Publication-Quality Figure Design

Hard rules

1. No fabricated citations. Every cited work must resolve to a verifiable
2. No claim without provenance. Every quantitative or factual claim
3. No silent failure. Every script invocation, API call, or tool use must declare its exit status and what to do on non-zero. A skill that silently swallows errors is a violation.

When to use

Trigger this skill when any of the following apply:

• Producing a figure for a journal article, preprint, thesis chapter, or grant.
• Preparing supplementary figures.
• A reviewer or co-author has flagged a figure for legibility, resolution, or color issues.
• You need a multi-panel composite and want consistent style across panels.
• You are finalizing exploratory plots for a public talk, poster, or web page.

When NOT to use

• Pure exploratory data analysis — use seaborn defaults or plotly.express directly.
• Slide-only graphics — see ors-scientific-visualization-slides-design.
• Poster-only graphics — see ors-scientific-visualization-poster-design.
• Schematic / cartoon figures — see ors-scientific-visualization-schematics-diagrams.
• Color and accessibility audits of an existing figure — see ors-scientific-visualization-color-and-accessibility.

Prerequisites

• Python 3.9+ with matplotlib, numpy, pandas, and ideally seaborn and plotly.
• For R users: a current ggplot2 and (optionally) patchwork / cowplot for composition.
• One journal's author guidelines in hand (column widths, accepted formats).
• A working directory layout: figures/, data/, src/ (recommended).

Core workflow

1. Decide the target. Identify the destination (journal name, screen vs. print) before drawing anything. This fixes width, font size, and DPI.
2. Pick the chart family. Match the question to the chart: distribution, comparison, relationship, composition, change over time. See the decision table below.
3. Set global style once. Apply a publication rcParams block (or theme_set in R) so every panel in the manuscript shares font, color cycle, and DPI. Do not restyle per-panel.
4. Encode data, not chartjunk. Remove top/right spines, gridlines on bar charts, 3D effects, shadows, and decorative backgrounds. Direct-label whenever feasible.
5. Label with units. Every axis gets a name plus a unit in parentheses, e.g. Time (min), Concentration (µM). Use sentence case, not ALL CAPS.
6. Use redundant encoding. Pair color with shape, line style, or marker — never rely on hue alone. See the color-and-accessibility skill.
7. Compose panels with gridspec / patchwork. A 2×2 layout with shared legends, consistent margins, and bold panel labels (A, B, C, D).
8. Export in the right format. Vector (PDF, SVG, EPS) for plots; raster (TIFF, PNG) at 300+ DPI for photos and microscopy. Never export plots as JPEG.
9. Verify at final size. Open the file at 100% intended print size and read every label. If you cannot, shrink your screen to ~25% and read.

Code patterns

A. Publication rcParams (apply once, project-wide)

import matplotlib as mpl

mpl.rcParams.update({"
    "font.family": "sans-serif",
    "font.sans-serif": ["Arial", "Helvetica", "DejaVu Sans"],
    "font.size": 8,
    "axes.labelsize": 9,
    "axes.titlesize": 9,
    "xtick.labelsize": 7,
    "ytick.labelsize": 7,
    "legend.fontsize": 7,
    "axes.linewidth": 0.8,
    "lines.linewidth": 1.2,
    "lines.markersize": 4,
    "xtick.major.width": 0.8,
    "ytick.major.width": 0.8,
    "xtick.direction": "out",
    "ytick.direction": "out",
    "figure.dpi": 100,        # screen preview
    "savefig.dpi": 300,       # raster export
    "savefig.bbox": "tight",
    "pdf.fonttype": 42,       # embed real fonts, not Type 3
    "ps.fonttype": 42,
})

B. Vector vs raster decision matrix

Rule of thumb: plots go vector, images go raster. Never export a vector plot as JPEG.

C. DPI / size reference table (common journals)

Approximate final-figure widths. Always confirm in the journal's "Information for Authors".

Convert to matplotlib with figsize=(width_in, height_in). Height is yours to choose; 0.6–0.75 of width is a safe starting point for single-panel plots.

D. Single-panel line plot with error bars

import matplotlib.pyplot as plt
import numpy as np

# data: mean ± SEM per timepoint per group
t = np.array([0, 1, 2, 4, 8, 24])
ctrl_mean = np.array([1.00, 0.98, 1.01, 1.02, 0.99, 0.97])
ctrl_sem   = np.array([0.04, 0.05, 0.04, 0.06, 0.05, 0.05])
treat_mean = np.array([1.00, 1.10, 1.25, 1.40, 1.32, 1.18])
treat_sem   = np.array([0.05, 0.06, 0.07, 0.08, 0.07, 0.06])

fig, ax = plt.subplots(figsize=(3.5, 2.5))   # Nature single column
ax.errorbar(t, ctrl_mean,  yerr=ctrl_sem,  fmt="o-",
            label="Control", color="#0072B2", capsize=2, lw=1.0, ms=3.5)
ax.errorbar(t, treat_mean, yerr=treat_sem, fmt="s-",
            label="Treatment", color="#D55E00", capsize=2, lw=1.0, ms=3.5)

ax.set_xlabel("Time (h)")
ax.set_ylabel("Relative expression (AU)")
ax.spines["top"].set_visible(False)
ax.spines["right"].set_visible(False)
ax.legend(frameon=False, loc="best")
fig.tight_layout()
fig.savefig("figure1.pdf")            # vector
fig.savefig("figure1.png", dpi=300)   # raster fallback

E. Multi-panel with gridspec (matplotlib)

from string import ascii_uppercase
import matplotlib.pyplot as plt

fig = plt.figure(figsize=(7.2, 4.0))  # Nature double column
gs = fig.add_gridspec(2, 2, hspace=0.55, wspace=0.45,
                      left=0.08, right=0.98, top=0.92, bottom=0.10)

axes = [fig.add_subplot(gs[i // 2, i % 2]) for i in range(4)]

# axes[0].plot(...), axes[0].bar(...), etc.
# axes[0].set_xlabel("..."); axes[0].set_ylabel("...")

for ax, letter in zip(axes, ascii_uppercase):
    ax.text(-0.18, 1.06, letter, transform=ax.transAxes,
            fontsize=10, fontweight="bold", va="top", ha="right")
    ax.spines["top"].set_visible(False)
    ax.spines["right"].set_visible(False)

# Optional: shared legend at the figure level
# fig.legend(handles, labels, loc="lower center", ncol=4, frameon=False)
# fig.subplots_adjust(bottom=0.18)

F. Seaborn for statistical comparison

import seaborn as sns
import matplotlib.pyplot as plt

sns.set_theme(context="paper", style="ticks", font_scale=1.0,
              rc={"axes.spines.top": False, "axes.spines.right": False})

fig, ax = plt.subplots(figsize=(3.5, 2.8))
sns.boxplot(data=df, x="treatment", y="response",
            order=["Control", "Low", "High"], palette="colorblind",
            fliersize=0, ax=ax)        # hide outliers
sns.stripplot(data=df, x="treatment", y="response",
              order=["Control", "Low", "High"], color="black",
              alpha=0.4, size=2.5, ax=ax)
ax.set_xlabel("Treatment")
ax.set_ylabel("Response (µM)")

G. Plotly → static export

import plotly.io as pio

# Make Plotly use the same paper-style fonts
pio.kaleido.scope.default_format = "png"
pio.kaleido.scope.default_scale = 3   # ~300 DPI for 96 DPI default

fig.update_layout(
    font=dict(family="Arial, sans-serif", size=10, color="black"),
    plot_bgcolor="white",
    paper_bgcolor="white",
    margin=dict(l=60, r=20, t=40, b=50),
)
fig.update_xaxes(showline=True, linewidth=1, linecolor="black", mirror=False)
fig.update_yaxes(showline=True, linewidth=1, linecolor="black", mirror=False)

fig.write_image("figure_interactive.html")           # for the web
fig.write_image("figure_static.png", scale=3)         # for the paper

H. ggplot2 equivalent (R)

library(ggplot2)
library(patchwork)   # for multi-panel composition

p1 <- ggplot(df, aes(t, mean, color = group)) +
  geom_line(linewidth = 0.4) +
  geom_errorbar(aes(ymin = mean - sem, ymax = mean + sem), width = 0.15) +
  scale_color_manual(values = c(Control = "#0072B2", Treatment = "#D55E00")) +
  labs(x = "Time (h)", y = "Relative expression (AU)", color = NULL) +
  theme_classic(base_size = 8, base_family = "Arial") +
  theme(legend.position = "top",
        plot.margin = margin(5, 5, 5, 5, "pt"))

p1 + p1 + plot_layout(ncol = 1)         # compose with patchwork
ggsave("figure1.pdf", width = 3.5, height = 2.5, units = "in")

I. R / ggplot2 — paper theme

theme_paper <- function(base_size = 8, base_family = "Arial") {
  theme_classic(base_size = base_size, base_family = base_family) %+replace%
    theme(
      axis.line = element_line(linewidth = 0.4, colour = "black"),
      axis.ticks = element_line(linewidth = 0.4, colour = "black"),
      legend.key = element_blank(),
      strip.background = element_blank(),
      plot.margin = margin(4, 4, 4, 4, "pt")
    )
}

Color choice (quick reference; full audit in color-and-accessibility)

• Sequential continuous data: viridis, cividis, plasma, magma.
• Diverging with a meaningful center (e.g., correlation, log fold-change): RdBu_r, BrBG, PuOr.
• Categorical up to ~6 groups: Okabe-Ito (#E69F00, #56B4E9, #009E73, #F0E442, #0072B2, #D55E00, #CC79A7, #000000).
• Always test in grayscale (plt.imshow(..., cmap="Greys") after a color transform) and with a CVD simulator.

Typography and legibility

• Sans-serif throughout. Arial / Helvetica for print, DejaVu Sans as the matplotlib fallback.
• Sentence case for axis labels: Time (min), not TIME (MIN).
• Always include units in parentheses — never leave bare axis labels.
• Use µ, not u. Use the minus sign (U+2212), not a hyphen. Use × for multiplication in legends.
• Bold sparingly: only for the one element the reader should anchor on.

Multi-panel conventions

• Panels labeled with bold uppercase A, B, C, D (Nature uses lowercase a, b, c — match the journal).
• Place labels in the top-left of each panel, just outside the axes, not inside the data.
• Share x or y axes when the scales are identical, to make side-by-side comparison honest.
• Align the baselines of bar charts and the y-axes of like units.

Scale bars and inset axes

• For microscopy, prefer a scale bar over a stated magnification in the caption.
• Draw the scale bar in the lower-right of the panel, in white if the background is dark, black otherwise.
• For geographic or spatial figures, include a north arrow AND a scale bar (or scale text).

Statistical annotation

• Always state the n in the figure or caption (n = 3 biological replicates).
• Mark significance with asterisks and report the test in the caption (* p < 0.05, ** p < 0.01, two-sided t-test).
• Show individual data points (jittered or strip) on top of summary bars/boxes; the dot cloud carries more information than the mean alone.
• Specify the dispersion measure (SD, SEM, 95% CI) — never let the reader guess.

Legends, captions, and the reader

• A figure legend inside the panel is for symbol-to-series mapping. A figure caption is the standalone explanation.
• Direct-label small multiples; reserve the figure-level legend for series comparisons.
• Captions should answer: what is plotted, what are the n and the error measure, what does the test show, and what the reader should conclude.

Common pitfalls

• PDF font fallback to Type 3: garbled text on the publisher's end. Set pdf.fonttype=42.
• Raster at 72 DPI: blurry when the figure is scaled up. Set savefig.dpi=300 and verify with pdfimages -list.
• Color-only encoding: indistinguishable for ~8% of male readers. Add markers, line styles, or panels.
• Junk axis: top and right spines, double axes, broken axes without a break mark.
• Truncated y-axis on a bar chart: exaggerates differences. Start at zero unless scientifically justified, and mark the break.
• Inconsistent styles across panels: different fonts, sizes, or color cycles between figure A and figure B. Set rcParams once, at the top of the project.
• "Preview" figure saved as the final: the test version of figure1.png ships in the submission. Version-control your outputs.

Validation

• Open the final PDF in Illustrator/Inkscape and confirm fonts are outlines-or-embedded, not substituted.
• Run pdfimages -list figure.pdf and check every embedded image is ≥ 300 DPI.
• Convert the PDF to grayscale (gs -sDEVICE=pdfwrite -sColorConversionStrategy=Gray ...) and check the figure still tells its story.
• Run a CVD simulator (Coblis, Sim Daltonism) over a PNG export and confirm series remain distinguishable.
• Compare panel-by-panel: same font size, same color cycle, same line weight, same axis style.
• Print a single panel at target size on paper. If the caption is illegible from 2 feet, the figure is not publication-ready.

Open alternatives

References

Internal cross-links to other ors-* skills:

• ors-scientific-visualization-color-and-accessibility — pick and audit palettes.
• ors-scientific-visualization-schematics-diagrams — non-data illustrative figures.
• ors-scientific-visualization-slides-design — adaptation for projection.
• ors-scientific-visualization-poster-design — adaptation for large format.
• ors-bioinformatics-omics-* — domain-specific plot families (volcano, MA, UMAP, circos, etc.).
• ors-scientific-writing — caption and figure-callout guidance.

External resources (do not fabricate exact paths):

• matplotlib official documentation — matplotlib.org/stable/contents.html
• matplotlib rcParams reference
• seaborn tutorial — seaborn.pydata.org/tutorial.html
• ggplot2 documentation — accessed via ?ggplot2 in R, or the official tidyverse site
• patchwork R package (multi-panel ggplot2)
• ColorBrewer 2.0 (Cynthia Brewer) — colorbrewer2.org
• viridis colormap family — documentation site
• WCAG 2.1 contrast guidelines — w3.org/TR/WCAG21/

Changelog

• 1.0.0 (2026-06-10): Initial adaptation by Pradyumna Jayaram from scientific-visualization (K-Dense Inc.). Removed AI image-generation scripts, restructured around the publication rcParams + gridspec pattern, added ggplot2 R patterns and a journal-width reference table.

Cross-references

Other skills in this category:

• color-and-accessibility
• poster-design
• schematics-diagrams
• slides-design