Teach Drylab Your Lab's Protocol
Save your lab's QC thresholds, naming rules, statistics and figure style once as a custom skill, then name it in your chats.
For PIs, core facilities and anyone who repeats the same kind of analysis: write your lab's conventions down once as a custom skill in Knowledge, then name it in your chats. Drylab looks the skill up and uses your thresholds and figure style instead of picking its own. Check its first answer to make sure it did (see below).
When to use it
- Everyone in the lab should use the same QC cutoffs, for example 200-6000 genes per cell and under 15% mitochondrial reads.
- You always name output files the same way, so results from different people line up.
- Your lab has fixed statistical choices: the test, the multiple-testing correction, the significance cutoff.
- Journals or your PI expect one figure style: file type, resolution, colors, font size.
Prompt examples
The protocol to save (paste it into SKILL.md Content when you create the skill, below the short header the template starts with):
# Morgan Lab scRNA-seq protocol
## QC
- Keep cells with 200 to 6000 detected genes.
- Keep cells with less than 15% mitochondrial reads.
- Report how many cells each filter removed.
## Clustering
- Use Leiden clustering with resolution 0.5.
## Statistics
- Find markers with the Wilcoxon test and Benjamini-Hochberg correction; call genes with adjusted p < 0.05.
## Figures and file names
- Save every figure as a 300 dpi PDF.
- Name files <sample>_<plot>.pdf, for example PBMC1_umap.pdf.A chat that relies on it:
Follow my custom skill "Morgan Lab scRNA-seq protocol". Run QC and clustering on @pbmc_sample.h5ad and tell me how many cells each QC filter removed.Check the rules before running anything:
Use my custom skill "Morgan Lab scRNA-seq protocol". Before I upload my next sample, list the QC thresholds, clustering resolution and figure format it tells you to use. Don't run any analysis.Same protocol, different sample:
Apply the Morgan Lab scRNA-seq protocol to @pbmc_sample2.h5ad and save the UMAP and the QC violin plots with our file names.Steps
Save the protocol (once)
- Open Knowledge in the left sidebar and click Create Skill.
- Choose Write Manually. Enter a Skill Name, such as "Morgan Lab scRNA-seq protocol", and a Description that says when it applies, such as "Use for any single-cell RNA-seq QC, clustering or figures in the Morgan Lab."
- In SKILL.md Content, keep the short header at the top (between the
---lines, updatenameanddescriptionto match) and replace the rest of the template with your protocol. - Click Create Skill. It appears under Custom Skills.
Already have the protocol in a paper, a PDF or an earlier analysis? Choose From Materials or From Conversation instead, and Drylab drafts the skill for you. See Create a skill.
Use it in a chat
- Start a new chat and name the skill in your request, in quotes, as in the examples above. Drylab looks up your custom skills by name and description.
- Check the answer before you trust the results: Drylab should name the skill and repeat your thresholds and file names. If it uses other values or says it can't read the skill, mention your protocol as a file instead (next section).
To change a rule later, open the skill, click Edit, change it and Save. Each saved change is a new version in Version History, so you can Restore an older one.
Create and edit skills in the web app: Knowledge isn't in the Desktop App's sidebar.
Or mention a protocol file
In the web app, save the same text as lab_protocol.md in your project's Project Files (upload it, or ask Drylab to save it there), then mention it in each request: Follow @lab_protocol.md and run QC on @pbmc_sample.h5ad. Drylab reads exactly that file. In the Desktop App, put the file anywhere in your project folder. See Mention Files and Resources.
Tips
- Name the skill in your request. Drylab doesn't apply your custom skills to every chat on its own; naming the skill makes sure it looks for it.
- Keep the protocol short and specific. Numbers, test names and file-name patterns work better than general advice. One skill per kind of analysis is easier to reuse than one long document.
- Use the description to say when it applies, for example "Use for paired tumor vs normal bulk RNA-seq". It helps Drylab pick the right skill when you have several.