Crafting Scientific Visualizations – Creative Process & Best Practices

Publiceret Juli 2014

Scientific visualization is a field that combines the complexities of science, the technical rigor of programming, the challenges of effective teaching and the creative possibilities of art and design. Publications, conferences and workshops devote considerable attention to the tools, techniques and data sets used in scientific visualization. Despite educational research that informs us on how visualizations impact target audiences like students, there is little discussion about the process of crafting visualizations and how it impacts those involved. Many artists and animators report anecdotally that scientists with whom they collaborate to create visualizations often gain new insights into their science: ‘visual thinking' triggered during the planning of a visualization is thought to put familiar data into a new light. Why? What is it about this process that triggers such realizations? Several reasons are likely.

First, visualization is fundamentally about data integration. It is about piecing together evidence - often from disparate (not to mention contradictory) sources into one coherent story. Although researchers consider many different types of data when thinking about their field and planning experiments, there is something unique about the requirements of creating visual representations of data that stimulate new types of thinking.

Second, when tasked with depicting a structure or process, gaps in knowledge inevitably become salient - they become holes in the visual tapestry or narrative. Visualization is not the only way to notice these ‘negative spaces' in the data, but it is a powerful process by which to make them more apparent. Important ‘next step' experiments can reveal themselves as a result. Incidentally, this is one of the important benefits of crafting a ‘model figure' - it is a process that requires a clear definition of the model's characteristics prior to visual representation and therefore clarifies one's thinking (Iwasa, 2010). Although in this respect it is conceptually no different than the writing or teaching process, visualization focuses our attention on specific aspects of the data that we may not otherwise consider.

Third, visualization is inherently a collaborative process whereby a content expert often partners with a scientifically-knowledgeable visualization expert. This often leads to considering alternate points of view as one attempts to communicate or teach the material. It leads to fruitful discussions and questions that challenge the expert to externalize and clarify ideas that may otherwise remain buried or unchallenged in the mind's eye. One often ‘take sides' in the crafting of a visualization - although it IS about integration of multiple data sets, it is also often about picking ONE point of view from which to tell a story. As noted above, in the same way that the writing process can catalyze clarity of ideas by committing words to paper, visualization catalyzes mental models to images in a way that challenges the illustrator, animator or programmer to produce a possible representation - as opposed to an ensemble of partially-formed scenarios in one's mind.

A related but unfortunate observation is that while creating visualizations requires synthesis of disparate types of data and the crafting of a story, we seldom find visualizations that tackle depiction of competing scientific models or hypotheses. Are such visualizations inherently challenging to craft without the risk of losing the narrative thread? Or does it have more to do with the financial pressures and realities of commissioning visualizations (i.e. people who invest the time or budget to develop a visualization often do so to depict their own research and conclusions rather than offer a survey of the field and its competing viewpoints)? Regardless of the possible reasons, we should embrace and seek out opportunities to create visualizations that address multiple or competing models (Figure 1).

2014-3-mcgill-fig_01_alternateModels_small
Figure 1. Visualizing alternate models. Screenshot from a visualization of Notch signaling shows how a temporary split-screen layout can help viewers assess competing models for ligand engagement (created for Stephen Blacklow, Harvard Medical School. Visualization by Dan Nowakowski & Gaël McGill, Digizyme Inc., http://www.digizyme.com/notch.html)

There are several benefits to depicting these within the same narrative structure. For one, it changes the way in which the viewer engages with the visualization - instead of being a passive observer who experiences a linear narrative, presentation of competing models stimulates the viewer's involvement in data interpretation, integration and refinement: which model best suits the data? In other words, visualizations of competing models fundamentally shift the narrative structure away from a directed storytelling approach to one that invites the viewer to develop their own point of view. These kinds of visualizations could also lead to more acceptance by a community who may otherwise see a partisan/biased approach (one that potentially lowers the credibility in the visualization itself and its authors).

Incidentally, visualizations of competing models may also remind non-expert audiences (namely students) that these are only interpretations of scientific data - not the data themselves. Although highly engaging and immersive visualizations offer students inspirational entry points into science, we must also be wary that the visualization become synonymous with the science. In the same way that Magritte's provocative painting ‘Ceci n'est pas une pipe' reminds us of the dichotomy between object and representation, students must be reminded of the interpretative visual language that bridges the gap between data and visualization (Figure 2). Instead of blindly accepting the latter, we should help students become aware of this gap and engage them in becoming more critical viewers. In fact engaging students in the visualization process itself has been shown to yield beneficial outcomes and help instructors identify students' misconceptions (www.picturingtolearn.org).

2014-3-mcgill-fig_02_ceciNestPasUneProteine
Figure 2. Ceci n'est pas une proteine. Homage to Magritte's famous painting ‘La Trahison des Images.' In the same way that the surrealist painter provokes us to consider the impossible reconciliation of image and object, we must remind students of the interpretive distance between scientific visualizations and the data upon which they are based (Image by Gaël McGill, Digizyme Inc.)

Finally, addressing other aspects of the scientific visualization process may also help in gaining acceptance by the scientific community. For example, little attention is paid to the way in which visualizations are referenced. Unlike research papers or scientific reviews, visualizations seldom include any type of reference information, let alone a well-crafted, systematic bibliography commensurate with the number and variety of data sources used. In the case of static visualizations like images, a great example of well-referenced work can be seen in the series of articles written by David Goodsell to accompany some of his watercolor paintings of cellular landscapes (Goodsell, 2009). Eventually, viewers should be able to interrogate and link directly from different parts of an image - in the same way that digital publications now offer links to articles within the main text of a manuscript, such an interactive visualization referencing system would offer ‘in context' information to help viewers link background data with selected visual elements.

For dynamic visualizations - be they linear animations or interactive game-like experiences - the added challenge is to invent a reference system that not only reflects the many different uses for data in visualizations but also a system that makes these references seamlessly accessible within the constantly evolving visualization (Jantzen et al., 2014). This type of referencing would make it easier for viewers to discern the quality of underlying data used in visualizations and thereby increase overall credibility of the medium (Bio-cinema Verite, Nature Methods 2012).

As the field matures the need to identify additional best practices has also arisen through conferences like Copenhagen's SciViz in September 2012 (http://projects.au.dk/sciviz/) or publications like Nature Methods' series of short ‘Points of View' 1-pagers offering effective design concepts for data visualization (Wong & Kjærgaard). The interest in design best practices is likely born out of a desire to increase confidence in the accuracy of content and care in the crafting of these visualizations. In commercial and professional settings, a ‘best practice' typically refers to a method empirically shown to result in better outcomes. However, can the field as a whole agree on which are the desired outcomes and therefore which design best practices should be adopted?

Certain visualizations aim to engage and inspire novice audiences while others focus exclusively on data visualization to aid the analysis of results from experiments or simulations. While we may strive to identify and disseminate design best practices to improve the consistency and quality of scientific visualizations, we must also be cautious of prescriptive practices and standardization in a field where innovation in visual style can be an important part of the process.

In the same way that a painter would likely reject a system that requires her to use specific colors to depict certain elements of a scene, a scientific illustrator or animator would undoubtedly feel that freedom in depiction style is an important aspect of her work - in service not only of aesthetics, but also pedagogy. This is a key challenge in establishing scientific visualization best practices since the activity is a dynamic combination of scientific and design thinking. Which aspects are amenable to ‘codification' via best practices and which should be left alone?

In summary, one of the most meaningful and yet little recognized beneficial aspects of scientific visualization is the extent to which the conceptualization and production process can change one's understanding of the underlying science. Despite years or decades of knowledge in a particular area, a scientist can gain new perspective when creating a visualization. Better visibility as to the benefits of this process has the potential to change the scientific community's understanding of visualization and increase confidence in the resulting work. In addition we have reviewed other opportunities to increase the accuracy and credibility of scientific visualizations through improved referencing, depiction of competing hypotheses and overall more transparency as to the interpretive nature of scientific visualizations.

In light of these observations, I would offer that the line separating ‘pure' data visualization and more immersive visualizations is mostly fictitious. The latter can be just as much about data analysis and synthesis: fundamentally, good science visualization IS data visualization. Although in certain cases the desired outcome may not exceed that of audience engagement - a goal for which scientific accuracy is sometimes unnecessarily sacrificed - animators or illustrators should always strive to synthesize as much available data as possible. Only then can an informed decision be made as to what data can bear omission in their visualization - regardless of the target audience.

References

Bio-cinema verité? Nat. Meth. 9, 1127 (2012).

Frankel, Felice. Picturing to Learn (http://www.picturingtolearn.org)

Goodsell, D.S. Neuromuscular synapse. Biochem. Mol. Biol. Educ. 37, 204-210 (2009).

Iwasa, J.H. Animating the model figure. Trends Cell Bio. 20, 699-704 (2010).

Jantzen S., Jenkinson J., McGill G. Transparency in Film: Using Citation to Increase the Credibility of Scientific Animation (submitted)

Wong, B., Kjærgaard, R. S. Pencil and Paper, Nat Meth. 9, 1037 (2012).