Martin Krzywinski / Genome Sciences Center / Martin Krzywinski / Genome Sciences Center / - contact me Martin Krzywinski / Genome Sciences Center / on Twitter Martin Krzywinski / Genome Sciences Center / - Lumondo Photography Martin Krzywinski / Genome Sciences Center / - Pi Art Martin Krzywinski / Genome Sciences Center / - Hilbertonians - Creatures on the Hilbert Curve
This love's a nameless dream.Cocteau Twinstry to figure it outmore quotes

In Silico Flurries: Computing a world of snow. Scientific American. 23 December 2017

visualization + design

Getting into Visualization of Large Biological Data Sets

The 20 imperatives of information design

Martin Krzywinski, Inanc Birol, Steven Jones, Marco Marra

Presented at Biovis 2012 (Visweek 2012). Content is drawn from my book chapter Visualization Principles for Scientific Communication (Martin Krzywinski & Jonathan Corum) in the upcoming open access Cambridge Press book Visualizing biological data - a practical guide (Seán I. O'Donoghue, James B. Procter, Kate Patterson, eds.), a survey of best practices and unsolved problems in biological visualization. This book project was conceptualized and initiated at the Vizbi 2011 conference.

If you are interested in guidelines for data encoding and visualization in biology, see our Visualization Principles Vizbi 2012 Tutorial and Nature Methods Points of View column by Bang Wong.

Martin Krzywinski @MKrzywinski
Getting into Visualization of Large Biological Data Sets. M Krzywinski, I Birol, S Jones, M Marra (poster presentation) (PDF, Illustrator)

The 20 imperatives of information design


Create legible visualizations with a strong message. Make elements large enough to be resolved comfortably. Bin dense data to avoid sacrificing clarity.

Distinguish between exploration and communication.

Use exploratory tools (e.g. genome browsers) to discover patterns and validate hypotheses. Avoid using screenshots from these applications for communication – they are typically too complex and cluttered with navigational elements to be an effective static figure.

Do not exceed resolution of visual acuity.

Our acuity is ~50 cycles/degree or about 1/200 (0.3 pt) at 10 inches. Ensure the reader can comfortably see detail by limiting resolution to no more than 50% of acuity. Where possible, elements that require visual separation should be at least 1 pt part.

Use no more than ~500 scale intervals.

Ensure data elements are at least 1 pt on a two-column Nature figure (6.22 in), 4 pixels on a 1920 horizontal resolution display, or 2 pixels on a typical LCD projector. These restrictions become challenges for large genomes.

Show variation with statistics.

Data on large genomes must be downsampled. Depict variation with min/max plots and consider hiding it when it is within noise levels. Help the reader notice significant outliers.

Do not draw small elements to scale.

Map size of elements onto clearly legible symbols. Legibility and clarity are more important than precise positioning and sizing. Discretize sizes and positions to facilitate making meaningful comparisons.

Aggregate data for focused theme.

A strong visual message has no uncertainty in its interpretation. Focus on a single theme by aggregating unnecessary detail.

Show density maps and outliers.

Establishing context is helpful when emergent patterns in the data provide a useful perspective on the message. When data sets are large, it is difficult to maintain detail in the context layer because the density of points can visually overwhelm the area of interest. In this case, consider showing only the outliers in the data set.

Consider whether showing the full data set is useful.

The reader’s attention can be focused by increasing the salience of interesting patterns. Other complex data sets, such as networks, are shown more effectively when context is carefully edited or even removed.


Match the visual encoding to the hypothesis. Use encodings specific and sensitive to important patterns. Dense annotations should be independent of the core data in distinct visual layers.

Use the simplest encoding.

Choose concise encodings over elaborate ones.

Help the reader judge accurately.

Accuracy and speed in detecting differences in visual forms depends on how information is presented. We judge relative lengths more accurately than areas, particularly when elements are aligned and adjacent. Our judgment of area is poor because we use length as a proxy, which causes us to systematically underestimate.

Use encodings that are robust and comparable.

In addition to being transparent and predictable, visualizations must be robust with respect to the data. Changes in the data set should be reflected by proportionate changes in the visualization. Be wary of force-directed network layouts, which have low spatial autocorrelation. In general, these are neither sensitive nor specific to patterns of interest.

Crop scale to reveal fine structure in data.

Biological data sets are typically high-resolution (changes at base pair level can meaningful), sparse (distances between changes are orders of magnitude greater than the affected areas) and connect distant regions by adjacency relationships (gene fusions and other rearrangements). It is difficult to take these properties into account on a fixed linear scale, the kind used by traditional genome browsers. To mitigate this, crop and order axis segments arbitrarily and apply a scale adjustment to a segment or portion thereof.

Use perceptual palettes.

Selecting perceptually favorable colors is difficult because most software does not support the required color spaces. Brewer palettes exist for the full range of colors to help us make useful choices. Qualitative palettes have no perceived order of importance. Sequential palettes are suitable for heat maps because they have a natural order and the perceived difference between adjacent colors is constant. Twin hue diverging palettes, are useful for two-sided quantitative encodings, such as immunofluorescence and copy number.

Never use hue to encode magnitude.

Hue does not communicate relative change in values because we perceive hue categorically (blue, green, yellow, etc). Changes within one category have less perceptual impact than transitions between categories. For example, variations across the green/yellow boundary are perceived to be larger than variations across the same sized hue interval in other parts of the spectrum.


Well-designed figures illustrate complex concepts and patterns that may be difficult to express concisely in words. Figures that are clear, concise and attractive are effective – they form a strong connection with the reader and communicate with immediacy. These qualities can be achieved with methods of graphic design, which are based on theories of how we perceive, interpret and organize visual information.

Reduce unnecessary variation.

The reader does not know what is important in a figure and will assume that any spatial or color variation is meaningful. The figure’s variation should come solely from data or act to organize information.

Encapsulate details.

Including details not relevant to the core message of the figure can create confusion. Encapsulation should be done to the same level of detail and to the simplest visual form. Duplication in labels should be avoided.

Use consistent alignment. Center on theme.

Establish equivalence using consistent alignment. Awkward callouts can be avoided if elements are logically placed.

Respect natural hierarchies.

When the data set embodies a natural hierarchy, use an encoding that emphasizes it clearly and memorably. The use hierarchy in layout (e.g. tabular form) and encoding can significantly improve a muddled figure.

Palette for color blindness / Martin Krzywinski @MKrzywinski
This 15-color palette provides good discrimination for common color blindness types. Individuals with tritanopia cannot distinguish colors marked with ● and ◥. (hires)

Be aware of the luminance effect.

Color is a useful encoding – the eye can distinguish about 450 levels of gray, 150 hues, and 10-60 levels of saturation, depending on the color – but our ability to perceive differences varies with context. Adjacent tones with different luminance values can interfere with discrimination, in interaction known as the luminance effect.

Be aware of color blindness.

In an audience of 8 men and 8 women, chances are 50% that at least one has some degree of color blindness. Use a palette that is color-blind safe. In the palette below the 15 colors appear as 5-color tone progressions to those with color blindness. Additional encodings can be achieved with symbols or line thickness.

I have designed 15-color palettes for color blindess for each of the three common types of color blindness.


news + thoughts

Curse(s) of dimensionality

Tue 05-06-2018
There is such a thing as too much of a good thing.

We discuss the many ways in which analysis can be confounded when data has a large number of dimensions (variables). Collectively, these are called the "curses of dimensionality".

Martin Krzywinski @MKrzywinski
Nature Methods Points of Significance column: Curse(s) of dimensionality. (read)

Some of these are unintuitive, such as the fact that the volume of the hypersphere increases and then shrinks beyond about 7 dimensions, while the volume of the hypercube always increases. This means that high-dimensional space is "mostly corners" and the distance between points increases greatly with dimension. This has consequences on correlation and classification.

Altman, N. & Krzywinski, M. (2018) Points of significance: Curse(s) of dimensionality Nature Methods 15:399–400.

Statistics vs Machine Learning

Tue 03-04-2018
We conclude our series on Machine Learning with a comparison of two approaches: classical statistical inference and machine learning. The boundary between them is subject to debate, but important generalizations can be made.

Inference creates a mathematical model of the datageneration process to formalize understanding or test a hypothesis about how the system behaves. Prediction aims at forecasting unobserved outcomes or future behavior. Typically we want to do both and know how biological processes work and what will happen next. Inference and ML are complementary in pointing us to biologically meaningful conclusions.

Martin Krzywinski @MKrzywinski
Nature Methods Points of Significance column: Statistics vs machine learning. (read)

Statistics asks us to choose a model that incorporates our knowledge of the system, and ML requires us to choose a predictive algorithm by relying on its empirical capabilities. Justification for an inference model typically rests on whether we feel it adequately captures the essence of the system. The choice of pattern-learning algorithms often depends on measures of past performance in similar scenarios.

Bzdok, D., Krzywinski, M. & Altman, N. (2018) Points of Significance: Statistics vs machine learning. Nature Methods 15:233–234.

Background reading

Bzdok, D., Krzywinski, M. & Altman, N. (2017) Points of Significance: Machine learning: a primer. Nature Methods 14:1119–1120.

Bzdok, D., Krzywinski, M. & Altman, N. (2017) Points of Significance: Machine learning: supervised methods. Nature Methods 15:5–6.

...more about the Points of Significance column

Happy 2018 `\pi` Day—Boonies, burbs and boutiques of `\pi`

Wed 14-03-2018

Celebrate `\pi` Day (March 14th) and go to brand new places. Together with Jake Lever, this year we shrink the world and play with road maps.

Streets are seamlessly streets from across the world. Finally, a halva shop on the same block!

Martin Krzywinski @MKrzywinski
A great 10 km run loop between Istanbul, Copenhagen, San Francisco and Dublin. Stop off for halva, smørrebrød, espresso and a Guinness on the way. (details)

Intriguing and personal patterns of urban development for each city appear in the Boonies, Burbs and Boutiques series.

Martin Krzywinski @MKrzywinski
In the Boonies, Burbs and Boutiques of `\pi` we draw progressively denser patches using the digit sequence 159 to inform density. (details)

No color—just lines. Lines from Marrakesh, Prague, Istanbul, Nice and other destinations for the mind and the heart.

Martin Krzywinski @MKrzywinski
Roads from cities rearranged according to the digits of `\pi`. (details)

The art is featured in the Pi City on the Scientific American SA Visual blog.

Check out art from previous years: 2013 `\pi` Day and 2014 `\pi` Day, 2015 `\pi` Day, 2016 `\pi` Day and 2017 `\pi` Day.

Machine learning: supervised methods (SVM & kNN)

Thu 18-01-2018
Supervised learning algorithms extract general principles from observed examples guided by a specific prediction objective.

We examine two very common supervised machine learning methods: linear support vector machines (SVM) and k-nearest neighbors (kNN).

SVM is often less computationally demanding than kNN and is easier to interpret, but it can identify only a limited set of patterns. On the other hand, kNN can find very complex patterns, but its output is more challenging to interpret.

Martin Krzywinski @MKrzywinski
Nature Methods Points of Significance column: Machine learning: supervised methods (SVM & kNN). (read)

We illustrate SVM using a data set in which points fall into two categories, which are separated in SVM by a straight line "margin". SVM can be tuned using a parameter that influences the width and location of the margin, permitting points to fall within the margin or on the wrong side of the margin. We then show how kNN relaxes explicit boundary definitions, such as the straight line in SVM, and how kNN too can be tuned to create more robust classification.

Bzdok, D., Krzywinski, M. & Altman, N. (2018) Points of Significance: Machine learning: a primer. Nature Methods 15:5–6.

Background reading

Bzdok, D., Krzywinski, M. & Altman, N. (2017) Points of Significance: Machine learning: a primer. Nature Methods 14:1119–1120.

...more about the Points of Significance column

Human Versus Machine

Tue 16-01-2018
Balancing subjective design with objective optimization.

In a Nature graphics blog article, I present my process behind designing the stark black-and-white Nature 10 cover.

Nature 10, 18 December 2017