listen; there's a hell of a good universe next door: let's go.go there

# space: beautiful

Bioinformatics and Genome Analysis Course. Izmir International Biomedicine and Genome Institute, Izmir, Turkey. May 2–14, 2016

# visualization + design

Cover image accompanying our article on mouse vasculature development. Biology turns astrophysical. PNAS 1 May 2012; 109 (18) (zoom, PNAS)

# Creating the PNAS Cover

One of my goals in life, which I can now say has been accomplished, is to make biology look like astrophysics. Call it my love for the Torino Impact Hazard Scale.

Recently, I was given an opportunity to attend to this (admittedly vague) goal when Linda Chang from Aly Karsan's group approached me with some microscopy photos of mouse veins. I was asked to do "something" with these images for a cover submission to accompany the manuscript.

When people see my covers, sometimes they ask "How did you do that?" Ok, actually they never ask this. But being a scientist, I'm trained me to produce answers in anticipation of such questions. So, below, I show you how the image was constructed.

The image was published on the cover of PNAS (PNAS 1 May 2012; 109 (18))

## Tools

Photoshop CS5, Nik Color Efex Pro 4, Alien Skin Bokeh 2 and a cup of coffee from a Rancilio Silvia.

## source images

Below are a few of the images I had the option to work with. These are mouse embryonic blood vessels, with a carotid artery shown in the foreground with endothelial cells in green, vascular smooth muscle cells in red and the nuclei in blue.

Of course, as soon as I saw the images, I realized that there was very little that I needed to do to trigger the viewer's imagination. These photos were great!

Mouse carotid arteries. (zoom)
Mouse carotid arteries. (zoom)
Mouse carotid arteries. (zoom)
Mouse carotid arteries. (zoom)
Mouse carotid arteries. (zoom)
Mouse carotid arteries. (zoom)
Mouse carotid arteries. (zoom)
Mouse carotid arteries. (zoom)
Mouse carotid arteries. (zoom)

## memories of star trek

Immediately I thought of two episodes of Star Trek (original series): Doomsday Machine and the Immunity Syndrome, as well as of images from the Hubble Telescope.

Enterprise is about to be consumed by a horror tube: a planet killer! (The Doomsday Machine)
Enterprise heads into a giant amoeba. Who eats whom? I'll let you guess. (The Immunity Syndrome)
Orion nebula (M42) as seen by the Hubble telescope. (zoom)

I though it would be pretty easy to make the artery images look all-outer-spacey. They already looked it.

## centerpiece image

And then I saw the image below.

A particularly spectacular image of a mouse carotid artery. I'm thinking 10 on the Torino scale. (zoom)

## constructing the cover

### background

The background was created from the two images shown here. The second image was sampled three times, at different rotations.

Images used for background. (zoom)
Images used for background. (zoom)
Layer composition for background elements. (zoom)

The channel mixer was used to remove the green channel and leave only red and blue.

Background elements for PNAS cover image. (zoom)

### middle ground

The next layer was composed of what looked like ribbons of blue gas. This was created by sampling the oval shapes from the source images. Here the red channel was a great source for cloud shapes, and this was the only channel that was kept. The hue was shifted to blue and a curve adjustment was applied to increase the contrast.

First set of middle ground elements, before adjustments. (zoom)
First set of middle ground elements, after channel adjustments. (zoom)
Second set of middle ground elements. (zoom)
Layer composition for middle ground elements. (zoom)

When the foreground and middle ground elements were combined, the result was already 40 parsecs away.

Background and foreground elements for PNAS cover image. (zoom)

### foreground

The foreground was created from the spectacular comet-like image of a mouse artery. Very little had to be done to make this element look good. It already looked good.

I applied a little blur using Alien Skin's Bokeh 2 to narrow the apparent depth of field, masked out elements at the bottom of the image and removed some of the green channel. The entire blue channel was removed altogether (this gave the tail of the comet a mottled, flame-like appearance).

Foreground element, before adjustments. (zoom)
Foreground element, after channel adjustments. (zoom)
Layer composition for foreground element. (zoom)

### post processing

Initial composition of background, middle ground and foreground elements. (zoom)
40% localized application of Nik's Tonal Contrast (Color Efex 4 plugin) to increase structure in red channel. (zoom)
50% blend with Nik's Pro Contrast (Color Efex 4 plugin). (zoom)

And here we have the final image.

Final PNAS cover. Spacey! (zoom)

# Gene Volume Control

Thu 11-06-2015

I was commissioned by Scientific American to create an information graphic based on Figure 9 in the landmark Nature Integrative analysis of 111 reference human epigenomes paper.

The original figure details the relationships between more than 100 sequenced epigenomes and genetic traits, including disease like Crohn's and Alzheimer's. These relationships were shown as a heatmap in which the epigenome-trait cell depicted the P value associated with tissue-specific H3K4me1 epigenetic modification in regions of the genome associated with the trait.

Figure 9 from Integrative analysis of 111 reference human epigenomes (Nature (2015) 518 317–330). (details)

As much as I distrust network diagrams, in this case this was the right way to show the data. The network was meticulously laid out by hand to draw attention to the layered groups of diseases of traits.

Network diagram redesign of the heatmap for a select set of traits. Only relationships with –log P > 3.9 are displayed. Appears on Graphic Science page in June 2015 issue of Scientific American. (details)

This was my second information graphic for the Graphic Science page. Last year, I illustrated the extent of differences in the gene sequence of humans, Denisovans, chimps and gorillas.

# Sampling distributions and the bootstrap

Thu 11-06-2015

The bootstrap is a computational method that simulates new sample from observed data. These simulated samples can be used to determine how estimates from replicate experiments might be distributed and answer questions about precision and bias.

Nature Methods Points of Significance column: Sampling distributions and the bootstrap. (read)

We discuss both parametric and non-parametric bootstrap. In the former, observed data are fit to a model and then new samples are drawn using the model. In the latter, no model assumption is made and simulated samples are drawn with replacement from the observed data.

Kulesa, A., Krzywinski, M., Blainey, P. & Altman, N (2015) Points of Significance: Sampling distributions and the bootstrap Nature Methods 12:477-478.

Krzywinski, M. & Altman, N. (2013) Points of Significance: Importance of being uncertain. Nature Methods 10:809-810.

# Bayesian statistics

Thu 30-04-2015

Building on last month's column about Bayes' Theorem, we introduce Bayesian inference and contrast it to frequentist inference.

Given a hypothesis and a model, the frequentist calculates the probability of different data generated by the model, P(data|model). When this probability to obtain the observed data from the model is small (e.g. alpha = 0.05), the frequentist rejects the hypothesis.

Nature Methods Points of Significance column: Bayesian Statistics. (read)

In contrast, the Bayesian makes direct probability statements about the model by calculating P(model|data). In other words, given the observed data, the probability that the model is correct. With this approach it is possible to relate the probability of different models to identify one that is most compatible with the data.

The Bayesian approach is actually more intuitive. From the frequentist point of view, the probability used to assess the veracity of a hypothesis, P(data|model), commonly referred to as the P value, does not help us determine the probability that the model is correct. In fact, the P value is commonly misinterpreted as the probability that the hypothesis is right. This is the so-called "prosecutor's fallacy", which confuses the two conditional probabilities P(data|model) for P(model|data). It is the latter quantity that is more directly useful and calculated by the Bayesian.

Puga, J.L, Krzywinski, M. & Altman, N. (2015) Points of Significance: Bayes' Theorem Nature Methods 12:277-278.

Puga, J.L, Krzywinski, M. & Altman, N. (2015) Points of Significance: Bayes' Theorem Nature Methods 12:277-278.

# Bayes' Theorem

Wed 22-04-2015

In our first column on Bayesian statistics, we introduce conditional probabilities and Bayes' theorem

P(B|A) = P(A|B) × P(B) / P(A)

This relationship between conditional probabilities P(B|A) and P(A|B) is central in Bayesian statistics. We illustrate how Bayes' theorem can be used to quickly calculate useful probabilities that are more difficult to conceptualize within a frequentist framework.

Nature Methods Points of Significance column: Bayes' Theorem. (read)

Using Bayes' theorem, we can incorporate our beliefs and prior experience about a system and update it when data are collected.

Puga, J.L, Krzywinski, M. & Altman, N. (2015) Points of Significance: Bayes' Theorem Nature Methods 12:277-278.