| name | life-sciences |
| description | Use for bench biology: designing and reading wet-lab experiments, choosing controls, telling biological from technical replicates, randomisation and blinding, batch effects, dose-response curves, image and blot quantification, antibody and cell-line validation, and animal work under ARRIVE/IACUC. Reach for it whenever the task involves in vitro or in vivo data, assays, plates, staining, cytometry or a protocol headed for institutional review.
|
| version | 1.0.0 |
| author | Remedy |
| tags | ["research","biology","wet-lab","experimental-design","replicates","biosafety"] |
| requires | [] |
| tools | ["lit_search","lit_fetch","cite_add","power_analysis","stats_assumptions","stats_effect_size","data_profile","manuscript_check","skill_activate"] |
| triggers | ["\\b(wet[- ]?lab|in vitro|in vivo|western blot|RT-?qPCR|\\bqPCR\\b|\\bELISA\\b|flow cytometry|immunohistochem\\w+)\\b","\\b(cell (?:line|culture|viability)|knock(?:out|down)|CRISPR|plasmid|transfect\\w+|passage number)\\b","\\b(biological replicates?|technical replicates?|\\bIACUC\\b|\\bIBC\\b|biosafety level|BSL-?[1-4])\\b","\\b(dose[- ]response curve|\\bIC50\\b|\\bEC50\\b|\\bLD50\\b|vehicle control|sham (?:group|control))\\b"] |
Life sciences (bench work)
Call skill_activate(skill="research-method") first — it owns question
framing, evidence standards, preregistration, citation honesty and how to say
"we do not know". Do not restate any of that here. This pack is only what is
different at the bench.
Where the method stops
Do the ordinary published science: culture, assays, blots, qPCR, cytometry,
imaging, dose-response, protocols written to ARRIVE. Do not supply
operational detail for pathogen enhancement (host range, transmissibility,
immune escape), toxin production or purification, select-agent handling, or
acquisition routes for controlled biological material. That is not a footer
disclaimer — it is where the procedure ends. Point the owner at the process
that governs the work (IBC for biosafety and recombinant work, IACUC for
animals, IRB for human subjects and tissue, institutional export control for
shipping) and help them prepare that submission instead of routing around it.
Say it once, plainly, and carry on with the work you can do.
The unit of replication — settle this before anything else
Ask two questions: what varied independently, and what got measured twice?
- Biological replicate — an independent unit: a separate animal, a
separate donor, an independently thawed and separately treated culture.
Only these license a claim about the population.
- Technical replicate — the same unit measured again: three wells split
from one culture, triplicate qPCR from one cDNA, two fields of one slide.
These estimate measurement noise. Average them, then use n = biological
units.
- Pseudo-replication is the commonest wet-lab statistics error. 24 wells
from 3 cultures is n = 3. 60 cells imaged from 3 mice is n = 3 unless you
fit a mixed model with animal as a random effect.
Write the definition into the legend before running any test: "n = 4
independent cultures; each point is the mean of 3 technical wells."
power_analysis sizes on the biological n. If the owner reports n as wells
or cells, ask which unit was randomised rather than assuming.
Procedure
- Design before pipetting. Name the comparison, the replication unit,
the controls, the randomisation and the blinding. Size with
power_analysis from a pilot SD or a published effect; with no defensible
prior, say so and call the run exploratory.
- Controls in the same plate, run and day as the samples — positive,
negative, vehicle, untreated, isotype, no-template, no-RT, sham. See
references/experimental-controls.md; a missing control is rarely
recoverable afterwards.
- Randomise and blind allocation, plate position and scoring. Blinding
matters most wherever a human reads the outcome.
