Dataset selection

The Dataset selection page decides which datasets, and which samples within each dataset, every other page analyses.

Structure

The sidebar lists the binding and perturbation datasets under two headings. Each dataset row has a switch that activates it and a filter button that opens its filter modal. Hovering over a dataset’s name shows its description. The modal has two columns: characteristics shared across datasets on the left, each with an “Apply to all datasets” toggle that copies the setting to every dataset with that characteristic, and dataset-specific characteristics on the right. “Queue Filters” stages the modal’s settings; “Reset” clears them.

Experimental-condition columns with defined levels appear as checkboxes, each labelled with the level’s definition. Selecting values of another characteristic, such as carbon source, narrows the condition checkboxes to the conditions that occur with it. When the dataset already has filters, the modal opens with those other characteristics set to the values that occur with the filters, and offers only the conditions that match.

Edits are staged, not applied. While any switch or filter differs from what is applied, the sidebar shows “Dataset selection has changed. Click Apply Changes to update.” and the Apply Changes button is highlighted; clicking it commits every staged change at once. A switch whose state is staged but not yet applied is drawn yellow.

The workspace shows the dataset intersection matrix, with one row and one column per active dataset:

  • a diagonal cell shows the dataset’s regulator and sample counts under its filters. Clicking it opens a modal that says whether every regulator is represented by one sample and, if not, which characteristics distinguish a regulator’s samples;
  • an off-diagonal cell (upper triangle) shows the number of regulators the two datasets share. Clicking it opens a modal whose “Select common regulators” button restricts every dataset to that shared set immediately, as a regulator_locus_tag filter; the cell stays highlighted while that restriction is in force.

Below the matrix, Regulators by Dataset lists each regulator against the active datasets it appears in, with a search box to pick out regulators.

First load

The page opens with the default selection already applied: the datasets marked active_default in the collection config’s tags are switched on, and DEFAULT_DATASET_FILTERS (tfbpshiny/utils/vdb_init.py) is applied. The defaults are chosen so that every dataset has exactly one sample per regulator.

Export

When at least one dataset is active, the sidebar footer offers an export. The download is not the data itself but a small kit, tfbpshiny_export-<datetime>.tar.gz, that the user runs on their own machine to pull the data through labretriever. Extracting it creates tfbpshiny_export-<datetime>/ with:

  • brentlab_yeast_collection.yaml: a copy of the collection config, so the script resolves the same HuggingFace repositories and configs as the app;
  • fetch_data.py: one block per active dataset, embedding the SQL and bound parameters of metadata_query and full_data_query under the current filters. Run, it queries labretriever.VirtualDB for each dataset and writes one subdirectory per dataset containing metadata.csv, annotated_features.csv and, when a description is available, a README.md;
  • requirements.txt: the dependencies fetch_data.py needs (labretriever);
  • README.md: how to create a virtual environment, install the requirements and run the script.

Effect on other pages

The active datasets and the applied filters determine the datasets and samples the Binding, Perturbation, Comparison and Figures pages draw from. Filters set on a primary dataset apply to all of its promoter-set and peak-calling variants.