Bio Data Visualization Color Palettes

Version Compatibility

Reference examples tested with: viridis 0.6+, RColorBrewer 1.1+, scico 1.5+ (Crameri colormaps in R), khroma 1.12+ (Tol/Crameri palettes in R), matplotlib 3.8+, colorcet 3.0+, ggsci 3.0+, colorspace 2.1+.

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

• Python: pip show <package> then help(module.function) to check signatures
• R: packageVersion('<pkg>') then ?function_name to verify parameters

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

Color Palettes for Scientific Visualization

"Pick a color palette" -> Choose a colormap that (a) is perceptually uniform along the relevant data axis, (b) remains interpretable under common color-vision deficiencies, (c) prints correctly to grayscale, and (d) matches the data type — sequential, diverging, cyclic, or qualitative.

• R: viridis::viridis, scico::scale_color_scico, khroma::color, RColorBrewer::brewer.pal
• Python: matplotlib.colormaps, colorcet, seaborn.color_palette, cmcrameri.cm

The Three Modern Standards

1. Perceptual uniformity -- equal data steps produce equal perceived color steps. viridis (van der Walt 2015), cividis (Nuñez 2018), and the Crameri family (batlow, roma, vik) are designed for this. Jet, rainbow, and red->green are not.

2. Color vision deficiency safety -- ~6% of males have deuteranopia / protanopia (red-green deficiency). cividis was explicitly designed to be near-identical under normal and CVD viewing (Nuñez 2018 PLOS ONE 13:e0199239). The Okabe-Ito 8-color qualitative palette (popularized in Wong 2011 Nat Methods 8:441) is the CVD-safe categorical default.

3. Grayscale monotonicity -- a perceptually-uniform sequential colormap has monotonically increasing luminance. Convert the figure to grayscale; if the order is still readable, the colormap is luminance-monotonic. This is the single most actionable test.

Palette Type by Data Type

The Crameri Scientific Colormaps

Crameri 2020 Nat Commun 11:5444 documented the prevalence of misleading palettes (rainbow, red-green) across published science and released a family of perceptually-uniform CVD-safe colormaps via Zenodo (doi:10.5281/zenodo.8409685). Key entries:

library(scico)
# Sequential
ggplot(df, aes(x, y, fill = value)) + geom_tile() +
    scale_fill_scico(palette = 'batlow')
# Diverging
ggplot(df, aes(x, y, fill = lfc)) + geom_tile() +
    scale_fill_scico(palette = 'vik', midpoint = 0)

from cmcrameri import cm
import matplotlib.pyplot as plt
plt.imshow(data, cmap=cm.batlow)         # sequential
plt.imshow(data, cmap=cm.vik, vmin=-vmax, vmax=vmax)   # diverging, symmetric

viridis Family (matplotlib default since 3.0)

library(viridis)
scale_color_viridis_c(option = 'viridis')   # default: dark blue -> yellow
scale_color_viridis_c(option = 'magma')     # black -> red -> yellow
scale_color_viridis_c(option = 'inferno')   # black -> purple -> yellow
scale_color_viridis_c(option = 'plasma')    # purple -> pink -> yellow
scale_color_viridis_c(option = 'cividis')   # CVD-optimized
scale_color_viridis_c(option = 'turbo')     # jet-like but perceptually uniform

plt.imshow(data, cmap='viridis')   # 'magma', 'inferno', 'plasma', 'cividis', 'turbo'

cividis is the only viridis-family colormap optimized for CVD. Use it for any figure intended to remain interpretable under deuteranopia/protanopia.

Okabe-Ito Categorical Palette (Wong 2011)

The 8-color CVD-safe categorical palette. Memorize the hexes:

okabe_ito <- c(
    '#E69F00',  # orange
    '#56B4E9',  # sky blue
    '#009E73',  # bluish green
    '#F0E442',  # yellow
    '#0072B2',  # blue
    '#D55E00',  # vermilion
    '#CC79A7',  # reddish purple
    '#000000'   # black
)
scale_color_manual(values = okabe_ito)

Available as palette.colors(8, 'Okabe-Ito') in R 4.0+, scale_color_manual(values = palette.colors(8, 'Okabe-Ito')). In matplotlib, colorblind style or manual hex list.

For DE plots, the canonical assignment is Up = #D55E00 (vermilion), Down = #0072B2 (blue), NS = #999999 (grey).

ColorBrewer (Harrower & Brewer 2003)

library(RColorBrewer)
display.brewer.all()                    # interactive palette browser
display.brewer.all(colorblindFriendly = TRUE)   # CVD-safe subset only
brewer.pal(n = 8, name = 'Dark2')       # qualitative
brewer.pal(n = 9, name = 'YlOrRd')      # sequential
brewer.pal(n = 11, name = 'RdBu')       # diverging

ColorBrewer's CVD-safe sequential and diverging palettes are publication-defaults. For qualitative beyond 8 colors, switch to Tol/Polychrome — ColorBrewer qualitative tops out at 12 (Set3).

Scientific Journal Brand Palettes

library(ggsci)
scale_color_npg()       # Nature Publishing Group
scale_color_aaas()      # Science (AAAS)
scale_color_lancet()    # Lancet
scale_color_jama()      # JAMA
scale_color_jco()       # JCO
scale_color_nejm()      # NEJM

These are CVD-imperfect — use journal palettes for stylistic compliance, not for accessibility. Verify by colorblindness simulation (below).

CVD Simulation -- The Mandatory Check

library(colorspace)
# Simulate deuteranopia / protanopia on a palette
cvd_emulator(palette, type = 'deutan')
cvd_emulator(palette, type = 'protan')
cvd_emulator(palette, type = 'tritan')

# Visual side-by-side
demoplot(palette, type = 'heatmap')

# colorspacious provides CVD simulation
from colorspacious import cspace_converter
# or use a CVD-safe palette by construction (cividis, Okabe-Ito, Crameri)

If a palette is unreadable under deutan simulation, do not use it for accessible figures. Period.

Grayscale Monotonicity Test

library(scales)
show_col(viridis(10))           # full color
show_col(grey(seq(0, 1, length = 10)))   # equivalent grayscale gradient

In practice: save the figure as PNG, open in an image editor, desaturate. If the data order is still readable, the colormap is luminance-monotonic. If it shows arbitrary "rings" or "bands," the colormap is non-monotonic — fix before submitting.

Rainbow / jet fails this test catastrophically. viridis and cividis pass.

Diverging Palette Setup (LFC, z-score)

library(circlize)
col_fun <- colorRamp2(c(-2, 0, 2), c('#0072B2', 'white', '#D55E00'))
# Symmetric around 0; ALWAYS use symmetric bounds for signed data

import matplotlib.pyplot as plt
plt.imshow(data, cmap='RdBu_r', vmin=-2, vmax=2)   # symmetric
# do NOT use vmin=data.min(), vmax=data.max() for diverging data

The most common diverging-palette error is asymmetric bounds (vmin=min, vmax=max) which mis-aligns zero with the white midpoint.

Custom Palette Construction

# Discrete categorical
my_palette <- c('Control' = '#0072B2', 'Treatment' = '#D55E00', 'Vehicle' = '#009E73')
scale_color_manual(values = my_palette)

# Continuous gradient between custom colors
colorRampPalette(c('#0072B2', 'white', '#D55E00'))(100)

from matplotlib.colors import LinearSegmentedColormap
cmap = LinearSegmentedColormap.from_list('cvd_div', ['#0072B2', '#FFFFFF', '#D55E00'])

When building a custom diverging palette: pick endpoints with similar luminance (so neither side dominates), pass through pure white at the midpoint (NOT light gray), and verify with the grayscale test.

Common Failure Modes

Asymmetric bounds on diverging data

Trigger: vmin=data.min(), vmax=data.max() on signed data with skewed distribution.

Mechanism: Zero no longer maps to the midpoint (white) of the diverging palette.

Symptom: Half the cells visually look "below zero" but are actually positive; reviewer confusion.

Fix: vmax = max(abs(data.min()), abs(data.max())); then vmin = -vmax. Or pre-clip data to a fixed range.

Categorical palette with too many colors

Trigger: 15+ groups all on one colormap.

Mechanism: Human color discrimination saturates around 8-10 distinct hues.

Symptom: Groups look identical; legend has no information value.

Fix: Facet by category, or aggregate small groups into "Other," or use a categorical+marker-shape combination.