Bio Reporting Figure Export

Version Compatibility

Reference examples tested with: matplotlib 3.8+, numpy 1.26+, ggplot2 3.5+, ggrastr 1.0+, ragg 1.2+

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show matplotlib then help(matplotlib.figure.Figure.savefig); introspect matplotlib.rcParams if a key is renamed
• R: packageVersion('ggplot2') then ?ggsave

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Publication-Ready Figure Export

"Export this figure for the journal" -> Save the plot so each part of it survives production: vector structure stays crisp, the dense data layer stays a reasonable file size, text stays editable, and color survives the print conversion.

• Python: fig.savefig('fig.pdf', dpi=600) with publication rcParams set
• R: ggsave('fig.pdf', width=89, units='mm', device=cairo_pdf)

The Load-Bearing Idea: A Figure Is Four Layers

A publication figure is not one object; it is four superimposed layers, and export means giving each the representation it needs:

1. Vector structure - axes, ticks, spines, gridlines, fit lines, error bars, annotations. Resolution-independent; must stay vector so the typesetter can scale it to 89 mm without pixelation.
2. Raster data layer - the dense part: a scatter with 10^5-10^7 points, a heatmap, a micrograph. Drawing a million points as a million vector circles makes a hundreds-of-MB PDF that crashes Illustrator. This layer wants to be pixels.
3. Type - every glyph. Editors need it to stay selectable text, not flattened paths or pixels baked into the raster.
4. Color encoding - the data-to-color mapping. A scientific choice (perceptual uniformity, color-vision-deficiency safety, grayscale survival) that also interacts with the RGB->CMYK conversion the journal performs without asking.

The expert move is the hybrid figure: rasterize only layer 2, keep layers 1 and 3 vector, embed editable fonts, and pick a colormap that survives CMYK and grayscale. Everything below serves that.

A reproducibility framing: a figure is a pure function of (data, code, theme, font availability). If any of those is unpinned, the figure is not reproducible.

The Hybrid Figure (rasterize only the dense layer)

The single most important export skill for single-cell (UMAP/tSNE) and GWAS (Manhattan) figures. A fully-vector million-point scatter is unopenable and gets rejected by the typesetter's RIP; rasterizing just the data layer keeps file size sane while axes and text stay crisp at any zoom.

import matplotlib.pyplot as plt
fig, ax = plt.subplots(figsize=(3.5, 3.0))     # physical inches = the real size control
ax.scatter(x, y, s=2, rasterized=True, zorder=0)   # 10^6 points -> embedded raster
ax.plot(xfit, yfit, color='black', zorder=2)        # stays vector
ax.set_xlabel('UMAP1')                              # stays vector text
fig.savefig('fig.pdf', dpi=600)                     # dpi governs ONLY the rasterized layer

In a vector container, savefig(dpi=...) sets the resolution of the embedded raster patch and does nothing to the vector parts. ax.set_rasterization_zorder(0) rasterizes every artist below a z-order cleanly. In R use ggrastr::rasterise(geom_point(size=0.3), dpi=600) (wraps any geom since 0.2.0), keeping theme_* vector.

Fonts: Keep the Text Editable

The most common typesetter complaint about matplotlib PDFs is un-editable text. matplotlib's default pdf.fonttype is 3 (Type 3), which embeds glyphs as PostScript procedures that import into Illustrator as ungrouped paths and cannot be re-selected as text. Set 42 (TrueType, wrapped) so text stays selectable and editable. Fix it once, globally:

import matplotlib as mpl
mpl.rcParams['pdf.fonttype'] = 42      # TrueType, editable text in PDF
mpl.rcParams['ps.fonttype']  = 42      # same for EPS/PS
mpl.rcParams['svg.fonttype'] = 'none'  # SVG: emit real <text>, reference the font by name

Type 42/TrueType/Type 3 fonts are subsetted (only used glyphs embedded); Type 1 are not. With large glyph sets (CJK) Type 42 can bloat the PDF - a real tradeoff. svg.fonttype='none' keeps words editable in Inkscape/Illustrator but the viewer must have the font (else it substitutes); the default 'path' outlines every glyph (portable, uneditable). Default to editable text; only outline if a specific production desk asks.

In R, the base pdf() device has weak font handling and inconsistent cross-OS rendering; use cairo_pdf (embeds fonts and supports alpha). For raster, ragg::agg_png()/agg_tiff() render anti-aliased text better than cairo/base devices. Rule of thumb: showtext for vector devices, ragg for raster.

Color Space: The Author Works in RGB, Print Is CMYK

Screens are additive RGB; offset print is subtractive CMYK. matplotlib and ggplot2 author in RGB only - there is no honest path to a true CMYK figure from them. The journal's pipeline converts RGB->CMYK, and because the CMYK gamut is smaller than sRGB, saturated out-of-gamut colors shift: pure RGB blue (#0000FF) and vivid green/cyan come back muddier and darker on paper. The neon scatter that pops on screen can print gray.

What to do: pull colors slightly off full saturation (they survive conversion better); soft-proof downstream in Illustrator/Photoshop with a CMYK profile if it matters; and submit RGB - Nature, Science, Cell, and PLOS all explicitly want RGB, not CMYK, because their pipeline does the conversion and online is RGB anyway. If a legacy desk demands CMYK, convert downstream with an explicit profile and re-check that nothing shifted; do not fake it with a colorspace flag.

Transparency: EPS Has No Alpha

EPS/PostScript do not support alpha - matplotlib's PS backend renders partially-transparent artists as opaque, so an alpha-blended overplotted scatter loses its density information on EPS export. PDF and SVG support alpha natively; prefer them when transparency carries meaning. If a journal forces EPS and transparency is needed, rasterize that layer (rasterized=True) or re-encode density as hexbin/2D-KDE. savefig(transparent=True) makes the background transparent for slide overlays, not for print.

DPI Is Meaningless for Vector

A vector PDF has no inherent resolution; it renders sharp at any zoom. DPI governs only raster formats (PNG/TIFF) and the rasterized data layer inside a vector file. The real size control is the physical figure size in inches/mm - design at the journal's exact column width from the start; rescaling a 300-dpi raster to 200% halves its effective resolution. Font sizes are in points (1 pt = 1/72 in) independent of DPI.

The DPI tiers follow the IMAGE CLASS, because print reproduces tone via halftone dots (follow the target journal's own numbers; these are the common production convention):

savefig.dpi is the file resolution; figure.dpi is the on-screen resolution. Independently of DPI, very thin strokes (below ~0.25 pt / 0.1 mm) can drop out or thicken unpredictably at the printer's RIP even in a vector file - keep hairlines at or above the journal's minimum line weight.

Format Decision

Never JPEG for line/text figures - block compression rings along high-contrast edges (gray halos on text, fringing on thin lines). For TIFF, use LZW (near-universal reader support) for 8-bit figures; use ZIP for 16-bit (LZW can inflate 16-bit files).