How to hand your sequencing data to an analyst (and get a better result back)
The file formats, metadata and written questions that turn a two-week analysis into a three-day one — with a copy-paste checklist.
Do this on your own data: the cheapest way to avoid a week of back-and-forth is to check the data before you send it anywhere. The free QC report tells you in minutes whether the matrix is complete and readable: omics.hstgenomics.com/qc-check. What you should expect to get back: omics.hstgenomics.com/sample.
Want it done rather than explained? We do analysis only — send a count matrix, an h5ad or Seurat object, Cell Ranger output or a public accession and get publication-ready results with the code that produced them, at a fixed price agreed before any work starts: omics.hstgenomics.com/quote.
Most delays in an outsourced analysis are not analysis. They are a week of emails establishing what the samples are and what the question is. Ten minutes of preparation removes almost all of it.
Send the right file, not the biggest one
For single-cell work, the ideal handover is the filtered feature-barcode matrix from Cell Ranger (the
filtered_feature_bc_matrix folder or its h5), one per sample. Raw FASTQ is only necessary when the
alignment itself must be redone — a different reference genome, a custom transgene, or a suspected mapping problem.
Sending 300 GB of FASTQ when 300 MB of matrices would do costs days of transfer and adds nothing.
For bulk RNA-seq, a gene-by-sample count matrix plus the sample table is enough for differential expression. For spatial, send the whole Space Ranger output directory including the tissue image — the image is data, and an analysis without it is not a spatial analysis.
Send a sample table that answers the obvious questions
One row per sample, with: sample ID exactly as it appears in the file names, subject or animal ID, group or condition, timepoint, sex, batch or processing date, chemistry version, and anything you already know is odd about that sample. The subject ID is the field people forget, and it is the one that determines whether the statistics can be done correctly at all.
Write the comparisons down
Not "find what is different" — that is not answerable. Write the contrasts you want:
- "Treated versus control, within each cell type, at day 7."
- "Day 7 versus day 0 within treated animals."
- "Does the proportion of cell type X change with treatment?"
Three lines like these determine the entire analysis plan, and they are the difference between getting an answer to your question and getting a competent analysis of a different one.
Say what the deliverable is for
A figure for a grant application, a first look to decide the next experiment, and a figure set for submission to a specific journal are three different jobs with different levels of polish. Saying which one saves both sides from guessing.
Flag the constraints early
Deadlines, data that cannot leave your institution, embargoes, consent restrictions on human data, and any requirement that specific software be used (some collaborators insist on Seurat objects). All of these are easy to accommodate up front and expensive to discover halfway through.
Copy-paste checklist
□ Filtered count matrices (or h5 / h5ad / Seurat object), one per sample □ Sample table: sample ID, subject ID, group, timepoint, sex, batch, chemistry □ Species and reference genome used □ The contrasts I want, written as sentences □ What the output is for (grant / next experiment / submission to journal X) □ Deadline and any data-handling restrictions □ Anything already known to be wrong with a sample
Anyone competent can start work the same day with this. Without it, the first three days are spent reconstructing your experiment from file names.