Vitessce

Résumé fourni par la source

Patch Changes Make cell set selection changes paint immediately on large datasets: the scatterplot views now defer set selection and color values into a non-urgent render (via useDeferredValue), so the sets manager checkbox updates in the first commit and the per-observation recoloring follows, and the contour stratification pass only runs when contours are visible or the points layer is hidden, instead of on every selection change. (@vitessce/scatterplot-embedding, @vitessce/scatterplot-gating) (#2571) Three follow-up optimizations surfaced by the 2M-cell dataset. The remaining dense obs-by-feature loaders (AnnData obsFeatureColumns, plain matrix zarr, CSV) now share an allocation guard: an over-budget estimate logs a warning before the attempt, and an allocation failure is rethrown as a descriptive MatrixTooLargeError naming the matrix. The spatial views build their lasso/rectangle-selection quadtrees lazily on first use instead of on every data update. Zarr data sources memoize node opens per path, so a node's metadata documents (.zattrs/.zarray/.zgroup) are read once per data source instead of once per access pattern. loadNumericForDims uses the contiguous sliced read only when the requested dims actually share chunks, and per-dim reads when each dim has its own chunk column, avoiding needless 2D assembly on per-column-chunked embeddings. (@vitessce/abstract, @vitessce/csv, @vitessce/zarr, @vitessce/spatial, @vitessce/spatial-beta) (#2571) Adds support to provide a list of segments (cell-ids) to the neuroglancer view to avoid dependence on obsSets to generate the view (@vitessce/neuroglancer, @vitessce/example-configs, @vitessce/schemas) (#2547) Cache zarr store reads through the react-query client so concurrent requests for the same chunk share one download (fixes duplicate embedding chunk fetches), load contiguous embedding dims with a single sliced read, and read single-level categorical obs sets as raw codes end-to-end — building the sets tree, membership lookups, and the scatterplot color encoding from typed arrays instead of per-observation strings, with a fallback to the string-based route for scored, multi-level, or non-categorical columns. (@vitessce/zarr-utils, @vitessce/zarr, @vitessce/sets-utils, @vitessce/types, @vitessce/scatterplot-embedding, @vitessce/scatterplot-gating) (#2568) Improve obs set performance on large datasets: accumulate set colors, memberships, and color indices in linear rather than quadratic time; color the embedding and gating scatterplots from a positional typed array instead of an observation-ID-keyed Map; and build the observation set membership encoding in a web worker, dispatched during idle time and transferred as typed arrays, falling back to the main thread when workers are unavailable. Also fixes treeToMembershipMap reporting duplicate set paths for leaf sets shallower than the height of their hierarchy. (@vitessce/sets-utils, @vitessce/workers, @vitessce/scatterplot, @vitessce/scatterplot-embedding, @vitessce/scatterplot-gating, @vitessce/types, @vitessce/zarr, @vitessce/csv, @vitessce/json) (#2567) Added a layer coordination type (spatialLayerLabel) that allows users to specify a label for the layers' (@vitessce/layer-controller-beta, @vitessce/neuroglancer, @vitessce/constants-internal, @vitessce/all, @vitessce/example-configs) (#2552) Name the obs set that holds observations with a missing categorical value with a shared placeholder, so it renders identically in the sets manager, tooltips, and plots, and so selections of it survive view config serialization. (@vitessce/utils, @vitessce/sets-utils, @vitessce/tooltip) (#2568) Show a loading overlay while a lasso/rectangle selection is being calculated and applied. At millions of observations the hit test and the selection-driven re-render block the main thread for seconds; the selection layer now signals a busy state, lets the browser paint the view's loading indicator first, and clears it in the same commit that shows the applied selection. (@vitessce/gl, @vitessce/scatterplot, @vitessce/scatterplot-embedding, @vitessce/scatterplot-gating, @vitessce/spatial, @vitessce/spatial-beta) (#2571) Read individual features from CSR-encoded obsFeatureMatrix stores by scanning indices/data in chunks instead of densifying the whole matrix, coalescing concurrent selections into one scan and refusing scans beyond the browser's allocation budget with a descriptive MatrixTooLargeError; warn before full-matrix densification that exceeds the budget and report allocation failures descriptively; fix CSC selection on int64 indptr and the silent last-column result for unknown features on dense matrices; document that csc_matrix is the preferred sparse encoding. (@vitessce/zarr, @vitessce/zarr-utils, @vitessce/utils, @vitessce/error, @vitessce/vit-s, docs) (#2570) Speed up per-observation work on large datasets: stratify expression values and embedding arrays (violin/dot plots, scatterplot contours) from positional set indices instead of string-keyed map lookups per observation, using raw categorical codes when available; build the observation index map only when a set selection resolves; skip embedding re-alignment and the expression index mapping when the observation indices are the same array; and build the scatterplot's selection quadtree on first use rather than on every embedding change. stratifyArrays no longer takes a sampleId-to-obsIds map argument. (@vitessce/sets-utils, @vitessce/vit-s, @vitessce/scatterplot, @vitessce/scatterplot-embedding, @vitessce/statistical-plots) (#2571) Upgrade zarrita package version to fix bug reading sharded zarr v3 data. (@vitessce/spatial-zarr, @vitessce/zarr-utils, @vitessce/zarr) (#2546) Add an obsColors data type (one RGB color per observation), an obsColors.csv file type, a useSegmentationMultiObsColors multi-level data hook, and support for the new obsColors option of obsColorEncoding in the Neuroglancer view. (@vitessce/constants-internal, @vitessce/schemas, @vitessce/csv, @vitessce/vit-s, @vitessce/neuroglancer, @vitessce/all) (#2547) Versioned Sites Docs Demo

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.