User:Karsten Theis/AI tutorial
From Proteopedia
This is a tutorial for AI how to help with making 3D figures in Jmol. If it helps humans too, so much the better.
|
Contents |
My take on high-quality figures
This is paraphrased from Image:Proteopedia rubrics.pdf.
A 3D figure should be content-rich, visually clear, fun to explore, and understandable. If it is part of a series of figures, they should work together well.
- Content-rich: If you can, tell a complete story in a single figure rather than having to click through multiple figures. Journal articles sometimes limit the number of figures to as low as 5, and so they are information-packed.
- Visually clear: There should be a clear sense of what the focus of the figure is (the foreground), and how this focal point connects to the whole (the background). There are multiple aspects that can help achieve a clear image (choice of initial zoom level, color, of representation, slabbing and fogging, line widths, initial orientation and center of rotation)
- Fun to explore: The viewer should be invited to rotate the figure and to zoom in and out. Interactive elements (hiding and displaying structural features, buttons to get back to the original view or to other standard views, animation, morphs) can guide this exploration, as well as reminders about Proteopedia's and Jmol's built-in features (measure distances and angles, invitation to view in stereo, pop-out the image to maximize viewing area).
- Understandable: The viewer should be able to understand the visual elements (color coding, which representation for what) and the source of the data (PDB ID, typically) through the caption, legends on the viewing window, permanent and hover labels, and the text around the green link.
Annotated scripts
Here are some scripts. If you click on the title, you can see the resulting figure in the Jmol window above.
Detailed views
These have a focal point which should be centered and visible in the foreground. The focal point could be a ligand, an (empty) ligand binding site, an interface (e.g. dimer interface), a single interaction (e.g. disulfide bridge), or a single molecule. The background could be the entire remainder of the structure, or carefully selected parts of it, or blank (e.g. for a single molecule).
Typically, the foreground is shown at the detail of single atoms (represented by wireframe, spacefill, both "ball-and-stick", surfaces), and it is the center of rotation. Often, the background is shown with thinner lines, uniform color, or not showing single atoms but groups (represented by backbone, cartoon, meshribbon, ribbon, strand, trace, or surface). If the viewer is encouraged to rotate 360 degrees, the background should not interfere with seeing the foreground (use thin lines, slabbing, fading/fogging "zshade"). Transparency should be used with caution in Jmol because it can add visual clutter (spacefill and thick wireframe as well as cartoons add clutter, strands of constant thickness work well).
Overall views
These don't have a focal point. They should be centered on everything displayed. If using a representation you can look through (e.g. backbone or cartoon rather than surface or spacefill), fading/fogging "zshade" may be helpful to distinguish front and back.
