DepMap Explorer prompt

Analyze public DepMap data to identify selective cancer dependencies, then create a **downloadable, self-contained HTML application** for exploring and verifying the results.

Build the finished application and provide the files, not just a plan or source-code listing.

## Inputs

**Data source:** DepMap 24Q4 Public, version 1.

**Editable biological comparison:**

- Group of interest: **microsatellite-unstable (MSI) cancer cell lines**
- Comparison group: **microsatellite-stable (MSS) cancer cell lines**

These groups are the main replacement targets when adapting this prompt to other cancer lineages, molecular subtypes or mutation-defined groups.

## Question

Which gene knockouts impair the group of interest more strongly than the comparison group?

## Retrieve the data

Use these direct downloads from the archived release:

- **CRISPRGeneEffect.csv:** [https://ndownloader.figshare.com/files/51064667](https://ndownloader.figshare.com/files/51064667)
- **OmicsSignatures.csv:** [https://ndownloader.figshare.com/files/51065726](https://ndownloader.figshare.com/files/51065726)
- **Model.csv:** [https://ndownloader.figshare.com/files/51065297](https://ndownloader.figshare.com/files/51065297)
- **README.txt:** [https://ndownloader.figshare.com/files/51065795](https://ndownloader.figshare.com/files/51065795)

Read the release documentation and download the complete files required for analysis. A preview or partial download is insufficient.

Calculate results from the retrieved data. Do not substitute published findings, remembered results or synthetic data.

## Define the groups

Join the files by ModelID. In the gene-effect and signature matrices, ModelID is stored in the first, unnamed column.

For the MSI comparison:

- Use `MSIScore` from `OmicsSignatures.csv`.
- Classify scores >20 as MSI and scores ≤20 as MSS.
- Exclude missing MSI scores.
- This file already contains model-level values; no sequencing-profile selection is required.

For an alternative comparison, use the appropriate documented annotations. Retrieve any additional annotation files from the same release and record their sources. Do not infer group membership from cell-line names.

Verify that each model contributes only once. Report unmatched models, exclusions and final group sizes.

## Analyze dependencies

For each gene:

1. Calculate the mean Chronos gene-effect score in each group.
2. Calculate the difference: **group of interest minus comparison group**. More-negative scores indicate stronger dependency.
3. Compare groups using a two-sided Welch’s t-test.
4. Apply Benjamini–Hochberg correction across all successfully tested genes.

Handle missing values explicitly and report each gene’s sample counts. Mark untestable genes without inventing statistics.

Identify the ten genes with the largest negative mean differences among those with adjusted P<0.05. Do not force a particular gene to appear.

## Build the HTML application

Deliver one HTML file containing embedded CSS, JavaScript and the real data needed for its controls. It must work locally without installation, external libraries or an internet connection.

Include:

- A concise explanation of the comparison and score direction.
- A ranked table of the ten selected genes, showing sample counts, group means, mean differences and adjusted P-values.
- A gene selector covering those genes.
- An interactive plot showing every included cell line for the selected gene.
- Selectable points revealing ModelID, cell-line name, lineage, group, MSI score where applicable, and exact gene-effect score.
- A searchable source-value table linked to the plot.
- Group counts, means and mean difference recalculated deterministically from the embedded source rows.
- Download buttons for the displayed source values and selected-gene results.
- A provenance panel identifying the release, source URLs, group definitions and exclusions.

Embed the source rows for the selected genes; the HTML does not need to contain the entire genome-wide matrix. All controls must operate using ordinary code, without AI calls.

## Verification and deliverables

Check selected source values directly against the original downloaded files, separately from the ranking calculation. Verify that displayed counts, means and differences match the analysis outputs.

Test that the HTML opens locally and that gene selection, point inspection, filtering and downloads work. Report any checks you could not perform.

Provide:

1. `depmap_dependency_explorer.html`
2. `selective_dependencies.csv` containing results for all tested genes.
3. `selected_gene_source_values.csv` containing the model-level values used in the explorer.
4. A runnable analysis script that retrieves the data and reproduces the results.

Briefly interpret the leading result, whether it reflects a broad group difference or a few extreme models, and any conspicuous exceptions.

Use the frozen 24Q4 source data for verification. Do not treat values from a newer live release as interchangeable.

If retrieval fails, identify the missing files and request them. Never claim that analysis or testing was completed when it was not.
​