---

name: scientific-figures description: | Comprehensive guide for creating publication-ready scientific figures. Covers technical requirements (DPI, formats, sizing), best practices for various plot types, panel figure assembly, caption writing, color accessibility, and language-specific tips for Python and R.

Integrates with claude-codex-gemini-consensus skill for figure generation during analysis execution and report generation phases. All figures undergo multi-model validation for scientific accuracy.

routing_description: | Use this skill when creating figures for scientific publications, manuscripts, posters, or presentations. Covers all aspects from technical specifications to aesthetic design and accessibility.

routing_keywords:

• figure
• figures
• plot
• plots
• visualization
• chart
• graph
• publication figure
• manuscript figure
• scientific figure
• ggplot
• matplotlib
• seaborn
• panel figure
• figure caption
• DPI
• resolution

---

Scientific Figures: Publication-Ready Guide

Comprehensive skill for creating publication-quality figures for scientific manuscripts, posters, and presentations.

Integration with Consensus Workflow

This skill integrates with the claude-codex-gemini-consensus workflow:

ANALYSIS EXECUTION (from analysis/plan.md)
    │
    ├── [EXECUTE] Run analyses per plan
    │
    ├── [FIGURES] Generate figures using this skill  ← YOU ARE HERE
    │   ├── Create standalone figure scripts (analysis/scripts/fig01_*.py)
    │   ├── Export to analysis/figures/
    │   ├── Write publication-ready captions
    │   └── Validate with Codex + Gemini
    │
    └── [REPORT] Include figures in analysis/report.md

Figure validation requirements:

• All figures reviewed by Claude + Codex + Gemini before inclusion in report
• Scientific accuracy verified against analysis results
• Technical specifications checked (DPI, format, sizing)
• Accessibility validated (colorblind-safe, readable fonts)

---

Core Design Principles

Clarity Beats Decoration

• Prefer 2D plots; avoid 3D effects, excessive gradients, chartjunk
• Use direct labeling over legends when feasible (reduces eye movement)
• Maintain consistent typography, colors, and axis styling across the entire manuscript
• Each panel should answer: "What should the reader learn in 5 seconds?"

Show the Data, Not Just Summaries

• For small/medium n (≤30-50 per group), show individual data points + summary statistics
• Avoid bar charts for continuous distributions; prefer box/violin + overlaid points
• When using summary statistics, always report: measure of center (mean/median) + measure of spread (SD/IQR/CI)

Statistical Transparency

• Define error bars explicitly in every figure caption (SD, SEM, 95% CI)
• Report exact n for each group/condition
• Show exact P-values where journal allows (not just "P < 0.05")
• State correction method for multiple comparisons (Bonferroni, FDR, etc.)

---

Technical Requirements

Resolution (DPI) Standards

Figure Type	Minimum DPI	Recommended DPI	Notes
Line art (diagrams, graphs with lines)	600	1000	Black/white only, no grayscale
Halftone (photographs, images)	300	300-400	Continuous tone images
Combination (text + images)	500	600	Mixed line art and halftone
Color figures	300	300-400	CMYK for print, RGB for online

Warning: Web images (72 DPI) are NOT acceptable for publication. Always capture/export at print resolution.

File Formats

Format	Use Case	Pros	Cons
TIFF	Final submission	Lossless, universally accepted	Large file size
EPS	Vector graphics	Scalable, editable	Not for photographs
PDF	Vector + raster combined	Preserves layers, fonts	Variable support
PNG	Screen/web, drafts	Lossless compression	Not ideal for print
SVG	Vector graphics, web	Fully scalable, editable	Limited journal support
JPEG	Never for publication	-	Lossy compression artifacts

Recommended workflow:

1. Create figures in vector format (SVG/PDF)
2. Export final as TIFF (300-600 DPI) for submission
3. Keep source files for revisions

Figure Dimensions

Nature family journals:

• Single column: 88 mm (3.46 inches)
• 1.5 column: 120 mm (4.72 inches)
• Double column: 180 mm (7.09 inches)
• Maximum height: 230 mm (9.06 inches)

General guidelines:

• Design at FINAL print size (don't scale down later)
• Square subplots: ~3 x 3 inches each
• Aspect ratios: 4:3 or 16:9 for presentations

Font Specifications

Element	Size Range	Recommended
Axis labels	8-12 pt	9-10 pt
Tick labels	7-10 pt	8-9 pt
Legend text	7-10 pt	8-9 pt
Panel labels (A, B, C)	10-14 pt	12 pt bold
Figure title	10-12 pt	10 pt bold

Font families:

• Sans-serif (Arial, Helvetica): Cleaner for figures, better screen readability
• Serif (Times): Match manuscript body text
• Keep consistent throughout all figures

Line Weights and Markers

Element	Minimum	Recommended	Notes
Plot lines	0.5 pt	1.0-1.5 pt	Thicker for emphasis
Axis lines	0.5 pt	0.8 pt	Frame the plot
Grid lines	0.25 pt	0.3-0.5 pt	Subtle, don't dominate
Error bars	0.5 pt	1.0 pt	Match data lines
Markers	3 pt	5-8 pt	Visible at final size

Warning: Avoid hairlines (<0.25 pt). They may disappear in print or become invisible when scaled.

---

Panel Figures (Multi-Panel Layout)

Panel Labeling Conventions

Journal Style	Format	Example
Nature, Science	Lowercase bold	(a), (b), (c)
Cell, NEJM	Uppercase bold	A, B, C
AHA journals	Uppercase bold Arial 12.6pt	A, B, C
JMR	Lowercase, lower-left	(a), (b), (c)

Best practices:

• Position labels consistently (typically upper-left corner)
• Do not box labels or add periods after them
• Use 12 pt bold sans-serif (Arial/Helvetica)
• Maintain 2-3 mm padding from plot edges

Hierarchy for Complex Figures

When panels have sub-components:

Preferred: A, B, C → i, ii, iii → (1), (2), (3)
           Ai, Aii, Bi, Bii

Avoid:     Aa, Ab, Ba, Bb (confusing hierarchy)

Layout Guidelines

GOOD: Horizontal reading order         BAD: Vertical-first layout
┌─────┬─────┬─────┐                    ┌─────┬─────┐
│  A  │  B  │  C  │                    │  A  │  D  │
├─────┼─────┼─────┤                    ├─────┼─────┤
│  D  │  E  │  F  │                    │  B  │  E  │
└─────┴─────┴─────┘                    ├─────┼─────┤
                                       │  C  │  F  │
                                       └─────┴─────┘

• Arrange panels left-to-right, top-to-bottom
• Align baselines and axes where appropriate
• Use consistent spacing between panels (3-5 mm)
• Submit multi-panel figures as ONE file

Panel Narrative Logic

Order panels to tell a coherent story:

Setup → Key Result → Validation → Mechanism → Robustness/Generalization

Panel Position	Typical Content
A, B	Study design, cohort overview, methods schematic
C, D	Primary findings (main hypothesis test)
E, F	Supporting evidence, mechanistic insight
G+	Validation cohort, sensitivity analyses, extensions

Principles:

• Each panel = one message; move secondary analyses to Supplement
• Same axis limits when comparisons are intended
• Same transforms (log, z-score) for comparable panels
• Consistent color/symbol mapping across all panels

---

Figure Types: When to Use What

Categorical Comparisons

Plot Type	Use When	Avoid When
Bar chart	Comparing categories, counts, proportions	>10 categories, showing distributions
Grouped bar	Comparing categories across groups	Many groups (cluttered)
Stacked bar	Part-to-whole relationships	Comparing middle segments
Dot plot	Few categories, precise values	Large datasets

Bar chart requirements:

• Start y-axis at zero (unless log scale)
• Include error bars with clear definition (SD, SEM, 95% CI)
• Order bars meaningfully (by value or logical grouping)

Distributions

Plot Type	Use When	Avoid When
Histogram	Single distribution shape	Comparing groups
Box plot	Comparing distributions across groups	Need to see all data points
Violin plot	Distribution shape + comparison	Few data points per group
Density plot	Smooth distribution estimate	Discrete data
Beeswarm/strip plot	Small-n, show all points	Large datasets

Box plot conventions:

• Box: IQR (25th-75th percentile)
• Line: Median
• Whiskers: 1.5×IQR or min/max
• Points: Outliers beyond whiskers
• Always define in caption

Relationships & Correlations

Plot Type	Use When	Avoid When
Scatter plot	Two continuous variables	Overplotting (>1000 points)
Line plot	Time series, connected observations	Unordered data
Heatmap	Matrix data, correlations, expression	Few variables
Bubble chart	Three variables (x, y, size)	Too many bubbles

Scatter plot requirements:

• Include correlation coefficient if relevant (r or ρ)
• Add regression line with confidence band if showing relationship
• Consider jittering for overlapping discrete values
• Use transparency (alpha) for dense data

Survival Analysis (Kaplan-Meier)

Required elements:

1. Step function curves for each group
2. Censoring marks (tick marks or dots) — define in caption
3. Number at risk table below x-axis
4. P-value from log-rank test (or specify test used)
5. Hazard ratio with 95% CI if comparing groups
6. Median survival with 95% CI for each group

Example structure:

┌────────────────────────────────────────┐
│  [Kaplan-Meier curves with CI bands]   │
│  [Censoring tick marks on curves]      │
├────────────────────────────────────────┤
│  Number at risk:                       │
│  Group A: 120  98  76  54  32  18  8   │
│  Group B: 115  85  62  41  25  12  4   │
│           0   12  24  36  48  60  72   │
│                 Time (months)          │
└────────────────────────────────────────┘

Forest Plots (Meta-analysis, Subgroups)

Required elements:

1. Study/subgroup labels on left
2. Effect sizes (HR, OR, RR) with 95% CI
3. Horizontal lines spanning CI
4. Vertical line at null effect (HR=1, OR=1)
5. Point size proportional to weight/sample size
6. Summary diamond for pooled estimate
7. Heterogeneity statistics (I², Q, p-value)

Volcano Plots (Differential Expression)

Required elements:

1. X-axis: log2 fold change
2. Y-axis: -log10(p-value) or -log10(FDR)
3. Horizontal threshold line (significance cutoff)
4. Vertical threshold lines (fold change cutoffs)
5. Color coding: up (red), down (blue), NS (gray)
6. Label top significant genes/features
7. Report total counts in each category

Heatmaps

Required elements:

1. Clear row and column labels (or clustering dendrogram)
2. Color bar with scale and units
3. Appropriate color palette (see Accessibility section)
4. Clustering method stated in caption if used
5. Normalization/scaling method stated

ROC and Precision-Recall Curves

ROC curves:

1. X-axis: False positive rate (1 - Specificity)
2. Y-axis: True positive rate (Sensitivity)
3. Diagonal reference line (random classifier)
4. AUC with 95% CI (bootstrap or DeLong)
5. Define positive class explicitly in caption

Precision-Recall (PR) curves:

• Use when classes are imbalanced (rare outcomes) — ROC can be misleading
• X-axis: Recall (Sensitivity)
• Y-axis: Precision (PPV)
• Horizontal reference line at prevalence (baseline)
• Report AUPRC (Area Under PR Curve)

Caption must include:

• Definition of positive class
• Number of positives and negatives in test set
• Confidence interval method (bootstrap replicates, DeLong)
• Whether curves are from training, validation, or held-out test set

---

Image Integrity (Microscopy, Gels, Blots)

Critical for peer review. Image manipulation is a leading cause of retractions.

Acceptable Adjustments

Allowed	Not Allowed
Global brightness/contrast	Selective region enhancement
Cropping (with disclosure)	Splicing from different images
Color balance correction	Cloning/removing objects
Uniform scaling	Non-linear distortions

Key rules:

1. Apply adjustments to the entire image, not selectively
2. Retain original/raw files for all images
3. Disclose any non-trivial processing in Methods or caption
4. If cropping, indicate in caption; show full image in Supplement if requested

Specific Requirements

Microscopy:

• Include scale bars (calibrated, not estimated)
• State magnification, microscope type, acquisition settings
• For representative images: state number of replicates (e.g., "Representative of n=5 independent experiments")

Gels and Blots:

• Show full uncropped blot in Supplement (many journals require this)
• If lanes are rearranged, clearly demarcate with dividing lines
• State exposure time and whether image was inverted
• Never remove bands or splice non-adjacent lanes without clear disclosure

Radiology/Medical Imaging:

• Remove or anonymize patient identifiers
• State window/level settings if relevant
• Include anatomical orientation markers

Reference

HHS Office of Research Integrity: Image Guidelines

---

Color Guidelines

Colorblind-Accessible Palettes

~8% of men and ~0.5% of women have color vision deficiency. Half of published biological figures are partially or fully inaccessible.

Recommended palettes:

Palette	Type	Use Case	Package
viridis	Sequential	Continuous data	matplotlib, viridis (R)
cividis	Sequential	Colorblind-optimized viridis	matplotlib
plasma	Sequential	High contrast continuous	matplotlib
RdBu (diverging)	Diverging	Centered data (z-scores)	ColorBrewer
Okabe-Ito	Qualitative	Categorical (8 colors)	Manual definition

Colors to AVOID:

• Red + Green combinations (most common CVD)
• Red + Black combinations
• Rainbow color schemes
• Pure red for important elements

Okabe-Ito palette (8 colorblind-safe colors):

#E69F00  Orange
#56B4E9  Sky Blue
#009E73  Bluish Green
#F0E442  Yellow
#0072B2  Blue
#D55E00  Vermillion
#CC79A7  Reddish Purple
#999999  Gray

Additional Accessibility Strategies

1. Use shapes in addition to colors (circles, squares, triangles)
2. Use line styles (solid, dashed, dotted) with colors
3. Test in grayscale — figure should still be interpretable
4. Add direct labels instead of relying on color legends
5. Use thick lines (1.5-2 pt minimum)

Testing Tools

• Viz Palette (online): https://projects.susielu.com/viz-palette
• colorblindr (R): Simulates CVD on ggplot figures
• Color Oracle (desktop): System-wide CVD simulation

---

Caption Writing

Structure (IMRAD for Figures)

Figure X. [TITLE — bold, one sentence describing the figure]

[DESCRIPTION — what is shown, not the results]
(a) Description of panel a. (b) Description of panel b.

[DEFINITIONS — all abbreviations, symbols, colors]
Error bars represent mean ± SEM (n = X per group).
*P < 0.05, **P < 0.01, ***P < 0.001 (test name).

[STATISTICS — exact values]
HR = 0.65 (95% CI: 0.48–0.88), P = 0.005 (log-rank test).

[SOURCE — if adapted from published work]
Adapted from [Reference] with permission.

Caption Checklist

• Brief title (bold) summarizes the entire figure
• Each panel described in order (a, b, c...)
• All abbreviations defined (even if defined in text)
• All symbols explained (*, †, ‡)
• Color coding explained
• Error bar definition (SD, SEM, 95% CI) with n
• Statistical test named with degrees of freedom
• Exact P-values (not just <0.05)
• Sample sizes for each group
• Scale bars defined (for images/microscopy)
• Figure is understandable without reading main text

Examples

Good caption (Kaplan-Meier):

Figure 2. Overall survival by treatment group.

Kaplan-Meier curves showing overall survival for patients receiving
treatment A (blue, n = 120) versus treatment B (orange, n = 115).
Shaded areas represent 95% confidence intervals. Tick marks indicate
censored observations. HR = 0.65 (95% CI: 0.48–0.88), P = 0.005
(log-rank test). Median survival: treatment A, 18.3 months
(95% CI: 15.2–21.4); treatment B, 12.1 months (95% CI: 9.8–14.4).

Good caption (multi-panel):

Figure 3. Immune cell infiltration predicts treatment response.

(a) Representative immunohistochemistry images showing CD8+ T cell
infiltration in responders (top) and non-responders (bottom).
Scale bars, 100 μm. (b) Quantification of CD8+ cells per mm²
(n = 45 responders, n = 38 non-responders). Box plots show median
and IQR; whiskers extend to 1.5× IQR. ***P < 0.001 (Mann-Whitney U test).
(c) ROC curve for CD8+ density as predictor of response. AUC = 0.82
(95% CI: 0.74–0.90). Dashed line indicates random classifier.

---

Python Implementation

Setup: Publication Style

import matplotlib.pyplot as plt
import matplotlib as mpl
import seaborn as sns
import numpy as np

# Publication-quality defaults
def set_publication_style():
    """Configure matplotlib for publication-quality figures."""
    plt.style.use('seaborn-v0_8-whitegrid')

    mpl.rcParams.update({
        # Figure
        'figure.figsize': (3.5, 3.5),  # Single column width
        'figure.dpi': 300,
        'figure.facecolor': 'white',
        'savefig.dpi': 300,
        'savefig.bbox': 'tight',
        'savefig.pad_inches': 0.05,

        # Font
        'font.family': 'sans-serif',
        'font.sans-serif': ['Arial', 'Helvetica'],
        'font.size': 9,
        'axes.labelsize': 9,
        'axes.titlesize': 10,
        'xtick.labelsize': 8,
        'ytick.labelsize': 8,
        'legend.fontsize': 8,

        # Lines
        'lines.linewidth': 1.5,
        'lines.markersize': 5,
        'axes.linewidth': 0.8,
        'xtick.major.width': 0.8,
        'ytick.major.width': 0.8,

        # Grid
        'axes.grid': True,
        'grid.alpha': 0.3,
        'grid.linewidth': 0.5,

        # Legend
        'legend.frameon': False,
        'legend.loc': 'best',

        # Math text
        'mathtext.fontset': 'stix',
        'mathtext.default': 'regular',

        # Font embedding (CRITICAL for PDF)
        'pdf.fonttype': 42,  # TrueType fonts (editable text)
        'ps.fonttype': 42,   # TrueType for PostScript
        'svg.fonttype': 'none',  # Keep text as text in SVG
    })

set_publication_style()

Creating Figures

# ALWAYS explicitly set figure size
fig, ax = plt.subplots(figsize=(3.5, 3.5))

# Use constrained_layout for multi-panel figures
fig, axes = plt.subplots(2, 2, figsize=(7, 7), constrained_layout=True)

# Add panel labels
for i, ax in enumerate(axes.flat):
    ax.text(-0.15, 1.05, chr(65 + i),  # A, B, C, D...
            transform=ax.transAxes,
            fontsize=12, fontweight='bold',
            va='bottom', ha='right')

Colorblind-Safe Palettes

# Viridis (sequential)
cmap = plt.cm.viridis
colors = cmap(np.linspace(0, 1, 5))

# Okabe-Ito (categorical)
okabe_ito = ['#E69F00', '#56B4E9', '#009E73', '#F0E442',
             '#0072B2', '#D55E00', '#CC79A7', '#999999']
sns.set_palette(okabe_ito)

# Or use seaborn's colorblind palette
sns.set_palette('colorblind')

Saving Figures

def save_publication_figure(fig, filename, formats=['svg', 'tiff', 'pdf']):
    """Save figure in multiple formats for publication."""
    for fmt in formats:
        fig.savefig(
            f'{filename}.{fmt}',
            format=fmt,
            dpi=300,
            bbox_inches='tight',
            pad_inches=0.05,
            facecolor='white',
            edgecolor='none'
        )
        print(f"Saved: {filename}.{fmt}")

# Usage
save_publication_figure(fig, 'analysis/figures/fig01_survival')

Common Plot Templates

# Kaplan-Meier survival plot
from lifelines import KaplanMeierFitter
from lifelines.plotting import add_at_risk_counts

fig, ax = plt.subplots(figsize=(4, 4))
kmf = KaplanMeierFitter()

for group, color in zip(['A', 'B'], ['#0072B2', '#D55E00']):
    mask = data['group'] == group
    kmf.fit(data.loc[mask, 'time'],
            data.loc[mask, 'event'],
            label=f'Group {group}')
    kmf.plot_survival_function(ax=ax, ci_show=True, color=color)

ax.set_xlabel('Time (months)')
ax.set_ylabel('Survival probability')
ax.set_ylim(0, 1.05)
add_at_risk_counts(kmf, ax=ax)
ax.legend(frameon=False, loc='lower left')

# Box plot with individual points
fig, ax = plt.subplots(figsize=(3.5, 4))

# Box plot
sns.boxplot(data=df, x='group', y='value', ax=ax,
            palette='colorblind', width=0.5,
            fliersize=0)  # Hide outlier points from boxplot

# Overlay individual points
sns.stripplot(data=df, x='group', y='value', ax=ax,
              color='black', alpha=0.5, size=4, jitter=0.2)

ax.set_xlabel('')
ax.set_ylabel('Measurement (units)')

# Heatmap with proper colorbar
fig, ax = plt.subplots(figsize=(5, 4))

im = ax.imshow(matrix, cmap='viridis', aspect='auto')
cbar = plt.colorbar(im, ax=ax, shrink=0.8)
cbar.set_label('Expression (z-score)', fontsize=9)

ax.set_xticks(range(len(x_labels)))
ax.set_xticklabels(x_labels, rotation=45, ha='right')
ax.set_yticks(range(len(y_labels)))
ax.set_yticklabels(y_labels)

Rasterization for Dense Plots

# Rasterize only the scatter points (many points), keep axes as vector
fig, ax = plt.subplots(figsize=(4, 4))

scatter = ax.scatter(x, y, alpha=0.5, s=10, rasterized=True)
# rasterized=True prevents huge file sizes while keeping text as vector

fig.savefig('figure.pdf', dpi=300)  # Rasterized elements at 300 DPI

---

R Implementation

Setup: Publication Style

library(ggplot2)
library(ggpubr)
library(viridis)
library(cowplot)

# Publication theme
theme_publication <- function(base_size = 9, base_family = "Arial") {
  theme_bw(base_size = base_size, base_family = base_family) +
    theme(
      # Panel
      panel.grid.minor = element_blank(),
      panel.grid.major = element_line(linewidth = 0.3, color = "grey90"),
      panel.border = element_rect(linewidth = 0.8),

      # Axes
      axis.text = element_text(size = rel(0.9), color = "black"),
      axis.title = element_text(size = rel(1.0)),
      axis.ticks = element_line(linewidth = 0.5),

      # Legend
      legend.background = element_blank(),
      legend.key = element_blank(),
      legend.title = element_text(size = rel(0.9)),
      legend.text = element_text(size = rel(0.8)),

      # Strip (facets)
      strip.background = element_rect(fill = "grey95", color = NA),
      strip.text = element_text(size = rel(0.9), face = "bold"),

      # Plot
      plot.title = element_text(size = rel(1.1), face = "bold", hjust = 0),
      plot.subtitle = element_text(size = rel(0.9), hjust = 0),
      plot.margin = margin(5, 5, 5, 5)
    )
}

# Set as default
theme_set(theme_publication())

Colorblind-Safe Palettes

# Okabe-Ito palette
okabe_ito <- c("#E69F00", "#56B4E9", "#009E73", "#F0E442",
               "#0072B2", "#D55E00", "#CC79A7", "#999999")

scale_color_okabe <- function(...) {
  scale_color_manual(values = okabe_ito, ...)
}

scale_fill_okabe <- function(...) {
  scale_fill_manual(values = okabe_ito, ...)
}

# Viridis for continuous
# scale_color_viridis_c() / scale_fill_viridis_c()

# Viridis for discrete
# scale_color_viridis_d() / scale_fill_viridis_d()

Creating Multi-Panel Figures

library(patchwork)  # or cowplot

# Create individual plots
p1 <- ggplot(data, aes(x, y)) +
  geom_point() +
  labs(tag = "A")

p2 <- ggplot(data, aes(group, value)) +
  geom_boxplot() +
  labs(tag = "B")

p3 <- ggplot(data, aes(time, survival, color = treatment)) +
  geom_line() +
  labs(tag = "C")

# Combine with patchwork
combined <- (p1 | p2) / p3 +
  plot_annotation(
    title = "Figure 1",
    theme = theme(plot.title = element_text(face = "bold"))
  )

# Or with cowplot
combined <- plot_grid(p1, p2, p3,
                      labels = "AUTO",  # A, B, C...
                      label_size = 12,
                      label_fontface = "bold",
                      ncol = 2)

Saving Figures

save_publication_figure <- function(plot, filename,
                                    width = 3.5, height = 3.5,
                                    units = "in", dpi = 300) {
  # Save as multiple formats
  ggsave(paste0(filename, ".pdf"), plot,
         width = width, height = height, units = units, dpi = dpi)
  ggsave(paste0(filename, ".tiff"), plot,
         width = width, height = height, units = units, dpi = dpi,
         compression = "lzw")
  ggsave(paste0(filename, ".svg"), plot,
         width = width, height = height, units = units)

  message("Saved: ", filename, " (.pdf, .tiff, .svg)")
}

# Usage
save_publication_figure(p, "analysis/figures/fig01_survival",
                        width = 4, height = 4)

Common Plot Templates

# Kaplan-Meier survival plot
library(survival)
library(survminer)

fit <- survfit(Surv(time, status) ~ group, data = data)

p <- ggsurvplot(fit,
  data = data,
  risk.table = TRUE,
  pval = TRUE,
  conf.int = TRUE,
  palette = c("#0072B2", "#D55E00"),
  xlab = "Time (months)",
  ylab = "Survival probability",
  legend.title = "Group",
  legend.labs = c("A", "B"),
  risk.table.height = 0.25,
  ggtheme = theme_publication()
)

# Add HR annotation
hr <- summary(coxph(Surv(time, status) ~ group, data = data))
hr_text <- sprintf("HR = %.2f (95%% CI: %.2f-%.2f)",
                   hr$conf.int[1], hr$conf.int[3], hr$conf.int[4])
p$plot <- p$plot +
  annotate("text", x = Inf, y = 0.9, label = hr_text,
           hjust = 1.1, size = 3)

# Box plot with statistics
library(ggpubr)

p <- ggboxplot(data, x = "group", y = "value",
               color = "group", palette = "colorblind",
               add = "jitter", add.params = list(size = 2, alpha = 0.5)) +
  stat_compare_means(method = "wilcox.test",
                     label = "p.signif",
                     comparisons = list(c("A", "B"), c("B", "C"))) +
  labs(x = "", y = "Measurement (units)") +
  theme(legend.position = "none")

# Volcano plot
library(ggrepel)

p <- ggplot(de_results, aes(x = log2FC, y = -log10(padj))) +
  geom_point(aes(color = significance), alpha = 0.6, size = 1.5) +
  scale_color_manual(values = c("Up" = "#D55E00",
                                "Down" = "#0072B2",
                                "NS" = "grey60")) +
  geom_hline(yintercept = -log10(0.05), linetype = "dashed", color = "grey40") +
  geom_vline(xintercept = c(-1, 1), linetype = "dashed", color = "grey40") +
  geom_text_repel(
    data = subset(de_results, padj < 0.001 & abs(log2FC) > 2),
    aes(label = gene),
    size = 2.5, max.overlaps = 20
  ) +
  labs(x = expression(log[2]~"fold change"),
       y = expression(-log[10]~"adjusted p-value"),
       color = "") +
  theme(legend.position = c(0.9, 0.9))

# Heatmap with ComplexHeatmap
library(ComplexHeatmap)
library(circlize)

col_fun <- colorRamp2(c(-2, 0, 2), c("#0072B2", "white", "#D55E00"))

ht <- Heatmap(matrix,
  name = "Z-score",
  col = col_fun,
  show_row_names = TRUE,
  show_column_names = TRUE,
  row_names_gp = gpar(fontsize = 8),
  column_names_gp = gpar(fontsize = 8),
  clustering_distance_rows = "euclidean",
  clustering_method_rows = "ward.D2",
  row_dend_width = unit(10, "mm"),
  column_dend_height = unit(10, "mm"),
  heatmap_legend_param = list(
    title_gp = gpar(fontsize = 9),
    labels_gp = gpar(fontsize = 8)
  )
)

# Save
pdf("analysis/figures/fig_heatmap.pdf", width = 6, height = 8)
draw(ht)
dev.off()

---

Standalone Figure Scripts

For consensus workflow integration, create standalone scripts for each figure:

Directory Structure

analysis/
├── plan.md
├── scripts/
│   ├── fig01_survival.py      # Generates fig01
│   ├── fig02_volcano.R        # Generates fig02
│   ├── fig03_heatmap.py       # Generates fig03
│   └── tab01_baseline.R       # Generates table 01
├── figures/
│   ├── fig01_survival.pdf
│   ├── fig01_survival.tiff
│   ├── fig02_volcano.pdf
│   └── ...
├── figure_manifest.csv        # Optional: tracking file
└── report.md

Figure Manifest (Optional)

For complex manuscripts with many figures, maintain a tracking file:

figure_id,panel,script,input_data,output_file,caption_file,status
Fig1,A-D,fig01_survival.py,clinical_data.csv,figures/fig01_survival.pdf,captions/fig01.txt,complete
Fig2,A-C,fig02_volcano.R,de_results.csv,figures/fig02_volcano.pdf,captions/fig02.txt,pending_review
Fig3,A-F,fig03_heatmap.py,expression_matrix.csv,figures/fig03_heatmap.pdf,captions/fig03.txt,in_progress

This enables:

• Tracking figure generation status
• Linking figures to source data for reproducibility
• Coordinating multi-author figure revisions

Script template (Python):

#!/usr/bin/env python3
"""
Figure 01: Overall survival by treatment group

Generates: analysis/figures/fig01_survival.{pdf,tiff,svg}
Data source: data/clinical_data.csv
"""

import matplotlib.pyplot as plt
import pandas as pd
from pathlib import Path

# Configuration
OUTPUT_DIR = Path("analysis/figures")
FIGURE_NAME = "fig01_survival"

def create_figure(data):
    """Create the survival figure."""
    fig, ax = plt.subplots(figsize=(4, 4))
    # ... plotting code ...
    return fig

def main():
    # Load data
    data = pd.read_csv("data/clinical_data.csv")

    # Create figure
    fig = create_figure(data)

    # Save in multiple formats
    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
    for fmt in ['pdf', 'tiff', 'svg']:
        fig.savefig(OUTPUT_DIR / f"{FIGURE_NAME}.{fmt}",
                    dpi=300, bbox_inches='tight')

    print(f"Saved: {OUTPUT_DIR / FIGURE_NAME}")
    plt.close(fig)

if __name__ == "__main__":
    main()

Script template (R):

#!/usr/bin/env Rscript
#' Figure 02: Volcano plot of differential expression
#'
#' Generates: analysis/figures/fig02_volcano.{pdf,tiff,svg}
#' Data source: results/de_analysis.csv

library(ggplot2)
library(ggrepel)

# Configuration
OUTPUT_DIR <- "analysis/figures"
FIGURE_NAME <- "fig02_volcano"

create_figure <- function(data) {
  p <- ggplot(data, aes(x = log2FC, y = -log10(padj))) +
    # ... plotting code ...
    theme_publication()
  return(p)
}

main <- function() {
  # Load data
  data <- read.csv("results/de_analysis.csv")

  # Create figure
  p <- create_figure(data)

  # Save in multiple formats
  dir.create(OUTPUT_DIR, showWarnings = FALSE, recursive = TRUE)

  ggsave(file.path(OUTPUT_DIR, paste0(FIGURE_NAME, ".pdf")), p,
         width = 4, height = 4, dpi = 300)
  ggsave(file.path(OUTPUT_DIR, paste0(FIGURE_NAME, ".tiff")), p,
         width = 4, height = 4, dpi = 300, compression = "lzw")
  ggsave(file.path(OUTPUT_DIR, paste0(FIGURE_NAME, ".svg")), p,
         width = 4, height = 4)

  message("Saved: ", file.path(OUTPUT_DIR, FIGURE_NAME))
}

main()

---

Validation Checklist

Before submitting figures through the consensus workflow, verify:

Technical Requirements

• Resolution: ≥300 DPI for images, ≥600 DPI for line art
• Format: TIFF or PDF for final submission
• Dimensions: Within journal limits
• Fonts: 8-12 pt, embedded, sans-serif preferred
• File size: Under journal limit (typically 10-50 MB)

Scientific Accuracy

• Data correctly represented (values match source)
• Appropriate plot type for data type
• Error bars correctly calculated and labeled
• Statistical tests appropriate and correctly reported
• Axes start at appropriate values (zero for bar charts)
• No misleading truncation or scaling

Accessibility

• Colorblind-safe palette used
• Figure interpretable in grayscale
• Shapes/patterns used in addition to colors
• Sufficient contrast between elements
• Font size readable at final print size

Caption Completeness

• Brief title describes figure content
• All panels described in order
• All abbreviations defined
• Error bar definition included (with n)
• Statistical test and exact P-values reported
• Color/symbol coding explained
• Scale bars defined (for images)
• Sample sizes stated

Consensus Integration

• Figure script is standalone and reproducible
• Output saved to analysis/figures/
• Source data traceable
• Caption written and ready for report.md
• Submitted for Codex + Gemini validation

---

Quick Reference: Journal Requirements

Journal	Column Width	DPI	Format	Panel Labels
Nature	88/180 mm	300-600	TIFF/EPS	lowercase (a,b,c)
Science	86/180 mm	300-600	EPS/TIFF	UPPERCASE (A,B,C)
Cell	85/174 mm	300-1000	TIFF/EPS	UPPERCASE (A,B,C)
NEJM	86/178 mm	300	TIFF/EPS	UPPERCASE (A,B,C)
PNAS	86/178 mm	300-600	TIFF/EPS	UPPERCASE italic
PLOS	83/173 mm	300	TIFF/EPS	UPPERCASE (A,B,C)
Elsevier	90/190 mm	300-1000	TIFF/EPS	varies

Always verify with target journal's current author guidelines.

---

Sources and References

Journal Guidelines

• Nature Figure Guidelines
• Nature Initial Submission (n/error bars)
• Science Author Instructions
• Cell Digital Art Guide (PDF)
• Wiley Figure Preparation
• PLOS ONE Figure Requirements
• JCB Figure Guidelines

Best Practices & Tutorials

• CalTech Figure Caption Handout
• Matplotlib for Papers (GitHub)
• Publication-Ready Figures Tutorial

Color & Accessibility

• Viridis R Package
• Colorblind-Friendly Figures Guidelines (NKI)
• Crameri et al. (2024) - Choosing Suitable Color Palettes

Statistics & Image Integrity

• HHS ORI Image Guidelines
• Cumming et al. (2007) - Error Bars in Experimental Biology
• Cromey (2010) - Avoiding Twisted Pixels

R Packages

• ggpubr Package
• survminer Package
• ComplexHeatmap