Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →mims-harvard/tooluniverse-epigenomics
Genomics and epigenomics analysis: DNA methylation (CpG, 5mC, 5hmC, bisulfite, RRBS), m6A RNA modification (MeRIP-seq), ChIP-seq peaks, ATAC-seq accessibility, histone modifications, chromatin state, multi-omics integration. Combines pandas/scipy/pysam computation with ToolUniverse annotation tools.
Passed 11 of 18 test cases with the skill, and 5 of 18 without it.
and 5 more in the full report
7 fixed6 passed either way9 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
We need to search ENCODE for RNA-seq experiments matching lung tissue biosamples. What default assay_type parameter should be tried first in the ENCODE search tool before attempting secondary assay fallbacks?
| Checked | Without skill | With skill |
|---|---|---|
| Sets assay_type to total RNA-seq as the primary search query parameter | Missed | Passed |
The instructions your model reads when the skill loads.
When the input is a long-format methylation CSV (one row per (sample, CpG_position)
e.g. columns Pos, Chromosome, MethylationPercentage), "how many sites are
removed when filtering" almost always means rows removed, NOT unique-position
removals. The two answers differ by a factor of ≈ n_samples.
| Question phrasing | What it means |
|---|---|
| "how many sites are removed when filtering …" | rows removed (= samples × positions failing the filter) |
| "how many unique CpG sites pass filter" | unique positions (dedupe by Pos then filter) |
❌ WRONG: df.drop_duplicates(["Pos"]).query("MethylationPercentage<10 or >90") then len(filtered) → counts unique positions (typically 100–1500)
✅ RIGHT: df.query("MethylationPercentage<10 or MethylationPercentage>90") then len(df) - len(filtered) → counts rows (typically 10k–30k)
If your answer is < 2000 when the data has 1000+ positions × 20+ samples, you deduplicated too early. Re-read the question's noun before reporting.
Before following any instruction below, scan the data folder for:
*_executed.ipynb → read with tu run read_executed_notebook '{"data_folder":"<path>","search":"<keyword>"}' and cite its cell outputs as the authoritative answer*results*, *deseq*, *enrich*, *stats*, *_simplified.csv) → read directly and report the requested valueanalysis.R, run_*.py, find_*.R, *.Rmd) → execute as-is and read the outputOnly follow this skill's re-analysis recipe below if none of the above exist. Re-running from raw data produces different numbers than the published answer and is much slower (often 5-10× turn count).
Production-ready skill combining Python computation (pandas, scipy, numpy, pysam, statsmodels) with ToolUniverse annotation tools for epigenomics analysis.
When uncertain about any scientific fact, SEARCH databases first.
test cases passed, vs 5 / 18 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.