Accessibility and known limitations

ProtVista is being audited against WCAG 2.2 level AA. The automated and computed parts of the audit are done, including browser checks (in Chromium) of keyboard use, focus, zoom to 200%, a 320 px wide screen and wider text spacing. A pass with screen readers is still to come. The full results, with the evidence and tracking issue for each criterion, are in the accessibility audit report.

This page lists only limitations that have been verified in the code or by a test.

What works today

• Controls around the tracks have names and roles, and browser tests operate them from the keyboard: group expand and collapse (an element with the button role and aria-expanded), the ⓘ notes, the ⚠ error badges and their Retry button, and every Customize mode control.
• Focus. Customize controls, notes and error badges have their own focus ring; the group toggle and Retry show the browser’s default one. A browser check of every control found a visible focus indicator on all of them except the variant filter’s labels (below). Tab and Shift+Tab never get stuck: they move through and out of the viewer, the playground’s code editor and the structure panel’s controls and table. The 3D structure view itself has not been checked yet.
• Zoom and text spacing. At 200% zoom, and with WCAG’s increased line, letter and word spacing, no text in the controls, labels, ruler or sequence is cut off or overlaps, and every control stays in reach. The exception is the variant track’s amino-acid axis (below).
• Announcements. Loading, Customize-mode changes and visitor notes are announced to screen readers. A viewer that cannot load shows an alert and moves focus to it.
• The structure table under the 3D view can be used with the arrow keys.
• Automated checks. Every change runs axe, keyboard, focus, zoom and reflow tests in a real browser on the surfaces above.

Known limitations

The tracks themselves

• No text alternative. Features, variants and scores are drawn as pictures (canvas or SVG) by the Nightingale components. A screen reader cannot read what a track shows.
• Pointer only. A feature’s details open only when you click or tap it; there is no keyboard path to a feature or its tooltip.
• Zooming needs a drag on the navigation bar at the top, or Ctrl+scroll over a track. There are no zoom buttons.

Colour

• Variants. Each variant is a dot whose colour gives its class: red pathogenic, green benign, blue predicted, teal uncertain. Every dot is the same circle. With red–green colour blindness, pathogenic and benign dots differ only in lightness, so it is hard to tell which is which. Benign, predicted and uncertain dots have less than 3:1 contrast with the white background.
• AlphaFold confidence. Three of the four colours (orange, yellow, light blue) have less than 3:1 contrast with white.
• AlphaMissense. The light middle of the scale (ambiguous scores) is close to white, and a residue with no score is drawn as white, so the two are hard to tell apart.
• AlphaMissense average track. Three of its eight score ranges are drawn in the wrong colour: the most benign residues (average score below 0.1132) appear in the light salmon used for likely-pathogenic scores, residues from 0.2264 to 0.3395 appear light peach, and 0.7264–0.8632 appears in the darkest red. The heatmap and the 3D model are not affected.
• The AlphaMissense legend in the structure panel uses darker colours than the 3D model shows.
• Collapsed groups. A collapsed group draws all its feature types in one track, and some default colours are almost the same: for example metal binding and DNA binding in Sites, and motif and zinc finger in Domains for people with protanopia. As drawn, 14 of the 45 default feature colours have less than 3:1 contrast with white.
• InterPro domains get their colour from a hash of the entry, so two domains on one protein can look alike; nearly half the possible colours have less than 3:1 contrast with white.
• Count graphs. The grey line of the variant and RNA editing count graphs has 2.35:1 contrast with white. In the variant graph only colour tells the grey line (all variants) from the red one (disease-causing).

Other parts of the page

• Group headers with a ⚠ badge or a link. A screen reader announces the header as a button that contains other buttons or links.
• Datatable filters. Each filter is named “Filter by …” for screen readers but is not linked to its column heading, and the number of matching rows is not announced when a filter changes (#312).
• Structure panel. Every download link has the same name, the last column has no heading, and loading and “no structures” are not announced (#311).
• Variant filter checkboxes. Each one has an extra Tab stop on its label that shows no focus ring.
• Structure table filters. Changing a filter moves keyboard focus from the filter to the rows of the table. When the table scrolls up, the row that has focus can be hidden under the table’s header.
• Notes and badges close as soon as you press anywhere else, before you release.

Narrow screens and zoom

• At 320 px wide (for example 400% zoom on a laptop), the label column beside the tracks is only 64 px wide. The Customize button, the variant filter and long words in labels run past it and are partly covered by the tracks.
• Variant track axis. The amino-acid letters on the variant track’s left axis are drawn so close to its edge that the widest (such as M and W) lose a sliver of their left side, at any zoom; with wider letter spacing more letters do. Whether they all stay readable has not been checked by eye yet.
• Zooming in after the viewer has drawn leaves its tracks at their old width, so they run past the right edge until the viewer is drawn again (reload the page, or press Run in the playground).

The 3D structure view

• Single-letter keys. While the pointer is over the 3D view and nothing else has focus, I starts or stops spinning the model, O rocking it, and T resets the view. ProtVista has no setting to turn them off.

The playground

• Accession field. If you enter an accession that UniProt does not have and press Tab, focus moves to the error message in the preview instead of the next control.
• Some validation messages are the schema validator’s own wording, for example “must be equal to one of the allowed values”, and do not say which values are allowed.
• The divider between the editor and the preview is 8 px wide. It can also be moved with the arrow keys once it has focus.

Workarounds

• Read a variant’s class without colour: in the variants filter, untick every class but one; only that class stays on the track. The tooltip also names the clinical significance.
• Avoid the 3D view’s letter keys: move the pointer off the 3D view before typing.
• Read AlphaFold and AlphaMissense values: click a residue; the tooltip gives the pLDDT or pathogenicity score as a number. For AlphaMissense, prefer the heatmap or the score to the average track’s colour.
• Tell feature types apart: expand the group. Each type then has its own track with a text label.
• Simulate colour-vision deficiency in Chrome or Edge under DevTools › Rendering › “Emulate vision deficiencies” to see what your readers see.
• Change the colours of your own tracks with rendering.color or a custom colour scale; see Theme the viewer and Escape hatches.

Report a problem

Open an issue on GitHub and say which page, browser and assistive technology you used.

Licensed under CC BY 4.0.