What is a Sankey plot (and when not to use it)?

A Sankey plot is a directed flow diagram in which the width of each link indicates the quantity. It is designed to show how a total amount splits, flows, and sometimes re-splits across successive stages of a process.

Sankey plots are especially effective when you want readers to understand where things go, how much is lost, and how categories change over steps.

When should you use a Sankey plot?

Use a Sankey plot when your data represents flow through stages and when conservation or loss matters.

Typical and effective use cases include:

✔ Bioinformatics and genomics

• Read fate across an RNA-seq or ATAC-seq pipeline
  (raw reads → QC pass → aligned → assigned → filtered)

• Variant consequences across annotation tiers
  (all variants → coding → missense → pathogenic)

✔ Sample and project tracking

• Sample triage
  (collected → passed QC → sequenced → analysed)

• Cohort filtering
  (initial cohort → exclusions → final analysis set)

✔ Resources and cost breakdowns

• Cost or time allocation across pipeline steps

• Compute usage split by analysis stage

If the question is “where did the data go?”, a Sankey plot is often the right answer.

When should you not use a Sankey plot?

Sankey plots are powerful—but very easy to misuse.

Avoid them in the following situations:

✖ Time series data

If you want to show trends over time, use:

• Line charts

• Area charts

Sankey plots do not encode time well.

✖ Too many small categories

If you have dozens of very thin flows (≈ more than 40–50):

• The diagram becomes unreadable

• Important patterns disappear into visual noise

In these cases, use:

• Bar charts

• Faceted plots

• Or group small categories into “Other”

✖ Cycles or bidirectional networks

Sankey plots assume a left-to-right, acyclic flow.

They are not suitable for:

• Feedback loops

• Iterative processes

• Bidirectional interactions

For those, use:

• Network diagrams

• Chord diagrams

• Graph layouts

Design best practices (the ones reviewers actually like)

1) Stages and ordering matter

• Arrange stages logically from left to right

• Keep the same order across related figures

• Never reorder stages just to “make it look nicer”

Consistency builds trust.

2) Use colour with restraint

• Assign consistent colours to the same category across all stages

• Avoid rainbow palettes

• Use muted backgrounds

• Group related categories using colour families

If colour encodes meaning, explain it.

3) Labels and tooltips

• Node labels: short, precise, unambiguous
  (“Aligned reads”, not “Aligned”)

• Link values: show both absolute counts and percentages
  e.g. “700,000 (11.2%)”

For interactive figures, use tooltips.
For static figures, annotate only the key flows.

4) Handle small categories carefully

Very small flows (<1–2% of the total):

• Add clutter

• Distract from the main story

Group them into “Other”, and explain this choice in the caption.

5) Always state units and totals

Make it explicit what the widths represent:

• Reads

• Samples

• Variants

• Compute hours

A subtitle helps enormously, for example:

Total reads analysed: N = 6.5 million

6) Accessibility matters

• Use colour-blind friendly palettes

• Ensure sufficient contrast

• Provide a text summary or table alongside the figure

Remember: figures should support your story, not replace it.

7) Avoid cycles and double-counting

Sankey diagrams must not loop.

If your process is iterative:

• Summarise each iteration separately, or

• Switch to a network-based visualisation

Never allow the same quantity to be counted twice without explicit explanation.