Interactive visualization of RADseq data

RADseq data has a reputation for containing a lot of missing data, particularly between more distantly related samples, which has led to concern over how different distributions of missing data might influence downstream analyses. However, there hasn’t yet been a lot exploration of how missing data are distributed in empirical (or simulated) RADseq data sets, and so this is something I’ve been working on quite a bit, and which was the subject of my talk at the Botany conference this year. Someone asked about the code I used for visualizing shared data between samples, and so I’m making it avaliable here.

The basic functions are to parse the assembled RADseq data set (.loci file), calculate the number of loci shared between samples, and then plot it. I focused on one of the simplest ways to visualize shared data which is with a heatmap, although this is not necessarily ideal, since it limits us to viewing only the amount of data shared by any two samples at a time.

Another reason for this post is to plug the new Python plotting library Toyplot which I’m really excited about. I won’t delve into all the details of why I think it’s great, you can check out their site for that, but the main thing Toyplot has going for it is that by default it creates plots in html, making it easy to view figures on the web, and making them readily scalable, interactive, and easy to embed. The default aesthetic is also really quite pleasing, which in comparison to most plotting libraries is a pretty big deal. The code is also very minimalist which makes it easy to learn (I’m looking at you ggplot!)

An example

Below is an example plot I created from a RADseq data set for 34 oak trees from my recent publication. If you hover over the plot the cursor shows the amount of data for any given data point, cool huh? Testing just now it seems the interactivity is a little slow on my website compared to when the plot is rendered in nbviewer (see link below). I wonder if this has to do with how my site is hosted. Anyway, the plot was actually pretty easy to make. My full code to download the data from online, calculate a shared-data matrix, and plot it with interactive popups is available here as a jupyter notebook.

Obviously this isn’t perfect yet. Next up I need to add a scale bar and I would like to add a topology to the left of the heatmap showing the relationships of the taxa. In our paper we created a similar plot by simply combining the tree (made using ape) and a heatmap together in Inkscape (Illustrator clone) since Python still doesn’t have a very decent tree plotting library. Word has it Toyplot is adding tree/graph plotting functionality, which if it’s good, would be very exciting. I’ll certainly be keeping my eye on it as it develops.