From spectra to a biological claim

The sentence "protein X causes pathway Y" is a biological claim about mechanism. A treated-versus-control cell digest cannot carry it. The strongest sentence the data can support, after they have climbed a ladder of checks, is closer to this: under this design, protein X was higher in treated cells than in control cells. This page walks one fictional study down that ladder. The identification steps are bottom-up proteomics in plain language. The comparison the ladder is trying to earn is the subject of the differential abundance reference, used here as a description of the question, not as a statement that a particular study is running.

A raw file and a thresholded identification

The study is ordinary research. A cultured line, a treatment and a vehicle, independent flasks, harvested together and processed in a mixed order. Nothing in that design has been measured until a raw file exists for each injection and the file name joins a row on the sample sheet. A spectrum that cannot be traced to a flask is not evidence about the treatment.

Identification is the next step, and it is a threshold. Peptide-spectrum matches pass a stated false-discovery rate. Protein groups are built with a stated rule. The database version, the taxon and the contaminant list are the ones written in the report. Protein X at this stage means a group of accessions that explains the observed peptides under those settings. It does not mean the protein functioned, and it does not mean the isoform a reviewer likes best. If the group contains three accessions and no unique peptide separates them, the ladder can only carry the group.

A contaminant that shares the name of a protein you hoped to see ends the climb early. Keratin, serum albumin from the medium, and trypsin are identifications of the bench and the digest. They can be higher in one group because a glove or a medium change differed. Flag them. Do not promote them to pathway members. The Human Proteome Organization is where research protein evidence is argued in public. PRIDE and ProteomeXchange show how raw files and identifications are deposited when a study is shared. Your claim should be readable at that level of detail even if you never deposit it.

Quantification that follows the cell design

Quantification is a new experiment on the same files. It uses intensities or reporter ions, not the mere presence of a match, and it uses the design. Independent flasks are the biological replicates. Two injections of one digest are a technical pair. If the treated arm is three flasks and the control arm is three flasks, the contrast has three and three. If all six digests were labelled and combined, the design is different and the report has to say so.

Normalisation and missing values are part of the number. A protein that is missing in every control and present in every treated flask is interesting and still ambiguous: it may have risen from nothing, or the sampler may have failed to pick it when it was scarce. Show the pattern. Say whether anything was imputed. A fold change computed from imputed lows is a different object from a fold change computed from measured intensities.

The ladder does not require one named statistical package. It requires that the contrast match the design and that the report name the procedure. A ranking of ratios from unreplicated flasks is not that procedure. A test that ignores a batch the sample sheet already records is not that procedure either.

The contrast, then an orthogonal check

Only after the contrast exists do you have a sentence about abundance. "Protein group X was higher in treated flasks than in vehicle flasks, under this processing order, at this false-discovery threshold, with this missing-value rule" is long and fair. You may then ask whether the change is large enough, relative to the spread among flasks, to deserve a follow-up. The follow-up is not automatic.

An orthogonal check uses a different measurement. A blot with a characterised antibody, described in western blot from gel to membrane, can ask whether an immunoreactive band at the expected size also moves. It can disagree, because the blot sees epitopes and the mass spectrometer sees peptides. A targeted measurement of the same peptide in a new digest can ask whether the intensity change repeats. A second set of flasks can ask whether the contrast was a one-day event. None of these proves that protein X caused a phenotype. They test whether the abundance sentence survives contact with another method.

Pathway tools sit beside the ladder, not on it. They regroup the same list. A cluster of metabolic enzymes moving together is a prompt for a hypothesis. The NCBI Bookshelf is one public place to read cell-biology background while you form that hypothesis. The background chapter did not observe your flasks. Write the hypothesis as a hypothesis.

Batch and contaminants end claims

Two claim-killers appear late and should have been caught early. If every treated flask was digested on Monday, the fair sentence is that Monday differed from Tuesday. You do not own a treatment effect. Repeating the processing in a mixed order is the repair when sample remains. A formula that "corrects" a fully confounded batch is not a repair.

If the proteins that separate the groups are keratin and albumin, the fair sentence is about handling or medium. Look back at the contaminant flags before you open a pathway tool. A single unique peptide with a modification the fragments do not localise is a similar stop: report the peptide, and do not report a mechanism at a residue.

A third killer is silent imputation. If the controls are empty because of missing values you filled with a low number, the fold change is partly your fill rule. Say the rule in the same sentence as the fold change, or do not quote the fold change.

Research sentences are not medical advice

The cell study does not diagnose a person and does not prescribe a treatment. Even a human cell line is a research reagent. Misidentification of the line, contamination, and the biosafety level of any genetic modification are institutional issues. A careful abundance sentence does not settle them. Do not let the final slide use clinical language because the protein name is familiar from a textbook.

Timestamps when the power fails

Write the harvest time, the digest day and the injection order in the notebook next to the sentence you hope to earn. If a power cut splits digestion or the queue, the sentence has to mention the split or it is too strong. A later reader should see that flasks after the restart were a second batch. In a humid week, a pellet that thawed on the bench gets the same honesty: the abundance of oxidation-prone peptides may be the thaw. The ladder's sentence shrinks to what the design still supports. That shrinkage is the result.

Discussing a differential study

Bring the design, not the hoped-for verb. Say how many independent cultures, what the vehicle is, whether processing can be mixed, which database and contaminant list you require, and whether an orthogonal method is planned. Ask for a deliverable that stops at the sentence the design earns.

The differential abundance reference is the enquiry page for that contrast. The shotgun discovery proteomics reference and the protein identification by LC-MS/MS reference are the identification work underneath it. Send the design with the quote request. Those pages are how the question is named in a discussion. They do not mean the flasks have already been acquired, and they do not authorise a causal sentence the ladder cannot reach.