at in more detail. Or maybe these are the observations that we need to exclude in the analysis later on. What you could do is you could create a column by using some real state formula. They're hidden right in here. I just did label row state and what this formula does is it turns these labels into a zero one indicator. Any row that has a label will have a value of one and then zero otherwise.

I guess one more cool thing that I like to point out because it's actually hard to find, but ever since I found this I keep using it. It was actually, in the legend panel so an example of this is in this heat map. If you right-click on the legend, color key and under gradient. It actually opens up a lot more options for customization. The two that I'm thankful that JMP has this is the scale type and the range type.

Scale type, the default is linear. You could see that the values of my color key goes up in equal spacing. You could change this to a quantile, cut-off. Actually, if you set that, it totally brings you a very different perspective of your data. I need you to be careful or be more thoughtful about what it is that you want to show, but that is an option that you can use. In this view, you could see that the last row, maybe looks a little bit weird too, the fact that it's all yellows. Anything that is on the edge of a plate I'm always worried about them. That's another, question that you can ask the scientists. Let me go back to it again.

It's all switch it back to linear, but there's actually a lot more options in here that you can play around with. The range type is what I like to use. The default option and JMP... I think what JMP does is it rounds up or down your min and max and so we end up with numbers that look nice. But if it's important to you to see the exact range of your data, you could switch that easily in here. Now you know that the maximum value you have is 115, and the minimum value is minus 84. I'll just switch it back to default.

Then lastly, there's some esthetic options in here. They're all pretty straightforward. I do like using discrete colors, especially when you have a continuous gradient. This is what looks like continuous, and I'll switch back to the discrete. For some reason, my brain just reads this a lot easier than when it's continuous. I like to use that a lot too.

Finally, I just wanted to say that even though this may seem like a really simple idea, we're just putting together like graph builder panels. It actually ended up to be a really effective and useful tool in revealing insights into the data that we would have missed otherwise. Make sure that you play around with dashboards. I hope you learned something useful in my presentation, and thanks for your time.