Vis capstone address: Designing visualizations to enable molecular insights


Journal Article

Our laboratory research goal of understanding 3D molecular structures depends strongly on the development and use of visualization systems. We especially value their effectiveness for enabling scientific insights, in our original research process as well as in the communication of those results, which leads us to emphasize interactive functionality and pragmatic effectiveness over perfection as presentation images. Therefore we design alternative molecular visualizations tuned to the medium, the user, and the specific scientific problem at hand; feedback from our own use improves them further. Visualizations sometimes prompt scientific insight by suggesting metaphors from other realms, such as art. Examples influential in our work range from Greek vases to origami to Picasso. For students or colleagues approaching molecular structure from a verbal or algebraic pattern of thought, we work to develop their "3D molecular literacy". Best are interactive graphics exercises that let them explore, compare, and measure at their own pace, to master inherently 3D relationships such as handedness or hydrogen-bond geometry. In our structure validation and improvement work, one challenge is clearly visualizing the outliers from many quality measures simultaneously on the molecular structure, to allow both rapid location of 3D problem clusters and interactively-updated local detail as the model is modified. Our current implementations support effective and routine use in structural biology, but we seek further improvement of their completeness and clarity. A current quest is finding the right molecular problem and developing the measurement and analysis interface to show if an immersive virtual environment (the Duke DiVE) can provide superior scientific results over simpler systems. We plan to test interpreting the highly local experimental data that determine NMR structures. A recent visualization success that enabled new molecular insight is a simple but powerful way of interactively visualizing and clustering high-dimensional data. It enabled us to classify cleanly the noisy 7-dimensional dihedral-angle data that describes RNA backbone conformations, and to perceive the shortcomings of earlier attempts done either by automated clustering or by combining multiple 3D projections of the data.

Duke Authors

Cited Authors

  • Richardson, J

Published Date

  • September 1, 2006

Published In

Volume / Issue

  • 12 / 5

International Standard Serial Number (ISSN)

  • 1077-2626

Citation Source

  • Scopus