dataset_dropdown_opus

[cornhundred]

No description provided.

[cornhundred]

Pull Request Overview: Multi-Dataset Support with Dropdown Selector
Summary
This PR adds support for visualizing multiple datasets in the Landscape widget with an efficient dropdown-based switching mechanism that reuses the existing deck.gl context instead of rebuilding the visualization from scratch.
New Files (untracked, need to be added)

1. js/ui/dataset_dropdown.js (~94 lines)
   Creates a compact dropdown UI for dataset selection:
   Small dropdown (55px width) that expands on focus/click
   Shows short labels with full names on hover (tooltip)
   Handles async dataset switching with loading state
   Subscribes to current_dataset_index observable to stay in sync
2. js/ui/switch_dataset.js (~284 lines)
   Core logic for efficient dataset switching:
   State persistence: Saves/restores selected genes, clusters, image visibility, and UMAP/spatial view across dataset switches
   Layer updates: Clones deck.gl layers with new data and unique IDs instead of recreating them
   Cache invalidation: Properly disposes old image layers and creates new ones with unique cache keys
   Data reloading: Fetches new landscape_parameters, cell_metadata, cluster_metadata, meta_gene, etc.
   Modified Files
3. src/celldega/viz/widget.py (+62 lines)
   Python widget changes:
   New base_urls traitlet: Accepts list of {'url': str, 'label': str} dicts
   New cell_name_prefix traitlet: When True, trims dataset prefix from cell names (e.g., "dataset-name_cell-name" → "cell-name") to map correctly to LandscapeFiles
   Automatic conversion of various input formats (single URL, list of URLs, list of dicts) to standardized format
   Cell name prefix trimming applied to both meta_cell and UMAP data
4. js/widget.js (+6 lines)
   Passes new base_urls and cell_name_prefix parameters from Python model to landscape_ist.
5. js/viz/landscape_ist.js (+8 lines)
   Accepts base_urls and cell_name_prefix parameters
   Stores them in viz_state for use by dataset switching logic
6. js/obs_store/obs_store.js (+3 lines)
   New observables for dataset management:
   current_dataset_index: Tracks which dataset is currently displayed
   dataset_switching: Flag to prevent concurrent switches and disable certain updates during transitions
7. js/ui/ui_containers.js (+7 lines)
   Integrates the dataset dropdown into the image controls section of the UI.
   Key Features
   Multiple Dataset Support: Pass a list of URLs to base_url parameter
   Efficient Switching: No WebGL context accumulation - reuses existing deck.gl instance
   State Persistence: Selected genes/clusters, image visibility, and view mode persist across dataset switches (only valid selections for new dataset are restored)
   Cell Name Prefix Handling: Supports datasets where cell names have been prefixed for uniqueness
   Improved UX: Dropdown expands on interaction, shows tooltips with full dataset names
   Usage Example
   landscape = dega.viz.Landscape( technology='Xenium', base_url=[ {'url': 'https://bucket/dataset1/', 'label': 'Sample A'}, {'url': 'https://bucket/dataset2/', 'label': 'Sample B'}, ], adata=adata, cell_name_prefix=True, # If cell names have dataset prefixes height=500,)
   Bug Fixes (in working directory, may need commit)
   Fixed gene click from Clustergram causing solid red cells on first click (reordered update_selected_cats to run after update_cell_exp_array)
   Converted camelCase variables to snake_case for consistency
   Note: The new files dataset_dropdown.js and switch_dataset.js are currently untracked and need to be staged/committed.

[cornhundred]

@jaspreetishar and @huanlity - you all can follow this example to test locally https://colab.research.google.com/drive/1VBJUNQoldIjpWxyUOdSd7oIOThpC5mri?usp=sharing

[huanlity]

Works great on my end. Two unexpected things I found are:

1. if you turn off IMG and change the dataset, the new dataset will still display the underlying IMG.
2. While 1 is happening, the IMG sign is greyed out.

Fixing either one would work but fixing 1 will be better IMO.

[cornhundred]

@huanlity I'm ok fixing this in a subsequent pull request because we have several other pull requests that build off of this one and I want them to not diverge too much.