The visualization of temporal data by line graphs has been documented and popularized by William Playfair in the 18th century (Aigner et al, 2011; Beniger and Robyn, 1978). Today, time-dependent changes are still depicted by line graphs and ideally accompanied by a measure of uncertainty (Marx, 2013). Below, I provide a ‘walk-through’ for generating such a plot with R/ggplot2 to visualize data from time-series. For convenience, example data and an R-script that performs all steps is available here. The data that I used is from Mastop et al (2017). After the make-over with ggplot2, the graph looks like this:

Some time ago, my favorite (commercial) software package for making graphs was no longer supported due to a system upgrade. So I was looking for a powerful and flexible alternative for data visualization. I ended up with R and ggplot2. First of all, the elegant data visualization delivered by the ggplot2 package is hard to beat. On top of that, it is completely free and it has a large user base. The biggest obstacle (in my experience) was getting used to the tidy data format that is needed as input. The conversion of ordinary, spreadsheet type data into long, tidy data is dealt with in a previous blog, and again briefly explained below. Before we start, we need to load two packages (these need to be installed first):

>require(tidyr)
>require(ggplot2)

The example data (available here) is read from a file in csv format into a dataframe (you need to set the “working directory” to match the location of the file):

>df_wide <- read.csv("FRET-ratio-wide.csv", na.strings = "")

To find out wat the first six rows of the dataframe look like type:

>head(df_wide)

which returns:

  Time    cell.1    cell.2    cell.3    cell.4 ...   cell.12
1    0 0.9935818 0.9965762 0.9907055 0.9938881 ... 0.9859260
2    2 0.9972606 0.9979417 1.0014600 1.0041020 ... 1.0080610
3    4 1.0109350 1.0076620 1.0023690 1.0011580 ... 0.9993525
4    6 0.9959134 1.0018910 1.0034930 1.0011640 ... 0.9900663
5    8 0.9970548 0.9971041 0.9957852 1.0019420 ... 0.9979293
6   10 1.0038560 1.0018550 0.9897691 0.9976327 ... 1.0042090

The first column defines time and the other columns have data of individual cells corresponding to each of the timepoints. This data has a spreadsheet format, which is also named a ‘wide’ format. The first step is to convert it into a tidy data format, which is ‘long’. For details see my previous blog. The command to convert the df_wide dataframe into a tidy dataframe, df_tidy, is:

>df_tidy <- gather(df_wide, Cell, Ratio, -Time)

You can have a look at the first six rows of the dataframe to see what has changed:

>head(df_tidy)

Generating a basic graph

Once the data in the df_tidy dataframe is in long/tidy format, it can be used to generate a graph. The minimal instructions needed to display a graph are:

• The dataframe that is used as input: df_tidy
• The data used for the x-axis: Time
• The data used for the y-axis: Ratio

To plot the data with minimal instructions use the function qplot():

>qplot(data=df_tidy, x=Time, y=Ratio)

Points/dots are used by default, but for these data lines are more appropriate:

>qplot(data=df_tidy, x=Time, y=Ratio, geom='line')

In the resulting graph all the datapoints are connected by lines, and the curves are not grouped per cell. To imrove this use:

>qplot(data=df_tidy, x=Time, y=Ratio, geom='line', group=Cell)

Or goup the data per cell and give the data from each cell a different color:

>qplot(data=df_tidy, x=Time, y=Ratio, geom='line', color=Cell)

Generating a graph with layers

The ‘qplot’ command provides a quick way to plot the data. To have a bit more control we turn to ggplot2(). The syntax is slightly more complex, but it allows to plot multiple layers. To reproduce what was done with qplot we need:

>ggplot(df_tidy, aes(x=Time, y=Ratio)) + geom_line(aes(color=Cell))

The aes() function is used for mapping “aesthetics”.  The aesthetics specify how the variables from the dataframe are used to visualise those variables. In this case ‘Time’ is used for mapping the data onto the x-axis and ‘Ratio’ is used to map the data onto the y-axis. The geom_line() function specifies that the data is shown as a line. Inside geom_line(), aes(color=Cell) specifies that the line of each cell is mapped to a unique color. For more information on aesthetics and examples see this book chapter.

The geom_line() function defines that the data are displayed as lines. We can add another layer with a different geometric shape, geom_point(), to show the individual data as dots (in this example the dots will not have a color, since no aesthetic mapping is specified for geom_point):

>ggplot(df_tidy, aes(x=Time, y=Ratio))+ geom_line(aes(color=Cell))+ geom_point()

The thickness of the lines and the size of the dots (size) and transparency of the dots (alpha) can be adjusted, for example:

>ggplot(df_tidy, aes(x=Time, y=Ratio)) + geom_line(aes(color=Cell), size=0.8) +
  geom_point(size = 1, alpha=0.3)

The appearance of the plot depends on the sequence in which the layers are added. This is defined by the sequence in which the objects are added, i.e. the first object defines the first layer and next object is added on top. For instance, this code will generate a plot in which the lines are on top of the points:

> ggplot(df_tidy, aes(x=Time, y=Ratio)) + geom_point(alpha=0.3, size=0.5) +
   geom_line(aes(color=Cell), size=0.8)

To generate a plot in which the points are visible, we define them as the last object and hence, they are added on top of the lines:

>ggplot(df_tidy, aes(x=Time, y=Ratio)) + geom_line(aes(color=Cell), size=0.8) +
   geom_point(alpha=0.3, size=0.5)

Generating a graph with data summaries

The option of adding layers allows the addition of a summary of the data, for instance the average, or data for various types of error bars. To achieve this, a dataframe is defined to store the summary statistics:

>df_summary <- data.frame(Time=df_wide$Time, n=tapply(df_tidy$Ratio, df_tidy$Time, length), mean=tapply(df_tidy$Ratio, df_tidy$Time, mean))

This will generate a df_summary dataframe that has the average signal and the number of samples per time point.

To add the standard deviation and standard error of the mean:

> df_summary$sd <- tapply(df_tidy$Ratio, df_tidy$Time, sd)

> df_summary$sem <- df_summary$sd/sqrt(df_summary$n-1)

> head(df_summary)

Returns:

   Time  n      mean          sd          sem
0     0 12 0.9929529 0.003313210 0.0009989706
2     2 12 0.9989194 0.007343436 0.0022141294
4     4 12 1.0027538 0.004194419 0.0012646648
6     6 12 1.0002915 0.006014562 0.0018134585
8     8 12 0.9994599 0.002606858 0.0007859971
10   10 12 1.0007186 0.004189333 0.0012631315

Finally, we add the lower and upper bound of the 95% confidence interval:

>df_summary$CI_lower <- df_summary$mean + qt((1-0.95)/2, df=df_summary$n-1)*df_summary$sem
>df_summary$CI_upper <- df_summary$mean - qt((1-0.95)/2, df=df_summary$n-1)*df_summary$sem

We can start by plotting the average response:

>ggplot(df_summary, aes(x=Time, y=mean)) +
   geom_line(data=df_summary, aes(x=Time, y=mean), size=1, alpha=0.8)

We add the 95% confidence interval (95%CI) as a measure of uncertainty. Here we employ geom_ribbon() to draw a band that captures the 95%CI. To this end, we employ aes() inside geom_ribbon() to specify that the upper and lower limits of the confidence interval from df_summary define the borders of the ribbon. (Note that alternative methods that display uncertainty may be considered.)

>ggplot(df_summary, aes(x=Time, y=mean)) +
   geom_line(size=1, alpha=0.8) +
   geom_ribbon(aes(ymin=CI_lower, ymax=CI_upper) ,fill="blue", alpha=0.2)

The resulting figure only shows the data summary. Since it can be valuable to show the underlying data we will plot the data from individual cells as well. These data are present in another dataframe, yet it is possible to add these data to another layer. To demonstrate this, the original data from the individual measurements (from df_tidy) is added:

>ggplot(df_summary, aes(x=Time, y=mean)) +
   geom_line(size=1, alpha=0.8) +
   geom_ribbon(aes(ymin=CI_lower, ymax=CI_upper), fill="blue", alpha=0.2) +
   geom_line(data=df_tidy, aes(x=Time, y=Ratio, group=Cell))

Improving the presentation and annotation of the graph

The graph shown above displays all the data of interest, but it looks cluttered. In the final session the lay-out and annotation will be changed to improve the visualization and communication of the results (see also ‘Graphics for Communication’).

To make the individual data less pronounced, the lines are plotted in grey. Also, the graph looks better if the ribbon of the 95%CI is drawn on top, so we move the grey lines to the first layer and plot the average and 95%CI on top of that:

>ggplot(df_summary, aes(x=Time, y=mean)) +
   geom_line(data=df_tidy, aes(x=Time, y=Ratio, group=Cell), color="grey") +
   geom_line(size=1, alpha=0.8) +
   geom_ribbon(aes(ymin=CI_lower, ymax=CI_upper) ,fill="blue", alpha=0.2)

To indicate where we performed a manipulation of the system another layer can be added. In this experiment a stimulus was added at t=44 and inhibited at t=146. To reflect this, we add a grey box with the function annotate():

>ggplot(df_summary, aes(x=Time, y=mean)) +
   geom_line(data=df_tidy, aes(x=Time, y=Ratio, group=Cell), color="grey") +
   geom_line(size=1, alpha=0.8) +
   geom_ribbon(aes(ymin=CI_lower, ymax=CI_upper) ,fill="blue", alpha=0.2) +
   annotate("rect",xmin=44,xmax=146,ymin=-Inf,ymax=Inf, alpha=0.1, fill="black")

To change the default grey background from the area of the graph, the ‘theme’ can be changed:

>ggplot(df_summary, aes(x=Time, y=mean)) +
   geom_line(data=df_tidy, aes(x=Time, y=Ratio, group=Cell), color="grey") +
   geom_line(size=1, alpha=0.8) +
   geom_ribbon(aes(ymin=CI_lower, ymax=CI_upper) ,fill="blue", alpha=0.2) +
   annotate("rect",xmin=44,xmax=146,ymin=-Inf,ymax=Inf,alpha=0.1,fill="black")+
   theme_light(base_size = 16)

Finally, we change the labels of the x- and y-axis and add a title:

>ggplot(df_summary, aes(x=Time, y=mean)) +
   geom_line(data=df_tidy, aes(x=Time, y=Ratio, group=Cell), color="grey") +
   geom_line(size=1, alpha=0.8) +
   geom_ribbon(aes(ymin=CI_lower, ymax=CI_upper) ,fill="blue", alpha=0.2) +
   annotate("rect",xmin=44,xmax=146,ymin=-Inf,ymax=Inf,alpha=0.1,fill="black") +
   theme_light(base_size = 16) +
   ylab("Ratio YFP/CFP [-]") + xlab("Time [s]") +
   ggtitle("G-protein activation by histamine")

Final words

The end results is a nice and clean visualization of the data, which is an improvement over the original visualization (made with commercial software, see insets of figure 6 from Mastop et al (2017)). The instruction to plot graphs with ggplot() usually consists of several different functions and may be daunting at first sight. I hope that providing this ‘walk-through’ that shows how to build a graph layer-by-layer lowers the barrier to start using R/ggplot2 for visualization of (temporal) data.

Acknowledgments: Thanks to Franka van der Linden, Eike Mahlandt and Jenny Olins for testing and debugging the code.

(14 votes)

Loading...

Applications are invited for a four year PhD studentship funded by The Royal Society to conduct research under the supervision of Dr Amanda Sferruzzi-Perri at the Department of Physiology, Development and Neuroscience.

Title: Role of placental endocrine malfunction in the programming of disease in offspring

Closing date: 17^th June 2018

For more details about the project and application process please see the following webpage:

http://www.jobs.cam.ac.uk/job/17502/

(No Ratings Yet)

Loading...

APPLY NOW

(No Ratings Yet)

Loading...

The Laboratory of Early Mammalian Developmental Biology (LEMDB), University of South Bohemia, České Budějovice, Czech Republic (Alexander W. Bruce group).

The LEMDB is currently recruiting both Masters (Mgr) & Ph.D. level studentships to research projects related to cell-fate choice.

Research projects offered in relation to p38 mitogen-activated-kinase’s role in primitive endoderm/ PrE development, the regulation of spatial allocation of cells relevant to first differential cell-fates i.e. outer trophectoderm/TE & inner-cell-mass/ ICM and others…

University provided stipend guaranteed for 4-years (plus grant based salaries and performance related bonuses)

Please send your CV and covering letters to lab head Alexander W. Bruce at: awbruce@prf.jcu.cz

FaceBook: LEMDB facebook

(No Ratings Yet)

Loading...

In September, Development is hosting the third of its highly successful series of meetings focusing on human developmental biology. Held in the Wotton House estate near Dorking in Surrey and organised by Paola Arlotta, Ali Brivanlou, Olivier Pourquié and Jason Spence, the meeting will cover the latest developments and future prospects for this rapidly evolving field.

The early bird deadline is Thursday, 31st of May!

Find out more and apply here:

biologists.com/meetings/from-stem-cells-to-human-development-september-2018

(1 votes)

Loading...

A postdoctoral position is available starting in October 2018 at the Developmental Biology Institute of Marseille, France, located on the Luminy campus, in the group of Vincent Bertrand. The position is funded for 3 years.

The acquisition and maintenance of the differentiated state of a neuron are achieved by terminal transcription factors that are expressed throughout the life of the neurons. However, gene expression is often noisy, presenting a high stochastic variability. The objective of this project is to determine how the identity of a neuron can be established and maintained throughout the life of the animal in a reliable manner despite this variability. This question will be addressed using C. elegans, an excellent system to analyze the robustness of neuronal development. The project will combine recent genome engineering methods (CRISPR), quantitative in vivo imaging and single molecule FISH.

We are seeking a highly motivated candidate with a PhD in molecular biology, genetics, cell biology, developmental biology or neurobiology. The working language in the laboratory is English.

A letter of motivation, a CV and the names of two referees should be sent to Vincent Bertrand (vincent.bertrand@univ-amu.fr).

The announcement on IBDM website here.

(No Ratings Yet)

Loading...

A postdoctoral position is available in the laboratory of Dr. Sophie Astrof at Thomas Jefferson University to study roles of cell-extracellular matrix (ECM) interactions in cardiovascular development and congenital heart disease. We have recently discovered that progenitors within the second heart field (SHF) give rise to endothelial cells composing pharyngeal arch arteries. Projects in the lab focus on the role of ECM in regulating the development of SHF-derived progenitors into endothelial cells and their morphogenesis into blood vessels. The successful candidate will combine genetic manipulation, embryology, cell biology, and confocal imaging to study molecular mechanisms of micro-environmental sensing during vascular development.
Astrof laboratory is a part of a modern and well-equipped Center for Translational Medicine at Jefferson Medical College (http://www.jefferson.edu/university/research/researcher/researcher-faculty/astrof-laboratory.html) located in the heart of Philadelphia.
To apply, please send a letter of interest detailing your expertise, CV and names and contact information of three references to sophie.astrof@gmail.com

(No Ratings Yet)

Loading...

Short CLE peptides regulate a wide variety of processes during plant development. In the developing root, the receptors and co-receptors for CLEs have remained largely unclear, as have the relationships between different CLEs and different (co-)receptors. A new paper in Development addresses this problem by reporting a new receptor kinase involved in CLE signaling. We caught up with lead author Pauline Anne and her PI Christian Hardtke, Professor in the Department of Plant Molecular Biology at University of Lausanne, to find out more.

Christian, can you give us your scientific biography and the questions your lab is trying to answer?

Out of an early interest in genetics, I studied Biology at the University of Munich. I became particularly fascinated by Developmental Biology when I learned about the then still recent isolation of the Bicoid morphogen. At that point I was determined to work with Drosophila. Incidentally, Thomas Berleth, who had cloned Bicoid a few years earlier, worked with Gerd Juergens in the Genetics department at that time. They were recruiting an M.Sc. student to investigate the “Arabidopsis embryo patterning mutant monopteros”. I only realised during the interview that Arabidopsis was not some short germ-band insect, but an upcoming plant model organism. Maybe my judgement was clouded by the dense pipe smoke in Gerd’s office, but I committed to the project, which also became my Ph.D. In hindsight, it was a stroke of luck, I enjoy working with plants ever since.
Four years later I had cloned MONOPTEROS and walked enough along chromosomes. Thus, after briefly wrapping up my Ph.D. work at the University of Toronto with Thomas, I started my postdoc as an HFSP fellow in the lab of Xing-Wang Deng at Yale University, to learn biochemistry. It was an exciting time as the lab discovered the mechanistic basis of the COP9-COP1 signalosome-ubiquitin ligase system in Arabidopsis photomorphogenesis, and its operational conservation but differential output in distant eukaryotes. When I started my own lab at McGill University however, I turned to root system architecture with the aim to analyse its intraspecific natural genetic variation. Since then, we isolated several QTL alleles. Because I like to explore the molecular foundations of their developmental phenotypes, my lab’s focus has shifted in the years after I moved to the University of Lausanne. Nowadays we are mainly interested in deciphering molecular pathways that govern the patterned differentiation of plant vascular tissues, and in identifying the non-cell autonomous signals that are emitted from phloem vasculature to organize surrounding tissues.

And Pauline, how did you come to work in Christian’s lab and be involved with this project?

Christian and I share common interests. In my PhD, I characterised the phloem regulator OCTOPUS (OPS) and its interaction with the brassinosteroid phytohormone pathway. I wanted to continue working on phloem differentiation, and Christian is a leader in that field. We discussed possibilities, I wanted to continue exploring signalling pathways, so he proposed to analyse CLE26 sensing, which we suspected to play a role in phloem differentiation.

When did your lab first get interested in CLE peptide signalling in the root?

CH When we got stuck in our analysis of BREVIS RADIX (BRX), another positive regulator of phloem formation. Stephen Depuydt, a postdoc in my lab at the time, proposed to screen for suppressors of the phloem defects in brx mutants. Although I was rather sceptical about the prospect of healing the strong brx phenotype by second site mutation.
PA But it worked! Loss-of-function in BARELY ANY MERISTEM 3 (BAM3), a receptor kinase, was isolated as a perfect brx suppressor from this screen. They later found that BAM3 is the receptor for the phloem-specific CLE45 peptide, pointing to a role of autocrine CLE peptide signalling in phloem differentiation.

Can you give us the key results of the paper in a paragraph?

PA We isolated CLERK as a genetic component of CLE26 sensing in the root, but it is also required for perception of other CLE peptides. Moreover, CLERK encodes a co-receptor protein that acts in the phloem and is necessary to prevent premature protophloem specification.

CH Also, CLERK acts genetically in parallel to described CLE sensing pathways in the protophloem, corroborating the redundancy we suspected at the outset.

You have lots of CLE peptides, lots of receptors, lots of co-receptors, and lots of promiscuous receptor-ligand and receptor-co-receptor interactions. How does this system achieve any degree of specificity?

PA Although many exogenously applied CLE peptides suppress root protophloem differentiation, only few, like CLE45 or CLE26, are actually expressed in root development. Recent studies by colleagues suggest that some CLEs are induced by specific environmental stimuli, to shut down root growth in unfavourable conditions. So part of the answer probably lies in specific ligand expression.

CH …and in specific receptor and co-receptor expression. It is also noteworthy that some receptors are not only more promiscuous with respect to ligands than others, but that ligand binding affinities can vary by an order of magnitude. So even in presence of multiple receptor-coreceptor pairs, ligand interaction can likely trigger vastly different signal outputs. It will be a major challenge to map these interactions in time and space, also because cross-regulation between CLE-sensing receptors has been observed.

How might CLERK help you in future analysis of the CLE sensing pathway?

CH One surprising discovery was that in brx or ops mutants, CLERK is not required for CLE26 or CLE45 sensing. We can therefore perform a sensitized screen for CLE resistance, for example in a brx clerk double mutant background. This should enable us to home in on the CLE26 receptor, as well as receptors for other CLEs.

PA The system is more and more complex, and genetics does not always give the expected answers. But it looks like it is less complex in the root than in the shoot. For example, while our paper was under review, colleagues found that CLERK also plays a role in the shoot. However, there its action is masked by redundancy between four related co-receptors.

When doing the research, did you have any particular result or eureka moment that has stuck with you?

PA Yes, the day I received the whole genome sequencing results of our original clerk mutant! I was on vacation in Italy. As you guess, I could not wait to be back in the lab to analyse the results…and At2g23950 popped up, a receptor-like kinase. This was the best candidate I could expect at that moment.

And what about the flipside: any moments of frustration or despair?

PA Of course! Even if the project went well, there are more instances of frustration in research life than eureka moments. I think my main frustration was the very weak expression level of CLERK in the root tip. But on the upside, this forced me to hone my confocal microscopy skills…

What next for you after this paper?

PA In the short term, I will focus on the characterization of other candidates isolated from the screen. I am most interested in a downstream CLE signalling component that I isolated recently, a prey that the lab has long been after. In the long term, I hope to turn my projects into more publications, to be competitive in the academic job market.

And where will this work take the Hardtke lab?

CH It will force us deeper into investigating the mechanics of CLE perception. One unexpected result contributed by our collaborators, Michael Hothorn’s lab, was the observation that the CLERK ectodomain cannot interact with the ectodomains of a few known bona fide CLE receptors, in the presence of ligand. If CLERK is indeed a generic CLE co-receptor, which is a parsimonious assumption, then this indicates that something is missing to make the system work biochemically in isolation. Maybe the sensitized screen will deliver this piece of the puzzle.

Finally, let’s move outside the lab – what do you like to do in your spare time?

CH Depends on the time of year. In the winter, I take every opportunity to go skiing. Luckily, from Lausanne it only takes about one hour to reach the bottom of some of the most spectacular ski slopes on the planet. In the summer, I like to attend a random selection of sports events. In general, I like to watch people perform feats that are impossible for myself, like swimming 100 m in less than a minute…

PA …I am a mom of two young kids, 3 years and one and a half years old. I spend all my free time with them and I really enjoy it. I love to read books with them. But sometimes I also have to fold up to fit in a cardboard hut, to escape wolves or other monsters. Sometimes it is more tiring to be a mom than to be in the lab. But my favourite moments are with the kids outside, observing nature, playing with its treasures, showing them birds, insects and, of course, plants!

(1 votes)

Loading...

We are seeking to recruit a research technician to work in the research group of Dr Vikas Trivedi at the newly established European Molecular Biology Laboratory (EMBL) Barcelona. We are situated in the Barcelona Biomedical Research Park (PRBB), one of the largest infrastructures in Southern Europe dedicated to translational research and which is located on the Mediterranean coast. Work in the group focusses on studying self-organization in multi-cellular systems, through comparative study of embryos and embryonic organoids, using both zebrafish and mouse models.

The position is available for a period of 3 years, in the first instance and can be renewed, depending on circumstances at the time of the review.

The successful candidate will be responsible for providing technical support for all projects running in the laboratory and general lab management.

Tasks

• Cell culture and cell line maintenance of mammalian (primarily mouse) embryonic cells
• Maintenance of zebrafish colonies, setting up crosses
• Zebrafish embryology techniques (cell transplantation, dissection, zygotic injections etc.)
• Basic molecular biology (PCRs, in vitro transcription, cloning, Western Blots, in situ hybridization and immunohistochemistry)
• Overseeing day to day operations, general organization of the laboratory such as administration of several databases (plasmids, antibodies, cell and fish lines, etc)
• Maintaining supplies (preparing and ordering reagents), servicing equipment, developing new protocols and training new laboratory members
• Setting up routine imaging and cell sorting experiments and sample preparation for the same.
• In addition to the technical support, the successful candidate will be encouraged to carry out their own research project within available capacity

Qualifications and Experience

The successful candidate must have:

• a Bachelors or Master’s degree in biology or related life science field
• one or more years of experience with cell/tissue culture techniques and cell sorting
• background in basic molecular biology techniques
• experience with zebrafish research is ideal but not necessary
• basic understanding of imaging (e.g widefield microscopes) is desirable but not necessary. Any prior experience with standard packages such as ImageJ /FIJI would be an advantage
• strong interpersonal and verbal communication skills to interact with internal/external contacts at all levels in a friendly and professional manner. You will be a team worker in an international and collaborative environment.
• openness to learn new techniques and work independently
• well organized and reliable
• fluency in English; knowledge of Spanish will be a plus

The successful candidate will be expected to take over routine imaging and analyses over time. However, any skills needed in the process will be taught and the candidate is expected to have proactiveness and enthusiasm toward new techniques.

Application Instructions

Application deadline: 13 June 2018
Please apply online through the EMBL recruitment system

Further details about the application can be found here

Additional Information

The remuneration package comprises from a competitive salary, a comprehensive pension scheme, medical, educational and other social benefits, as well as financial support for relocation and installation, including your family. Further details can be found here

(No Ratings Yet)

Loading...

Visualizations for comparing datasets is a topic in all my data viz classes. Current solutions for comparing 2,3, 4 and more datasets are diverse and some are controversial. A one-fits-all solution does not exist, but there are well-working solutions, and some that should be avoided.

1-3 datasets

Comparing two or three datasets works well in Venn diagrams. Most people learn them in school, and if not they are intuitive*. Each dataset is shown as a circle and they are arranged such that all overlaps are shown. Done.

More datasets

Things get problematic when comparing more than three datasets. Mathematically, it is not possible to show all overlaps of four or more datasets with circles. One possibility is to leave out some overlaps, as is often done in Euler diagrams. In the example below the overlap between “oocyte stage 2-7” and “oocyte stage 9” is for instance not visualized (RNAs localized in the oocytes across development, see publication). I find it however confusing when data is left out and sometimes “no overlap” is an important information itself.

Venn himself devised the diagram comparing four and more datasets by switching from a circles to ellipses. Branko Grünbaum developed the ellipsoid representation further for comparison of five dataset by . Their strategies are used by the online tool Draw Venn (Yves Vandepeer, Univ of Gent) where you can make Venn plots by simply uploading your data there. A variation is used by Heberle et al here (publication). There is also an R package by Victor Quesada.

I find there are two problems for Venn diagrams with more than three datasets. First, it takes long to read them and extract all information: comparing four datasets gives a diagram with 15 regions/11 overlaps, five datasets gives a diagram with 31 regions/26 overlaps! I invariably end up writing the numbers down into my own table. Secondly, the areas can’t possibly be representative for the overlap size – and this is a lost information.

New: upset plots

An alternate solution, the upset plot, was developed by Niels Gehlenborg and Jake Conway. Presence of dataset elements in a given intersect is shown with a dot in a simple table. The size of the intersect is represented with a bar chart. Both are simple visuals that are easy to consume. Their package is available in R and simple to use.

Customising upset plots

While the upset plots are simple, I think they can be improved. In upset plots the intersect is shown above the actual datasets, that serve as the legend. Basically, one is forced to read the upset from the bottom up. By flipping the plot horizontally this caveat is overcome: now the datasets are on the left, where we typically read first, and the bar is shown on the right nicely accompanying the respective set. Another improvement is to clearly label the intersects e.g. “present in one set”, “two sets” and to group them visually. Additionally, I have also color-coded the datasets to provide a quicker way of orienting the reader.

Depending on your message, you will have to find the optimal ordering strategy. I visualized the subcellular enrichments of RNAs and how they change localization during the development of the fruit fly oocyte. I would want to learn e.g. what happens to the hundreds of specific RNAs that enrich at early stages? Do they remain localized at all stages? It turns out the majority gives up their specific subcellular enrichment and instead become distributed inside the cell while other RNAs (not visualized here) take their place (more information on the biology).

Note, I did all the fine-tuning of the upset plot with illustrator but most likely it is also possible in R directly.

* Be aware that more people than you expect do not know Venn diagrams & require an introduction!

(16 votes)

Loading...