How figures are made
The figure style, the sizes and resolutions, the statistics windows, the colours, and why no curve is ever smoothed.
Every figure of Geinium is drawn by one figure module with one set of rules. The figure on the Results page, the image file of the results package and the figure in the PDF report are the same drawing: tested on real jobs, the figure shown on the page and the package image were identical byte for byte. What you see is what you publish.
Sizes and widths
Figures are designed at their printed size, in the three column widths journals use: 85 mm (single column), 114 mm (one and a half columns) and 174 mm (double column, also the text width of the report). Every figure has a natural width and a list of widths it supports. When another width is requested, the figure is laid out again for that width: fewer ticks, a legend below the panels, panels rearranged. It is never scaled, so the type is always at its intended size. A figure that has no layout for a width keeps its natural width, and the export says so.
| Width | Figures with this natural width | Typical use |
|---|---|---|
| 85 mm | Radius of gyration, SASA, buried area, hydrogen bonds, MM-GBSA over time, components and correlation, eigenvalue spectrum, probability density, free-energy map | Single-column figures of a manuscript. |
| 114 mm | PCA in three dimensions, free-energy surface | Figures that need more room than one column but not the full page width. |
| 174 mm | RMSD, RMSF, secondary structure, interaction fractions and timeline, torsion profile, ligand properties, run quality and others | Multi-panel figures; full page width. |
Resolution and formats
PNG files are written at 300 dpi, or at 600 dpi on request, with a white background and the resolution stored in the file, so a layout program places them at the right size. SVG and PDF are vector files: text stays text, lines stay lines, and they can be scaled without loss. The output is deterministic: the same input gives byte-identical PNG and SVG files, and byte-identical PDFs when a fixed time stamp is given.
Fonts and lines
Figures use a standard sans-serif typeface, embedded in every PDF; sequences and atom names use a monospaced one. At printed size, tick labels are 7 pt, axis labels 8 pt, legends 7 pt, annotations 6.5 pt, and panel letters 9 pt bold lower case (a, b) outside the top left corner of a panel. Frames are open (left and bottom axes only, 0.6 pt), ticks point outwards, there is no grid, and the image carries no title: the title belongs to the caption, as in a journal. Axis labels read "Quantity (unit)". Data lines are 1.0 pt, or 0.6 pt for series with more than 2000 points; mean and reference lines are dashed.
Colours
The categorical colours are indigo, amber, cyan, garnet, violet and gold, completed by slate and ink. They were chosen by a search over simulated colour-vision deficiencies: under normal, protanope, deuteranope and tritanope vision the smallest colour difference between any two of the six is 22.5 CIE76 units, so they stay distinguishable for readers with the common forms of colour blindness. One quantity has one colour in every figure, for example C-alpha always indigo and the ligand fitted on the protein always garnet. Secondary structure uses violet for helices and deep gold for strands. Maps use colour scales whose lightness changes monotonically.
| Limit | Consequence |
|---|---|
| The six colours are at least 7.5 lightness units apart, but slate (heavy atoms) and violet (side chains) are not. | In a grey-scale print these two lines of the RMSD figure have the same tone and are told apart by position and legend order. The same holds for some compound lines and stacked contact classes. |
| Beyond eight series the colours repeat. | Repeated colours are drawn dashed, dotted or dash-dotted. |
No smoothing
Every time series is the plain line through its per-frame values. There is no running mean, no smoothed band and no envelope anywhere. A smoothed curve looks calmer than the data and hides exactly the jumps a reader needs to see, such as a ligand leaving its pocket. Up to 20,000 points a line is drawn point for point; maps have one column per frame down to the size of a 600 dpi pixel. Statistics are given in the table beside each figure, never drawn into it as a band.
Replicas in figures
With several replicas, each replica is drawn as its own line (solid, dashed, dotted) in its own colour. Replicas are never averaged into one curve, because an average of independent runs is not a trajectory and can show behaviour that none of them had. Maps over residues and time get one part per replica. Distributions pool the frames of all replicas, and the caption says so.
Statistics windows
The first part of a production run still carries the memory of the restrained equilibration.
The statistics of the tables therefore leave out the first 10 % of the production run (10 to 100 ns
for a 100 ns run), unless the project file sets analysis.start_ns. Time series still
show the whole run, with a dashed vertical line where the window starts, so nothing is hidden. The
note under every table names its window, for example "10-100 ns; the first 10 % ... excluded as
equilibration". Contacts and MM-GBSA use the window of their own stage, which the note also
states. With one replica, the standard deviation is over the frames of that trajectory; with
several, it is the standard deviation between the replica means. See
Trajectory analyses.
Checked before a report is written
Two checks tie the figures and tables to the data. During development every drawn series was
compared with its table to 1e-12. At every export, every number of every statistics table is
recomputed from the raw-data files of the package, with a tolerance of half a unit of its last
printed digit; a single mismatch stops the report (the finding numbers.raw_data). The
export log states the result, for example "229 of 229 equal".
Residue axes
Proteins with gaps, fusion inserts or several molecules keep their original residue numbers. Residue axes are compact: they leave out the numbers that are not in the structure, mark gaps with a separator, and label each molecule and segment (A1, A2) in a strip above the axis. The sequence figure of Structure preparation shows the layout on a receptor with a fusion insert. Residue labels in figures read "TYR 148"; a chain prefix is added only when the protein has several chains, and insertion codes are kept ("SER 100A").