Plotting Distributions

In Cheminformatics, we often deal with distributions.  We may want to look at the distributions of biological activity in one or more datasets.  If we're building predictive models, we may want to look at the distributions of prediction errors.   A recent Twitter thread on "dynamite plots", and the fact that they should be banned from all scientific discourse got me thinking about a few different ways to look at distributions.  I thought it might be useful to put together some code examples showing how one can plot distributions.   As usual, all of the code I used for analysis and plotting can be found on GitHub.

Dynamite Plots (don't use these)
To demonstrate this concept, we'll look at the dataset published by Lee et al based on an earlier publication by Wang et al, that I used in my last post.  This dataset consists of molecules which had a range of activities against eight different targets.  One way to plot this data is to use the classic dynamite plot.   These plots, which were named for the similarity to the dynamite detonators depicted in Roadrunner cartoons, show the mean of a distribution as a bar plot and the standard deviation as a whisker above the plot.




We've all seen these plots or the even worse case where the authors simply display a bar plot displaying the mean with no error bars.  As you'll see if you do a Google search for "dynamite plot", many statisticians and data geeks consider these plots to be an abomination.  There is a lot not to like about dynamite plots.
  • They only show the mean and standard deviation
  • They don't tell you anything about how the data is distributed
  • They only show the upper whisker, which makes comparisons difficult 
Box Plots
One alternative to the dynamite plot is the box plot.  Box plots show a distribution as a box, which represents the middle 50% of the distribution, with whiskers that define the limits of the data.  The line in the middle of the box typically represents the median of the distribution.  The distance from the bottom to the top of the box is referred to as the interquartile range (IQR).  The box is extended with whiskers that define 1.5 * IQR above and below the top and bottom of the box.  Any points outside the range defined by the whiskers are considered outliers and typically drawn as points.   As you can see below, the box plot provides much more information on the overall activity distributions.


Beeswarm Plots
An alternative to the box plot is the beeswarm plot.  In the beeswarm plot, all of the data points are plotted in sets according to their y-values.  If multiple points have the same value, they are arrayed horizontally.   One advantage to the beeswarm plot is the ability to use the color to show an additional property.  For instance, we could use a beeswarm plot to display the pIC50 of the compounds, then use the color to display the fold error in the predicted pIC50.




Violin Plots
Box plots tell us a lot more about how the data is distributed, but we could still be missing key parts of the picture.  The box shows us where the middle 50% of the distribution and limits are but it doesn't tell us things like whether the data has a bimodal distribution.  One way to look at these distributions is to use a violin plot.   A violin plot is similar to a box plot, but instead of a box and whiskers, the data is plotted using a distribution known as a kernel density estimate (KDE).  A KDE creates a smoothed curve that provides an estimate of the probability density of a distribution.  One can think of a violin plot as a KDE that has been turned on its side and mirrored.  The violin plot also shows a box plot at its center.



Ridge Plots
From the plot above, we can see that the pIC50 distributions for most of the targets appear Gaussian, but the distribution for cdk2, mcl1, and tyk2 seem somewhat bimodal.  While violin plots are useful, many people, including me, find them visually overwhelming.   Most of us are more used to looking at distributions that are displayed horizontally rather than vertically.

One visualization that has been receiving a lot of attention, the ridge plot, attempts to address these limitations of the violin plot by arraying a set of overlapping KDEs in a vertical grid.  I find the plot below the easiest to process.   For me, this plot really highlights the differences in the distributions.




There is no single ideal plot for distributions.  A lot of this depends on the aspects of the data that you want to examine or highlight.   Hopefully, this post, and the corresponding code will provide some new ways for you to look at your data. 

Comments

Post a Comment

Popular posts from this blog

AI in Drug Discovery 2023 - A Highly Opinionated Literature Review (Part I)

Image
Here’s the first part of my review of some interesting machine learning (ML) papers I read in 2023.  As with the previous editions , this shouldn’t be considered a comprehensive review.  The papers covered here reflect my research interests and biases, and I’ve certainly overlooked areas that others consider vital.  This post is pretty long, so I've split it into three parts, with parts II and III to be posted in the next couple of weeks.    I. Docking, protein structure prediction, and benchmarking II. Large Language Models, active learning, federated learning, generative models, and explainable AI III. Review articles 2023 was a bit of a mixed bag for AI in drug discovery.  Several groups reported that the deep learning methods for protein-ligand docking weren’t quite what they were initially cracked up to be.  AlphaFold2 became pervasive, and people started to investigate, with mixed success, the utility of predicted protein structures.  There were reports of significant advanc
Read more

Generative Molecular Design Isn't As Easy As People Make It Look

Image
I was taken aback by a recent CNBC article entitled “ Generative AI will be designing new drugs all on its own in the near future ”.  I should know better than to pay attention to AI articles in the popular press, but I feel that even scientists working in drug discovery may have a skewed perception of what generative AI can and can’t do.  To understand exactly what’s involved, it might be instructive to walk through a typical generative molecular design workflow and point out a few things.  First, these programs are far from autonomous.  Even when presented with a well-defined problem, generative algorithms produce a tremendous amount of nonsense.  Second, domain expertise is essential when sifting through the molecules produced by a generative algorithm.  Without a significant medicinal chemistry background, one can’t make sense of the results.  Third, while a few nuggets exist in the generative modeling output, a lot of work and good old-fashioned cheminformatics are required to ext
Read more

AI in Drug Discovery - A Highly Opinionated Literature Review (Part II)

Image
Picking up where we left off in Part I , this post covers several other ML in drug discovery topics that interested me in 2023.  Some areas, like large language models, are new, and most of the work is at the proof-of-concept stage.  Others, like active learning, are more mature, and several groups are starting to explore nuances of the methods.   Here’s the structure of Part II.  4. Large Language Models 5. Active Learning 6. Federated Learning 7. Generative Models 8. Explainable AI 9. Other Stuff 4. Large Language Models The emergence of GPT-4 and ChatGPT brought considerable attention to large language models (LLMs) in 2023.  In November and December, several large pharmas held “AI Day” presentations featuring LLM applications for clinical trial data analysis. Many of these groups demonstrated the ability of LLMs to ingest large bodies of unstructured clinical data and subsequently generate tables and reports based on natural language queries.  Aside from some very brief demos on co
Read more