Biovis/Visweek recap

Finally time to write something about the biovis/visweek conference I attended about a week ago in Providence (RI)... And I must say: they'll see me again next year. (Hopefully @infosthetics will be able to join me then). Meanwhile, several blog posts are popping up discussing it (see here and here, for example).

This was the first time that biovis (aka the IEEE Symposium on Biological Data Visualization) was organized. It's similar to the 2-year old vizbi, but has an agenda that is more focused on research in visualization rather than application of visualization in the biological sciences. Really interesting talks, posters and people.

The biovis contest
This first installment of biovis included a data visualization contest, focusing on "specific biological problem domains, and based on realistic domain data and domain questions". The topic this year was on eQTL (expression quantitative trait loci) data, and I'm really happy that Ryo Sakai -  now a PhD student in my lab - won the "biologists' favourite" award!! The biologist jury was impressed with the ease in which his visualizations of the eQTL data highlighted and confirmed prior knowledge, and how it suggested directions for further experiments. It was interesting to see that there was a huge variation in the submissions, going from just showing the raw data in an interesting way (which Sakai-san did) to advanced statistical and algorithmic munging of the input data and visualizing the end result (which the winner of the "dataviz professionals' favourite" award did). See how this relates to my previous post on humanizing bioinformatics?


Interesting talks - amazing (good & bad) talks
As this was the first time that I attended visweek, I was really looking forward to the high quality presentations/papers and posters. Overall, I got what I wanted. But there were some examples of papers and posters that I have major doubts about (taking into account that I have to be humble here in talking about people working in the field for far longer than I do).
One example that seemed pretty counterintuitive was a presentation by Basak Alper from Microsoft about a new set visualization technique that they baptized LineSets. The main issue that they want to solve is the visualization of intersections of >3 sets (up to 3 you'd just use Venn diagrams). Their approach is to connect the different elements from a set by a line; hence: linesets. However, I (and many others with me) felt that this approach has some very serious drawbacks. Most of all, it suggests that there is an implicit ordering of the elements, which there isn't. In the image below, for example, line sets were used to connect Italian restaurants (in orange) and Mexican restaurants (in purple). That's the only thing this visualization wants to do: tell me which of the restaurants are Italian and which are Mexican. But give this picture to 10 people, and every single one of them will think that the lines are actually paths or routes between these restaurants. Which they're not... The example below shows data that has specific positions on a map, but they demonstrate this approach on social networks as well.

LineSets

Another example comes from the biovis conference: TIALA or Time Series Alignment Analysis. Suppose you have the time-dependent gene expression values for a single gene, which you'd plot using a line plot. Now what would you do if you have that type of data for 100 genes? Would you put those plots into 3D? I know I wouldn't... And better still: would you then connect these plots so you end up with some sort of 3D-landscape? That's like connecting the tops of a barchart displaying categorical data with a line...

TIALA - Time Series Alignment Analysis

But of course there were plenty really good talks as well. Some of the talks I really enjoyed are those about HiTSEE (by Bertini et al) on the analysis of high-throughput screening experiments, EVEVis (Miller et al) on multi-scale visualization for dense evolutionary data, arc length-based aspect ratio selection (Talbot et al) which is an alternative to banking to 45 degrees, drawing road networks with focus regions (Haunert et al), and especially DICON which showed an amazing application of visual analysis of multidimensional clusters using healthcare data.

HiTSEE

EVEVis

Road networks with focus regions

DICON - interactive visual analysis of multidimensional clusters

Meeting interesting people
But of course this was very much about meeting interesting people as well. It was really nice to exchange ideas again with the biovis crowd (Nils Gehlenborg, Cydney, Tamara, Will Ray, ...), and I finally had the chance to have a chat with @filwd Enrico. All those discussions with Thorri from Icelandic DataMarket were both useful and fun (as was our day hanging out in town, chatting to the Occupy Providence woman (forgot her name, I'm afraid) and trying to find a good hat).
At the airport on my way back, as I was trying to find out how to get to Brussels (as our flights were cancelled due to the weather), a chap comes to me and introduces himself as someone from Belgium. From Leuven. From our very own faculty. So together with @infosthetics Andrew that now makes three of us :-)

Anyway: I'll definitely be back next year (have to play some more official role anyway) and already looking forward to it.

Humanizing Bioinformatics

I was invited last week to give a talk at this year's meeting of the Graduate School Structure and Function of Biological Macromolecules, Bioinformatics and Modeling (SFMBBM). It ended up being a day with great talks, by some bright PhD students and postdocs. There were 2 keynotes (one by Prof Bert Poolman from Groningen (NL) and one by myself), and a panel discussion on what the future holds for people nearing the end of their PhDs.

My talk was titled "Humanizing Bioinformatics" and received quite well (at least some people still laughed at my jokes (if you can call them that); even at the end). I put the slides up on slideshare, but I thought I'd explain things here as well, because those slides will probably not convey the complete story.

Let's ruin the plot by mentioning it here: we need data visualization to counteract the alienation that's happening between bioinformaticians and bright data miners on the one hand, and the user/clinician/biologist on the other. We need to make bioinformatics human again.

Jim Gray from Microsoft wrote a very interesting book "The Fourth Paradigm - Data-intensive Scientific Discovery". Get it. Read it. He describes how the practice of research has changed over the centuries. In the First Paradigm, science was very much about describing things; the Second Paradigm (last couple of centuries) saw a more theoretical approach, with people like Keppler and Newton defining "laws" that described the universe around them. The last few decades saw the advent of computation in the research field, which allowed us to take a closer look at reality by simulating it (the Third Paradigm). But just recently - so Jim Gray says - we're moving into yet another fundamental way of doing science. We have moved into an age where there is just so much data generated that we don't know what to do with it. This Fourth Paradigm is that of data exploration. As I see it (but that's just one way of looking at it, and it doesn't want to say anything about what's "better" than what), this might be a definition for the difference between computational biology and bioinformatics: computational biology fits within the Third Paradigm, while bioinformatics fits in the Fourth.

Being able to automatically generate these huge amounts of data (e.g. in genome sequencing) does mean that biologists have to work with ever bigger datasets, using ever more advanced algorithms that use ever more complicated data structures. This is not about just some summary statistics anymore; it's support vector machine recursive feature elimination, manifold learning and adaptive cascade sharing trees and stuff. Result: biologist is at a loss. Remember Dr McCoy in Star Trek saying "Dammit Jim, I'm a doctor, not an electrician/cook/nuclear physicist" whenever the captain let him do stuff that is - well - not doctorly? (Great analogy found by Christophe Lambert). It's exactly the same for a clinician nowadays. In order to do a (his job: e.g. decide on a treatment plan for a cancer patient), he has to first do b (set up hardware that can handle the 100s of Gb of data) and c (devise some nifty data mining trickery to get his results). Neither of which he has the time or training for. "Dammit Jim, I'm a doctor, not a bioinformatician". Result: we're alienating the user. Data mining has become so complicated and advanced, that the clinician is at a complete loss. Heck, I'm working at a bioinformatics department and don't understand half of what they're talking about. So what can the clinician do? His only option is to trust some bioinformatician to come up with some results. But this is a blind trust: he has no way of assessing the results he gets back. This trust is even more blind than the one you give the guy who repairs your car.

As I see it, there are (at least) four issues.

What's the question?
Data generation used to be really geared towards proving or disproving a specific hypothesis. The researcher would have a question, formulate some hypothesis around it, and then generate data. Although that same data could already be used to answer other unanticipated questions as well, this really became an issue with easy, automated data generation; DNA sequencing being a prime example. You might ask yourself "does this or that gene have a mutation that lead to this disease?", but the data you generate (i.c. exome sequences) to answer this question can be used to answer hundreds of other questions as well. You just don't know what questions yet...
Statistical analysis and data mining are indispensable for (dis)proving hypothesis, but what if we don't know the hypothesis? As many others in the field, I believe that data visualization can give us some clues at what to investigate further.

Let's for example look at this example hive plot by Martin Krzewinski (for what B means: see the explanation at the hive plot website). Suppose you're given a list of genes in E.coli (or a list of functions in the linux operating system) and the network between those genes (or functions). Using clever visualization, we can define some interesting questions that we can look into using statistics or data mining. For example: why do we see so many workhorse genes in E.coli? Does this reflect reality, and what would that mean? Or does it mean that our input network is biased? What is so special about that very small number of workhorse functions in linux that have that high connectivity? These are questions that we need to be presented to us.

What parameters should I use?
Second issue: the outcome from most data mining/filtering algorithms depend tremendously on the right parameters. But it can be very difficult to actually find out what those parameters should be. Does there actually exist a "right" set of parameters for this or that algorithm? Also, tweaking some arguments just a little bit can have vast effects on the results, while you can change other parameters as much as you want, but it won't affect the outcome whatsoever.

Turnbull et al. Nature Genetics 2010

Can I trust this output?
Issue number 3: if I am a clinician/biologist and a bioinformatician hands me some results, how do I know if I can trust those results? Heck, being a bioinformatician myself and writing some program to filter putative SNPs, how do I know that my results are correct? Suppose there are 3 filters that I can apply consecutively, with different combinations of settings.

Looking at exome data, we main information that we can use for assessing the results of SNP filtering are the fact that you should end up with 20k-25k SNPs, and a transition/transversion ratio of 2.1 (if I remember correctly). But there's many different combinations of filters that can give these summary statistics. The state of the art (believe it or not) is to just run many different algorithms and filters independently, and then take the intersection of the results...

I can't wrap my head around this...
And finally, there's the issue of too much information. Not just the sheer amount, but of different data sources. It's actually not really too much information per se, but too much to keep into one head. Someone trying to decide on a treatment plan for a cancer patient, for example, will have to combine data from heterogeneous datasets, multiple abstraction levels and multiple sources. He'll have to look into patient and clinical data, family/population data, MR/CT/Xray scans, tissue samples, gene expression data and pathways. That's just too much. His cognitive capacities are fully engaged in trying to integrate all that information, rather than in answering the initial question.

Visualization... part of the solution
I'm not saying anything new here when I suggest that data visualization might be part of the solution to these problems. As current technologies and analysis methods have alienated the end-user from his own results, visualization can reach over and cross this gap. The rest of the presentation is basically about some basic principles in data visualization, which I'll not go further into here.

All in all, I think the presentation went quite well:

Visualize This (by Nathan Yau) arrived...

Last Friday I received my long-anticipated copy of "Visualize This" by Nathan Yau. On its website it is described as a "practical guide on visualization and how to approach real-world data". You can guess what my weekend looked like :-)

Overall, I believe this book is a very good choice for people interested in getting started in data visualization. Not only does it provide the context in which to create visualizations (chapters 1, 2 and 9), it also handles different tools for creating them: R, protovis, flash.... Apart from chapter 3 that is dedicated entirely to that topic, different examples in the book were created using different tools, which gives people a good feel of what's possible in each and how "hard" or "easy" the coding itself is for the different options. Different chapters discuss different types of data that you could encounter: patterns over time, proportions, relationships, ...

There were some minor points in the book that I'd mention if they asked me to review it (but that's according to me, and I don't want to pretend to be an expert). First of all, it would have been nice if Nathan had gone a little bit deeper into theories behind what is seen as good visualization. In the first chapter ("Telling Stories with Data") he does mention Cleveland & McGill in a side-note, but I think that information (along with Gestalt laws, etc) definitely deserves one or two full paragraphs, if not half a chapter. I also don't completely agree with the use of a stacked barchart (about page 109). From my experience, they're worth less than the time it takes to create them. After all, it's impossible to compare any groups other than the one that is at the bottom (and therefore has a common "zero"-line). For example: look at the first picture below. This shows the number of "stupid things done" by women and men, stratified over 5 different groups (A-F). Although it is easy to compare total stupidity per group (group C is doing particularly bad), as well as that for men, we can't see which of the groups A, D or F scores the worst for women. And that's because they don't have a common origin. We could of course put the women next to the men, but then we'd loose the total numbers.

In the second plot, however, it is possible to compare women, men and totals. The bars for women are put next to those for men, but I've added a shaded larger bar at the back that shows the sum of the two. This plot was originally created in R using ggplot2, but I'm afraid I can't find back the reference that explained how to do this... Let me know if you can find it.

The contents of the book of course is not world-shattering. But that's not the point of the book. For people new to the field it's a great addition to their library (and I learned a thing or two myself as well). If you're interested in data visualization, go out and get it.

Visualizing the Tour de France

UPDATE: I encountered a blog post by Martin Theus describing a very similar approach for looking at this same data (see here).

Disclaimer 1: This is a (very!) quick hack. No effort was put in it whatsoever regarding aesthetics, interactivity, scaling (e.g. in the barcharts), ... Just wanted to get a very broad view of what happened during the Tour de France (= biggest cycling event each year).
Disclaimer 2: I don't know anything about cycling. It was actually my wife who had to point out to me which riders could be interesting to highlight in the visualization. But that also meant that this could become interesting for me to learn something about the Tour.

Data was copied from the Tour de France website (e.g. for the 1st stage). Visualization was created in processing.

The parallel coordinate plot shows the standings of all riders over all 21 stages. No data was available for stage 2, because that was a team time-trial (so discard that one). At the top is the rider who came first, at the bottom who came last. Below the coordinate plot are little barcharts displaying the distribution in arrival time (in "number of seconds later than the winner") for all riders in that stage.

The highlighted riders are: Cavendish (red), Evans (orange), Gilbert (yellow), Andy Schleck (light blue) and Frank Schleck (dark blue).

So what was I able to learn from this?

• Based on the barcharts you can guess which trips were in the mountains, and which weren't. You'd expect that the riders become much more separated in the mountains than on the flat. In the very last stage in Paris, for example, everyone seems to have arrived in one big group. Whereas for stages 12-14 the riders were much more spread. So my guess (and that's confirmed by checking this on the TourDeFrance website :-) is that those were mountain stages.
• You can see clear groups of riders who behave the same. There is for example a clear group of riders who performed quite badly in stage 19 but much better in stage 20 (and bad in 21 again).
• As the parallel coordinate plots were scaled according to the initial number of riders, we can clearly see how people left the Tour because the "bottom" of the later stages are empty.
• We see that Cavendish (red) has very erratic performance. And it seems to co-incide with trips where the arrival times are spread out (= mountain trips?). This could mean that Cavendish is good on the flats, but bad in the mountains. Question to those who know something about cycling: is that true?
• Philippe Gilbert started good (both on the flats and in the mountains), but became more erratic halfway through the Tour.

TenderNoise - visualizing noise levels

A couple of days ago I bumped into this tweet by Benjamin Wiederkehr (@datavis): "Article: TenderNoise http://datavis.ch/q9pIxq" It describes a visualization by Stamen Design and others displaying noise levels at different intersections in San Francisco. They recorded these levels over a period of a few days in order to get an idea of auditory pollution. More information is here.

Although this particular visualization might be very useful for the people involved, I would like to explain some of the issues that I have with it, coming from a data-visualization-for-pattern-finding viewpoint.

I think there are many things that might be gleaned from this data which are not possible with the current visualization:

• Is there a relationship between the noise patterns at different intersections? Based on the graphic at the bottom, we can conclude that on average noise level goes down during the night and up during daytime, but it would be nice if the visualization would give an indication of any aberrant patterns as well. Are there intersections that behave differently from others?
• I don't see a real use for changing the graphic over time. I suspect that small multiples of area charts would work better to demonstrate the change over time (as e.g. the visual used here). Using the current approach it is very difficult to see how particular intersections change over time because (a) the display changes and you loose temporal context, and (b) the resolution is so hight that the blobs just flicker.
• Concerning that flicker, it might be an option to bin the data in larger time blocks. For calculating the value in each block different approaches should be investigated, like the average value, the maximum, the minimum, or the most extreme value (be it maximum or minimum, based on comparison with the average).

It'd be interesting to get hold of these data and work on some alternatives (given the time...)