Commissioning a sequencing or proteomics study

Commissioning a sequencing or proteomics study means paying for a measurement that can answer a biological question you wrote down first. The expensive failure is a large data set that cannot support the sentence you later want to publish, because the reference genome was vague, the replicates were not biological, the samples were degraded, or the deliverable was a picture without a table. This guide applies the sourcing habit in how to write a laboratory sourcing enquiry to that kind of statement of work.

Method classes you may need to name include whole-genome sequencing, shotgun discovery proteomics, and protein identification by LC-MS/MS. Those pages are enquiry references. They do not say that EVRINTH operates the sequencers or the mass spectrometers. Sample-preparation consumables live in the sample preparation catalogue, and sequencing-related categories in the genomics and sequencing catalogue. The wider context is the genomics research pathway. Send the statement of work through laboratory procurement, the quote form, or contact. University groups can also begin from academic research.

The decision you are locking

You are locking four things before a library kit is opened or a protein is digested: the question, the reference database, the quality bar that can stop a sample, and the files that count as delivery. Everything else in the provider's brochure is a method of getting those four. If a quotation describes chemistry and does not answer the four, it is not yet a statement of work.

You are also locking the claim you will not make. A research variant list is not a diagnosis. A protein abundance table is not a clinical test. This page is not medical advice, and the study you commission from it should say the same in the consent and the report if human-derived material is involved. Ethical approval and legal permissions are yours to obtain. A sequencing provider cannot invent them.

Start from the question, then the method class

A question about sequence differences across a genome you can name is a whole-genome conversation. A question that only concerns protein-coding exons is an exome conversation, and it will not see the regions the capture ignores. A question about one amplicon may be a much smaller assay. A question about which proteins are in a gel band is an identification conversation. A question about how a proteome differs between two conditions is a quantitative discovery conversation, with a replicate plan and a false-discovery rule.

Short-read next-generation sequencing, as introduced in public overviews such as Illumina's technology note, produces many reads that are interpreted against a library design and a reference. Long reads answer different structural questions and have different failure modes. Do not write "NGS" in the statement of work and stop. Write the library type, whether RNA methods need to preserve strand, the read layout you expect the quotation to confirm, and the analysis you want back.

Proteomics at discovery scale is usually bottom-up: proteins are digested to peptides, peptides are separated and fragmented in a mass spectrometer, and spectra are matched to a sequence database. Identification of a band is a narrower version of the same idea. The database is not a detail. UniProt proteomes differ by species, by reviewed status, and by version. An old or wrong database moves the identifications. Lock the version in the file.

References, replicates, and batches

For a genome, cite the assembly. Ensembl and GenBank accessions exist so that "the human genome" or "the mouse genome" is not a vibe. A variant coordinate without the assembly is not a result another laboratory can check. For a non-model organism, say whether you need a public reference or a de novo assembly, and who does the annotation.

A biological replicate is a separately derived sample: another animal, another culture, another field plot. Technical replicates, resequencing the same library, measure the instrument. They do not measure the biology. Write the number of each. A design that sequences one pool of the treatment and one pool of the control cannot support a strong claim about variation between individuals. It can support a hypothesis you will test properly later. Say which of those you are buying.

Batches confound. A month's gap, a different extraction day, or a different cell passage can look like a condition if you are careless. State whether samples will be randomised or blinded, and ask how the provider's own run batches will be recorded. If all treated samples are in one batch and all controls in another, the statement of work should say that you accept that confound or that you refuse it.

Intake quality control is a stop rule

Nucleic-acid samples fail from insufficient amount, from fragmentation, from salt and solvents, and from DNA in an RNA library. Agree an integrity check and a minimum amount before the samples travel. A provider's intake assay is part of the purchase. If a sample fails, the rule should be to stop, tell you, and wait. "Run it anyway" produces a low-complexity data set that will tempt someone to interpret noise.

Read-level checks after sequencing belong in the deliverable. Tools in the class of FastQC summarise per-base quality, adapter content, duplication and other library pathologies. You do not have to mandate one software brand. You do have to say that a quality summary will be returned and that a failed library will be identified as failed.

Protein samples fail from too little material, from dominant keratin or from the wrong lysis buffer, and from degradation. A gel or a quantification method agreed in advance is the stop rule. Identification work should say whether you are sending a band, a lysate, or an immunoprecipitate, and what contaminants you expect.

Analysis and the shape of the deliverable

Ask for raw data and for the processed table. For sequencing, that may mean reads plus alignments or counts against the named reference, or a variant list with the caller and the filters written down. For proteomics, that means identified peptides and proteins, the database version, the false-discovery rule, and any quantification values with the method named. A PDF of enriched pathways, with no table and no statistics, is not a complete deliverable. It may still be a useful picture on top of the table.

HUPO, the Human Proteome Organization, is a public home for community discussion of proteomics standards. Ask the provider which reporting minimum they follow, and file the answer. Do not assume a logo implies a particular guideline.

If you will confirm a transcript result with quantitative PCR, specify that assay with the same care. The MIQE guidelines, published by Bustin and colleagues in Clinical Chemistry and available under doi:10.1373/clinchem.2008.112797, list the sample, assay, and analysis information a quantitative PCR report needs. Agree those fields before the plates are run. MIQE is not a proteomics guideline and not a diagnostic authorisation. It is a checklist against vague PCR reporting.

Software names and versions belong in the report. A result that cannot name its aligner, its quantifier, or its search engine cannot be repeated. Who may be contacted when a column in the table is ambiguous also belongs in the statement of work.

Branch points

If intake QC fails, do not interpret the run. Re-extract, replace the sample, or drop that sample from the design and say so. A design that silently loses the awkward replicates is biased.

If the report uses a different assembly or a different UniProt version from the one in the statement of work, send it back before you mine it for stories. Coordinates and protein accessions will not match your notebook.

If the only output is a pathway cartoon, ask for the underlying identifications and the statistics. If they do not exist, you bought a drawing. The sourcing page explains how to treat that as an exception against a numbered line, not as a surprise at the group meeting.

If the provider batches all of one condition together, record the confound next to any difference you see. You may still use the data as pilot evidence. You should not pretend the confound was not there.

Cold chain, heat, and the day the box arrives

Extracted RNA and many protein lysates do not survive an afternoon on a warm dock. The statement of work names the temperature, the coolant class, the receiver, and the rule if the coolant is gone on arrival. A hot season and a long road leg shorten a dry-ice hold. Build the handover around a person who is present, and around a freezer that is already cold. Power cuts at the receiving bench are part of the plan, not an anecdote after the samples are lost.

The same heat does not damage a carefully dried DNA spot the way it damages RNA in a buffer. The stability statement of the material you actually send is the one that counts. Write it down so the courier's habit does not become the scientific method.

What the report will not prove

It will not prove a clinical diagnosis. It will not, by itself, prove that a variant changes a protein's function, or that a protein difference causes a phenotype. Those claims need orthogonal experiments. It will not repair a biased design. Depth and instrument quality cannot invent replicates you did not collect.

It will not be comparable to another study that used a different reference, a different false-discovery rule, or a different quantification method, unless you do the work of comparing them. Write the methods so that comparison is possible later. That is the practical point of Ensembl accessions, GenBank records, UniProt versions, and a named search threshold.

What to put in the quotation request

Include the question, the organism, the sample type and number, the replicate definition, the reference accession or proteome version, the library or digest class, the intake QC and the stop rule, the analysis deliverable, the data files, and the cold-chain handover. Ask the provider to state what the method will not resolve. Ask which reporting minimum they follow.

Use the quote form or contact, and keep the numbered statement with the procurement note. Compare answers on those lines. A quotation that is silent on the database version is not yet cheaper or dearer in any scientific sense. It is incomplete. Complete it before a sample leaves the freezer.