import pandas as pd
import numpy as np
import plotly.express as pxParallel Categories (Python)
A parallel-categories (alluvial) plot shows how every individual observation flows across several categorical dimensions at once — the per-patient counterpart of a Sankey, which only shows aggregated volumes. Combinatorial trial data (genotype × treatment × response) is its home turf. plotly.express.parallel_categories builds it from a tidy frame, with counts encoded by ribbon thickness.
Example
Setup
- System Requirements: Cross-platform (Linux/MacOS/Windows)
- Programming Language: Python
- Dependencies:
pandas,numpy,plotly
Data Preparation
We simulate one row per patient (n = 420) across four categorical dimensions: biomarker status, treatment arm, response, and grade-3 toxicity — plus a colour signal (response) that carries through the ribbons.
np.random.seed(231)
n = 420
tri = pd.DataFrame({
'Biomarker': np.random.choice(['Positive', 'Negative'], n, p=[0.55, 0.45]),
'Arm': np.random.choice(['Standard', 'Intensified'], n, p=[0.6, 0.4]),
})
resp_p = np.where(
(tri.Biomarker == 'Positive') & (tri.Arm == 'Intensified'), 0.75,
np.where(tri.Arm == 'Intensified', 0.55, 0.35))
tri['Response'] = np.array(['Responder', 'Non-responder'])[
(np.random.random(n) < resp_p).astype(int)]
tri['Toxicity'] = np.where(
(tri.Arm == 'Intensified') & (tri.Response == 'Responder'),
np.random.choice(['Grade < 3', 'Grade 3'], n, p=[0.6, 0.4]),
np.random.choice(['Grade < 3', 'Grade 3'], n, p=[0.85, 0.15]))
# plotly 6 parallel_categories maps colour continuously: encode response 0/1
tri['ResponseCode'] = (tri.Response == 'Responder').astype(int)Visualization
All Four Dimensions, Coloured by Response
Ribbon colour = response outcome, so the question “who responds?” is answered by following one colour across all axes; hovering any ribbon identifies the exact patient segment.
fig = px.parallel_categories(
tri, dimensions=['Biomarker', 'Arm', 'Response', 'Toxicity'],
color='ResponseCode',
color_continuous_scale=['#e63946', '#66c2a5'],
labels={'Biomarker': 'Biomarker', 'Arm': 'Arm',
'ResponseCode': 'Response (0 = non, 1 = res)',
'Response': 'Response', 'Toxicity': 'Toxicity'})
fig.update_layout(width=880, height=440, font_size=11)
fig.show()Counts on the Axes
Setting counts via line.color counts is plotly-internal; the simpler editorial alternative is sorting the frame so large ribbons render last, on top — and stating n in the title.
order = (tri.groupby(['Biomarker', 'Arm', 'Response', 'Toxicity'])
.size().sort_values().index)
tri_sorted = pd.concat(
[tri[(tri.Biomarker == b) & (tri.Arm == a) &
(tri.Response == r) & (tri.Toxicity == t)]
for b, a, r, t in order])
fig = px.parallel_categories(
tri_sorted, dimensions=['Biomarker', 'Arm', 'Response', 'Toxicity'],
color='ResponseCode', color_continuous_scale=['#e63946', '#66c2a5'])
fig.update_layout(title_text=f'Patient flows (n = {n}), dominant paths on top',
width=880, height=440, font_size=11)
fig.show()References
- Plotly Technologies Inc. Parallel categories diagram in Python. https://plotly.com/python/parallel-categories-diagram/
- Kosara, R., & Bendix, F. (2006). Parallel Sets: Interactive Exploration and Visual Analysis of Categorical Data. IEEE Transactions on Visualization and Computer Graphics, 12(4), 558-568.