Outsourcing bioinformatics analysis: a checklist before you pay anyone

Handing your data to an external analyst is routine. The disappointments are also routine, and they are almost always caused by things that were never agreed in writing. This is the list to work through before money changes hands — with whoever you use, including us.

Scope

1. What exactly is delivered? Figures, results tables, the processed object, the code, the methods paragraph. Ask for the list. "A full analysis" is not a list.
2. Which statistical approach for differential expression? For a comparison between groups of subjects, the answer should involve pseudobulk aggregation, not a per-cell test. If the answer is vague, that is your signal.
3. How many rounds of revision are included, and for how long? Reviews come back months later. A revision window shorter than six months is not much use.
4. What happens if the data turns out to be poor? Agree in advance what a QC-failed sample costs and whether it can be dropped from the invoice.

Reproducibility

5. Will you receive the code? Not a description — the actual scripts, with parameters.
6. Are software versions recorded? Results change between versions of Seurat, scanpy and annotation references. Without versions the analysis is not reproducible, whatever else is delivered.
7. Can your own team re-run it? A container image or an environment file is the difference between "we can re-run this" and "we would have to pay again".

Data and confidentiality

8. Where does the data live during the work, and who can see it?
9. Is it deleted afterwards, on what timetable, and confirmed in writing?
10. Is the data ever reused — for benchmarking, marketing examples, or model training? The answer you want is no, in writing. Unpublished data is the most valuable thing you own.
11. If human data is involved, does the transfer comply with your consent and your institution's rules? This is your responsibility, not the vendor's, and it must be settled before a single file moves.

Authorship and credit

12. Agree the credit up front. Routine pipeline work is normally an acknowledgement; substantial intellectual contribution to design or interpretation may be authorship. Deciding this at submission time is how collaborations sour.

Two smaller things that save real time

Send count matrices rather than raw FASTQ when you can — it removes days of transfer and processing, and most analysis questions never touch the raw reads. And put your comparisons in writing before work starts: "treated vs control within each timepoint, and treated-day-7 vs treated-day-0" is a specification. "Find what is different" is not, and it is the single most common reason a delivered analysis does not answer the question that was asked.