Using the Legend

The legend panel lets you filter, customize, and explore protein annotations. Most annotations are categorical, but scalar numeric annotations such as length are handled a little differently.

Show And Hide Values

Click any legend row to toggle its visibility:

• Click once to hide that value.
• Click again to show it.
• Double-click to isolate that value and hide the rest.
• If only one value remains visible, clicking it restores the full set.

Reordering And Draw Order

Drag labels up or down to change draw order in the scatterplot. Items near the top are drawn on top of items below them.

The drag handle stays visible in each legend row so manual ordering is easy to discover. Dragging from any sort mode switches that annotation to Manual order.

The "Other" Group

Categorical annotations can group less frequent values into Other once the visible list exceeds the current legend cap.

• Click Other to inspect the grouped values.
• Extract values from Other to show them separately.
• Drag values into Other to regroup them.

Numeric annotations do not use Other. They are binned directly from the raw numeric values.

Transferred Annotations (EAT)

When the selected annotation carries values borrowed from nearby annotated proteins, a Predicted (transferred) section appears at the very top of the legend panel, above the separation strips. It holds a Show checkbox that adds the borrowed values to the plot and the legend rows, or removes them so those proteins fall back to N/A; a reliability filter — a mode dropdown (Hide below, Hide above, Keep between) with a slider, plus a second slider for the band — that hides predictions falling outside the chosen range; and a count of Observed against Predicted by EAT proteins. Curated values are never hidden by it in any mode, and the section is absent for annotations that have no transferred values.

See Transferred Annotations (EAT) for where the borrowed values come from and how to read them in the plot.

Separation Score Strips

When the dataset was prepared with statistics and the selected annotation was scored in the current projection, two strips appear near the top of the legend panel, above the category list: one for Silhouette, one for Davies–Bouldin. Each plots one dot per category, in that category's legend color, so you can see which values sit apart from the rest and which overlap.

• Hovering a dot highlights the matching legend row, and hovering a row highlights its dot.
• Clicking a dot toggles that category, exactly like clicking its legend row.
• While a filter or isolation narrows the view, the strips are replaced by the note Separation scores are hidden while the view is filtered.

See Separation Scores for what the numbers mean and how they are computed.

Settings

Click the cog icon in the top-right corner of the legend for advanced options.

Shared Settings

Setting	What it does
Max legend items	Categorical annotations: maximum number of visible values before grouping into Other. Numeric annotations: target maximum number of bins.
Shape size / Point size	Adjusts marker size in the scatterplot. Numeric annotations use the label Point size because numeric legends always render circles.
Show duplicate counts	Shows duplicate count badges and spreads overlapping points when you expand them.
Color palette	Picks the active palette for the selected annotation.

Sorting

Categorical annotations support three sort modes:

• By category size
• Alphabetical
• Manual order

A fourth mode, By separation, appears only when the selected annotation carries per-category separation scores for the current projection, and orders categories by how cleanly each one separates. It is display-only: it reorders the list without changing which categories fall into Other, and it keeps working while the view is filtered.

Numeric annotations support two sort modes:

• By numeric value
• Manual order

The arrow button next to the cog reverses the current sort direction:

• Numeric By numeric value: low-to-high vs high-to-low
• Numeric Manual order: reverses the current manual order
• Categorical By separation: best-separating first vs worst-separating first
• Categorical: preserves the legacy reverse behavior

Numeric Bin Distribution

When the selected annotation is numeric and a numeric gradient palette is active, an extra Bin distribution control appears:

• Linear
• Quantile
• Logarithmic

Quantile is the default distribution, automatically falling back to Linear when the number of distinct values is small enough that quantile edges would collapse. Logarithmic is only available when all non-null values are strictly positive; otherwise ProtSpace falls back to linear binning.

Reverse Gradient Direction

Numeric annotations also expose a Reverse gradient direction toggle in the settings dialog.

• Off: lower numeric values use the low end of the selected gradient and higher values use the high end.
• On: the gradient is flipped so higher numeric values receive the colors that would normally be used for lower values, and vice versa.

This setting is saved with the rest of the numeric legend settings and is restored on reload or bundle import.

Numeric Annotations

Scalar numeric annotations are stored as raw numbers and binned in the browser from the current numeric settings.

Numeric Labels

Numeric legend labels are display summaries for humans:

• They use min - max formatting instead of comparison syntax like <.
• Integer-only data stays integer-only in the legend.
• Decimal labels only appear when the source data actually needs decimal precision.
• Single-value bins are shown as a single number.

The exact runtime bin edges are still used for membership. The displayed label is a summary of the values observed in that bin.

Why Fewer Bins Can Appear Than The Max

For numeric annotations, Max legend items is a target cap, not a hard guarantee. You can request 10 bins and still see fewer when the data does not support all of them.

Common reasons:

• A linear or logarithmic interval ends up empty, so it is dropped from the realized legend.
• Quantile cut points collapse because many proteins share the same value.
• A constant numeric column produces exactly one bin.
• An all-null numeric column produces zero bins.
• Very narrow ranges can collapse after precision/topology normalization.

This is why you might see 10 requested bins become 9, and 9 become 8, without anything being wrong.

Predicted Annotations

Some annotations are computational predictions rather than experimentally curated records. When one is selected, the legend header shows a ⚡ Predicted badge next to the annotation title, followed by the note Computationally predicted, not experimentally curated. The same ⚡ marker appears next to the annotation in the control bar's annotation dropdown.

The badge marks columns produced by a machine-learning model, a de-novo sequence-topology predictor, or predicted 3D structure. Reference signature matches such as Pfam, CATH-Gene3D and SUPERFAMILY, and curated or factual UniProt and Taxonomy fields, are not badged. Any unknown column whose name starts with predicted_ is treated as a prediction.

Which built-in columns carry the badge is listed per annotation in the Annotations reference, which is generated from the same table the app reads.

The ⚡ badge is a property of the whole column and is decided from the annotation's name, not from its values, so it is not the same thing as the EAT badge, which marks an otherwise curated column in which some individual proteins carry a transferred value.

Colors, Gradients, And Shapes

Categorical Palettes

Categorical annotations use discrete palettes. ProtSpace keeps the established categorical options for contrast and colorblind safety:

• Kelly's Colors
• Okabe-Ito
• Tol Bright
• Set2
• Dark2
• Tableau 10

Colors stay attached to categories when you reorder them.

Numeric Gradients

Gradients are only shown for numeric annotations. ProtSpace currently ships five curated sequential gradients:

• Batlow: scientific-publication oriented sequential ramp from Fabio Crameri's Scientific Colour Maps (default)
• Cividis: colorblind-friendly sequential gradient
• Viridis: balanced perceptually uniform sequential gradient
• Inferno: high-contrast dark-to-bright ramp
• Plasma: vivid exploratory sequential ramp

These were chosen because they are either perceptually uniform, publication-friendly, accessibility-aware, or visually distinctive for ordered data. Rainbow-style maps are intentionally excluded because they are a poor default for numeric interpretation.

Numeric bin colors are derived from the gradient and the selected distribution. They are not manually edited per bin.

If an imported bundle or saved browser state references an unsupported numeric gradient ID, ProtSpace falls back to batlow.

Special Categories

Category	Color	Shape
Other	#999999	Circle
N/A	#DDDDDD	Circle

Shapes

Every category renders as a circle by default. To assign a different shape (square, diamond, plus, triangle-up, triangle-down) to an individual category, click the category's color/shape swatch in the legend, the popover has a Shape section underneath the color picker.

Numeric and multi-label annotations always render as circles; per-category shape assignment is not available for those.

Saved Settings

Legend settings are saved per dataset and per annotation in the browser.

• Saved examples: visibility, palette, ordering, numeric binning settings, and duplicate-count preferences
• Numeric binning settings such as palette, gradient direction, strategy, and target bin count are restored on reload/import
• Numeric hidden values and manual order are only restored when the current numeric topology still matches the saved one
• Use Reset in the settings dialog to clear saved preferences for the selected annotation

Styling From The Python CLI

protspace style can pre-set legend styling inside a bundle, but its keys are categorical only. For a numeric annotation the web app reinterprets or ignores most of them, because numeric legend entries are bin IDs such as num:quantile:150:200 rather than per-value keys.

Bin distribution and Reverse gradient direction are UI-only settings and cannot be authored from the CLI.

See How CLI keys are reinterpreted for numeric columns for the per-key reference.

Multi-Label Annotations

When proteins have multiple values, such as multiple EC numbers:

• Points display as pie charts
• Each slice represents one value
• All unique values appear in the legend

Next Steps

• Separation Scores - the strips at the top of the legend panel, and how well each category separates
• Control Bar Features - projections, filters, export, and import
• Viewing 3D Structures - AlphaFold integration