Changelog

Version History

0.7.5 2026-09-03

Coexpression panel in the expression table: correlate a searched gene (or rank the whole result) against the atlas metacells of the currently visible/filtered samples
Computed on the coexpression service as an asynchronous celery task (the browser polls it), so the panel stays responsive on large results
Per-gene Pearson correlation against the searched gene or a sequence-query's per-sample intensities; multi-artifact searches fall back to marker-specificity scores (enterprise accounts)

0.7.0 2026-08-31

Browse the whole database before any search: the empty page now shows the full sample × metadata table (cell types, organs, diseases, …) with real per-sample cell counts, powered by the cached sparse per-(sample, cell type) count table — one lightweight transfer per session
Sample-level significance for the enrichment engine: instead of cell-level fold enrichment, each active column is tested with a per-column Kruskal-Wallis omnibus and then each value against the complement within its own column (Mann-Whitney U + study-by-sample-blocked permutation, BH-FDR), automatically rolled up across every ontology level — a term is found at 'lineage'/'cancer-specific' depth even when its subtype is not significant
Significance shown directly in the table: per-column badges (↑/↓ + n.s./*/**/*** star notation) inside each metadata cell, green/red fold tinting, and the significance is always computed over the unfiltered population so filtering never changes the p/q values
Reworked enrichment panel: per-column scrolling subpanels, significance-star items (no clutter), metric selector (% positive cells / mean expression / cell count), and a publication-ready export (PNG / SVG / plot-data JSON plus a full significance CSV) with an off-screen renderer so exports work straight from the panel
Compare groups panel: compare already-searched genes across saved groups in the browser — searchable multi-select group dropdown (10 most recent), per-sample metric vectors (mean expression / % positive / cell count, or the coexpression score), pairwise Mann-Whitney U with study-by-sample-blocked permutation and BH-FDR shown as significance stars, per-group violin with significant pairs annotated, and publication export (high-res PNG / SVG / raw-data CSV / plot-spec JSON)
Both side panels are resizable in-flow panels with pressed toolbar toggles; the enrichment panel auto-opens on wide screens; significance computation runs in the background (with a busy indicator) using shared-rank statistics so the table stays instant even at database scale
Filter-bar suggestions are relevance-ranked (exact > prefix > word-prefix > substring) across more columns, and the pre-search/metadata table is fully filterable and searchable on every sample column
Under-the-hood performance: the pre-search table renders instantly from cached counts, the metadata joins are dictionary-encoded, and the significance engine was reworked several times (shared global ranking, value-centric omnibus) to make p-values effectively instant

0.6.8 2026-08-18

Search results now carry per-sample cell counts in the table response itself, so the Cells column and the min-cells-per-sample filter work without re-aggregating expression data in the browser
The table now loads from a filter-independent per-sample intermediate: the metadata server ships per-(key, sample) cell counts and per-group statistics once, and the browser re-derives every statistic and applies filters locally with the same semantics — filtering no longer re-aggregates server-side, cutting the metadata service's CPU load on keystroke-driven filter changes
The per-sample intermediate is ~3x smaller: per-key min/max/cells replace per-sample extrema, sample metadata is dictionary-encoded per field, and sample ids travel as integers
The full sample-metadata blob is no longer fetched eagerly after every search — metadata loads lazily (detail drawer, browse view, column/date interactions) and can be restricted to specific fields, cutting the largest per-search transfer (~32MB) from the hot path
Fix: the browse view refreshes in place when the field-limited sample metadata arrives after it opens, so cards and their publication enrichment appear without leaving and re-entering the view
Fix: ExpressionTable, CoverageViewer and EnrichmentPanel are now exposed as window globals, restoring the mobile FASTA-drop/presets/save-group/clear-filters shortcuts and the Send-to-enrichment buttons (previously the guards never matched, silently disabling them)
Fix: the per_sample table decoder sized the row loop from the keyCols spec (the active-column name list) instead of the metadata row count, so only a handful of rows rendered — the full result set renders again
Fix: the filter-bar suggestions now respect the filters already applied to the table — the server resolves filter columns to real sample-index attributes (and drops per-cell columns like cell_type) before querying the search index, and resolves the fallback polars path to actual sample columns, so typing e.g. “female” with a cell_type filter applied returns suggestions again
The per-sample intermediate is now cached client-side per view (search artifacts + active columns + display mode + ontology levels): filter and search keystrokes apply locally from the cached payload, so the full table response is no longer re-downloaded on every keystroke
Fix: the per-sample detail drawer fetches its metadata itself, for every row and field-limited to what the drawer shows — the “unknown study” placeholder no longer appears until the browse view loads the sample metadata, and opening a drawer no longer downloads the full per-sample record for all its samples at once
The browse view opens with a single compact, dictionary-encoded sample-metadata request (unique values per field instead of per-sample records) covering exactly the fields the browse groupings display, so it no longer downloads the full field slice in row-ordered batches (the progressive batch loader remains as a fallback for older metadata servers); the per-cell expression-matrix export modal fetches only sample_uuid/uuid plus the metadata columns it displays
The column picker now lists every metadata column the metadata server knows (via /api/metadata/columns) — previously it showed only the small filterable subset plus whatever columns the browser had derived from fully-loaded sample metadata, so once sample metadata became lazy the picker collapsed to a few preselected columns
Fix: opening the sample drawer from a browse card lazily fetches the missing per-sample metadata (field-limited, same as the table drawer) instead of showing unknown values for samples that had not been scrolled into view yet
Search now commits only on Enter in every search bar: while typing, only suggestions load (server-side); Enter applies the highlighted suggestion or, when none is selected, runs the typed text as a free-text “anywhere” query
Fix: the column picker retries loading the full metadata-column list after a search if the initial page-load fetch was empty, and the browse metadata load sources its sample ids from the table rows (available as soon as results render) instead of requiring the raw expression data to have been fetched in-session
Fix: the compact browse metadata load no longer marks the whole sample-metadata store as loaded — it is a partial (browse-fields-only) slice, and other on-demand paths (per-column sample fetches when toggling a column, date counts, detail aggregation) depend on that flag being false until a real fetch happens; the coverage page's column picker also unions the full metadata-column list
Fix: the sample drawer no longer treats partial per-sample records (e.g. a single field merged when toggling a column in the picker) as fully loaded — a drawer-completeness check makes it fetch the complete field-limited record, so the '+ details' panel shows all sample metadata again after column changes
Fix: the browse publication view no longer shows the same paper as several cards or filters on the wrong one — cards whose titles normalize to the same text are merged; title variants that share a DOI are merged when the titles are highly similar; and title variants that share no DOI are merged when they belong to the same study and their titles are near-identical (a shared DOI alone never merges distinct papers, and similarity alone never merges distinct papers either, so several studies can no longer collapse into a single card) — the union carries every title variant so the card title always matches what its filter applies and the Filter button selects all of its samples

0.6.7 2026-08-18

Fix: the columnar table response is decoded from the nested rows payload, so search results render again after the columnar layout change

0.6.6 2026-08-18

The composite row key is no longer sent in the columnar table response — it is rebuilt browser-side from the metadata columns, further reducing transfer (~49% smaller than the original JSON)

0.6.5 2026-08-18

Table responses are leaner: the search-context columnar format now dictionary-encodes sample ids, stores per-artifact stats sparsely, drops duplicated per-artifact metadata and reduces numeric precision to 6 significant digits — roughly 45% less data transferred with identical displayed results

0.6.4 2026-08-17

Ontology filtering is now synonym- and label-aware: typing e.g. 'colorectal cancer' finds the node 'Colorectal carcinoma' (and its synonyms), raw metadata values resolve to their ontology node, and selected terms are expanded to their id/label/synonyms when sent to the server so filtered rows are no longer dropped
Selecting a term in the ontology filter panel or its search box still selects the whole descendant subtree ('clear filters' and deselecting removes them)

0.6.3 2026-08-17

Selecting a term in the ontology filter panel (or in its search box) now also selects all of its downstream descendant terms, and selecting again removes them
Refreshing the page no longer keeps the previous session's loaded results: the explorer starts from a blank state, and past searches can be reloaded from the search-history panel

0.6.2 2026-08-17

Removing an individual search result (the X button) now removes it from the table immediately instead of waiting for a separate table refresh
Each search in the artifact panel gains a show/hide-all button that toggles the visibility of all of that search's sub-results at once

0.6.1 2026-08-17

Refreshing the page no longer re-fetches any persisted job results: the job/artifact list is restored, and results load individually on demand when you click an artifact in the panel (reduces load on the Explorer/backend from repeated page loads)

0.6 2026-08-17

Forced refresh of stale frontends: the running build is fingerprinted (FE_VERSION) and users on an older build are prompted to reload when they next load a page, and are blocked at search submission (with a fail-safe 'Continue with current version' option) until they do
Search jobs that return no genes now surface a clear 'no results' state instead of an empty table: a zero-cell placeholder artifact engages the amber no-results banner, which now also carries actionable suggestions (search unstranded / relax k-mer limits / remove filters)
The no-results banner is spaced from the table and restyled to match the suggestions banner language
dbSNP (rsID) queries now resolve to the variant alleles (named e.g. A>C_rs113488022) instead of the reference sequence
Samples in the detail drawer and the expression/coverage tables can be filtered by a minimum number of cells per sample
Per-sample cell counts shown in the sample drawer, and the sample drawer offers direct h5ad download via the live catalog-resolved data files

0.5.6 Latest 2026-08-06

Redesigned the sample detail panel in the Expression and Coverage explorers: facet filter chips, per-study sample cards with expandable metadata, keyboard navigation and a sample detail drawer with facet tiles, external links and the full metadata record
DOI, PubMed and source database are now prominent and clickable in sample details and the sample tables, and stringified-array metadata values (DOIs, PMIDs, accessions) are parsed properly
Fold enrichment is now statistically sound: one-sided Fisher's exact test with Benjamini-Hochberg FDR correction (q < 0.05), multi-valued cells split into separate values and missing values counted as their own category
Fixed a critical bug that prevented the detail panel module from loading and blocking the explorers from rendering, and added a self-healing reload for shared UI modules
Added cache-busting for static assets so deployments are picked up reliably without stale browser caches
Added a smart filter bar to the Expression and Coverage result tables with value suggestions, column context, active-filter chips and a row counter

0.5.5 2026-08-05

Fixed fold-enrichment calculation in Expression and Coverage explorers so rows without expression/coverage data (e.g. empty values) can no longer skew enrichment, and results no longer depend on the table sort order

0.5.4 2026-08-05

Added live under-maintenance mode with per-user allowlist, toggleable via CLI or Redis without a server restart
Maintenance state enforced on the web UI (503 page) and the API (503 JSON), with a public /api/maintenance status endpoint
Added help to the Coverage explorer
Fixed coverage parsing edge cases
Fixed coverage rendering performance for large jobs

0.5.3 2026-06-05

Added selected-sample Download all cells action in Expression Explorer for denominator tables
Kept web per-cell downloads bounded to 50 samples and directed larger workflows to the Python API
Separated all-cell metadata retrieval from search result retrieval so search jobs remain positive-cell focused
Clarified positive-cell matrix exports and independent all-cell denominator downloads in the UI
Proxied Python API all-cell export jobs through the portal while keeping the web UI sample limit
Aligned Python API coverage submission/results with the optimized Coverage Explorer aggregation path
Proxied direct msgpack cell-expression value exports for Python API clients

0.5.2 2026-05-22

Redesigned Analysis page layout to match application aesthetic
Fixed tab bar background coloring — tabs now integrate with frame header
Removed card-like chrome from analysis panels for seamless fill
Styled composition, marker, and DE tables to match expression table aesthetic
Fixed sidebar cell groups list with collapse toggle and refresh button

0.5.1 2026-05-22

Added version badge and changelog page
Fixed light mode accent colors — purple replaced with dark charcoal
Fixed sidebar coloring consistency across all pages
Improved filter panel and search panel icon styling
Fixed table frame header background to match color identity

0.5.0 2026-05-18

Added Analysis page with coexpression, differential expression, and composition tools
New sidebar-based navigation replacing top navbar
Updated color palette and logo styling
Enhanced dark mode with yellow/gold accent theme

0.4.3 2026-05-10

Fixed coverage browser zoom-in behavior
Improved enrichment panel explanation and UI
Added system status indicator in preferences menu
General dark mode enhancements across all panels

0.4.2 2026-05-05

Fixed AI query handling and added LLM disclaimer
Added support for new types of LLM natural language queries
Reorganized search drawer layout for better usability
Fixed UI spacing issues in analysis and coverage views

0.4.1 2026-04-28

Added interactive help guidance and tutorial overlay
Fixed sluggish animations and added tutorial button
Improved search history logging and tracking
Adjusted dark mode brightness for better readability

0.4.0 2026-04-20

Added filter presets with save/load functionality
Added coverage analysis heatmap and text export
Improved mobile layout with global pill navigation bar
Added stats page with aggregate metadata counts

0.3.4 2026-04-12

Fixed expression table compact view performance
Improved navbar text clarity and branding
Added filter text button and legend to expression tables
Enhanced mobile view with responsive controls

0.3.3 2026-04-05

Fixed genome browser rendering issues
Added initial mobile responsive view support
Improved brand name styling and logo colors
Fixed theme color for browser tinting

0.3.2 2026-03-28

Redesigned navigation bar layout
Fixed search wizard panel positioning
Improved overall visual style and spacing
Removed unnecessary motion animations for performance

0.3.1 2026-03-20

Added floating search wizard panel
Fixed card padding and corrected source names
Enhanced landing page with improved layout
Added level aggregation for ontology browser

0.3.0 2026-03-12

Added ontology-based hierarchical filtering (cell type, organ, disease)
Fixed gene table and enrichment table in coexpression view
Reorganized coverage viewer with genome browser integration
Improved download features with barcode support

0.2.3 2026-03-05

Fixed download behavior for barcodes and per-cell matrices
Added UUID display and filter support for barcode downloads
Limited number of downloadable samples for performance
Fixed ontology term matching and cell counts

0.2.2 2026-02-25

Added dark mode support across the entire application
Improved font rendering and positioning
Redesigned navigation bar with better spacing
Fixed Malva branding text and logo styling

0.2.1 2025-08-18

Added search drawer with direct input and AI modes
Added sample count display and logo animation
Fixed ontology filter integration in coverage browser
Improved coverage browser with custom tracks

0.2.0 2025-08-12

Added coverage browser with genome visualization
Added ORCID-based authentication system
Added user profile and job history pages
Added search logging and session management

0.1.0 2025-08-01

Initial release of Malva Explorer
Expression Explorer with search and filtering
Per-cell expression matrix export
Basic metadata column support

You're running the latest version of Malva Explorer

v0.7.5