Research Integrity Responsibilities of a Lab Lead

Lab culture is an integrity system, and when the lead stays silent about standards, shortcuts quietly harden into the norms everyone follows.

Checklist · 10 min read · Directive Publications Editorial Team

Core responsibilities

1. Authorship honesty — no gifts, no ghosts
2. Data stewardship — retention, access, honest exclusion rules
3. Image and figure integrity — no inappropriate manipulation
4. Mentoring under pressure — deadlines do not excuse fabrication
5. Corrections — fix the record when errors appear

Authorship honesty means the byline describes who did the work. A gift author is someone added for a reason that has nothing to do with the paper: the department head who never read a draft, the collaborator who supplied a reagent, the colleague you owe a favour. A ghost author is the reverse — someone whose contribution meets the criteria and whose name is missing, most often a technician, a statistician, a medical writer, or the student who left the group before submission. Both are decided by the lead, because in most labs the lead is the only person who can add or remove a name without paying for it. In practice: open a contribution note in the project folder at the first experiment, record who did what in the words you would use in a contributorship statement, and update it as people join and leave. Then read the list out before submission. What goes wrong is almost never a dispute about the first author; it is the person who left eighteen months ago whom nobody thought to contact, and who reads the paper when it appears.

Data stewardship is three separate promises. Retention: raw data and analysis code outlive the person who produced them, on lab-controlled storage, in a format someone else can open. If a departing student's laptop holds the only copy of a dataset, you do not have data, you have a dependency. Access: a second person can find the file, understand the naming, and rerun the analysis. Test that rather than assume it — pick a published figure, ask a lab member who did not make it to reproduce it from the archive, and watch where they get stuck. Honest exclusion rules: decide what counts as a failed run, an unusable sample or an outlier before you see which points hurt the result, and write the rule down where it can be checked later. Criteria invented after looking are the most common way an honest lab produces a wrong paper, and they feel entirely reasonable to the person applying them.

Image and figure integrity is where good intentions do the most damage, because the tools make the problem invisible. Adjustments to brightness or contrast that are applied to the whole image, are modest, and do not make a feature appear or disappear are ordinarily acceptable, and you say so in the legend. What is not acceptable: splicing lanes from different membranes into one panel without a visible divider and a statement, cloning or erasing anything, reusing a panel for a different condition, cropping to remove an inconvenient control, or adjusting until a band emerges from the background. The working rule is short enough to teach in one sentence — if you would not be comfortable describing the adjustment in the figure legend, do not make it. The structural fix is originals. Keep the unprocessed files straight off the instrument, with their metadata, filed with the manuscript rather than on one person's own drive, and name the acquisition and processing software in the methods. Labs get into trouble here not because someone faked a blot, but because two years later nobody can produce the file that would show they did not.

Mentoring under pressure is tested exactly once per person, and you rarely notice it happening. The first time a trainee brings you a result that wrecks the schedule, your reaction sets the price of honesty in your lab for everyone watching. Say the rule out loud before the situation arrives: when the experiment fails a week before the deadline, the deadline moves or the claim shrinks, and the data does not change. Be careful with ordinary phrases that are heard as instructions. "Can you make that figure work" and "have another look at those outliers" mean something specific coming from the person who signs the reference letter. Ask what the data shows instead, and make it explicit that missing a submission window is an acceptable outcome and inventing a replicate is not.

Corrections are your job rather than the first author's, because you are the person with the standing to write to an editor and the person who will still be reachable when the file is reopened. The section below sets out what to do.

Guides: Who Counts as an Author? · COPE Core Practices

A worked example

A postdoc sends you the final figure set the night before submission. One blot panel looks cleaner than you remember it. You ask for the original files, and it emerges that the lanes came from two membranes run on different days, were assembled into a single panel with no divider, and the composite then had its contrast raised to even out the background.

Nothing was invented. Every band is a real band. The panel is still wrong, because a reader looking at it will assume one experiment, one gel and one exposure, and that assumption is false. The decision rule that settles most figure questions is exactly that one: ask what a reader is entitled to conclude from the panel as drawn, and if the answer is something untrue, the panel changes.

Here the fix is cheap. Rebuild the figure with a visible dividing line between the spliced sections, state in the legend that the lanes were run on separate membranes, and apply any adjustment uniformly to the whole image. You lose an hour and gain a panel nobody can question. If the same figure had already been published, the fix is the one you would use for any error in the record: write to the journal, describe what the panel actually shows, and propose a corrected version.

How you handle the conversation matters more than the figure does. If the originals exist and the intent was cosmetic, treat it as a training gap, fix it with the person in the room, and turn it into a rule at the next group meeting rather than a story about someone. If the originals do not exist, stop and deal with that instead, because it is the more serious problem and it is unlikely to be confined to one figure.

Practical lab policies worth writing down

• Notebook / data deposit expectations
• Who can submit as corresponding author
• How authorship is revisited before submission
• How to raise concerns internally without retaliation

One page in the shared drive is enough. The point of writing it down is that the answer stops depending on who is most senior in the room when the question comes up. Walk each new member through it in their first week, and reread it yourself once a year, because the version you carry in your head drifts.

Notebook and data deposit expectations should say what gets deposited, where, under what naming scheme, and how often. Set the interval short enough that a deposit is a routine act rather than a project in itself. Say that analysis code is deposited alongside the data it produced, and that deposit happens before submission rather than after acceptance — the months between those two are where files disappear as contracts end.

Who can submit as corresponding author is a security rule as much as an etiquette one. That account receives the editor's mail, signs the licence, answers integrity queries and, if something goes wrong later, is the address the journal writes to years afterwards. State who may hold the role, that it uses an institutional address rather than a personal one, and who inherits the paper when that person leaves. A published paper whose only contact address is a dead university account becomes a problem at the worst possible moment.

How authorship is revisited before submission should name a moment, not an intention. A short meeting or a circulated message in which the list and the order are read out, each named person confirms their contribution, and anyone who has left the group is contacted, works because it forces the decision while the manuscript can still be changed. Held after the submission form is filled in, the same conversation is a dispute.

How to raise concerns internally without retaliation has to name someone other than you. "Come and talk to me" is a good instinct that fails in precisely the case it exists for, which is a concern about the lead. Name a second PI, a departmental integrity contact or an ombuds office, and state plainly that raising a concern in good faith will not affect authorship, references or funding. Then behave that way the first time somebody uses it, because that is the only version of the policy anyone believes.

When something goes wrong

Disclose early to the journal/institution. Delay is better than a later retraction cascade.

In the first hours, preserve rather than tidy. Take a copy of the raw files, notebooks, code and correspondence exactly as they stand, and tell everyone involved to change nothing. Do not ask the person concerned to quietly rerun the analysis and send you a corrected figure, however tempting that is. It destroys the evidence that the original was an honest error and leaves you unable to show what actually happened.

Then triage against a single question: does the error change a conclusion? If it does, the paper needs a substantial correction or a retraction, and the journal should hear from you now rather than when your own review is finished. If it does not, and the problem is presentation, a mislabelled axis or a wrong panel, a corrigendum usually settles it. If you cannot tell yet, write anyway — say what you know, say what you are checking, and give a date by which you will report back. Editors deal with honest uncertainty routinely. What they cannot work with is silence.

Tell your co-authors before or at the same time as the journal, including the ones who have moved on, and tell your research integrity office if there is any chance the cause is misconduct rather than error. The two processes do different jobs: the journal fixes the record, the institution establishes what happened. Keep your first message to the editor factual — the paper, what is wrong, how you found it, which results are affected, what you propose — and leave the attribution of blame out of it until somebody has actually established the facts.

The reason early disclosure is worth the discomfort is what it prevents. An error you report is an error. An error found by a reader after you knew about it becomes a question about everything else the lab has published, and that kind of attention rarely stops at one paper.

If you are starting a lab this year

Do these in order. Say your standards out loud at the first group meeting, in plain words, before there is any incident to attach them to. Write the one-page policy and put it where people actually look. Set up lab-controlled storage and make the first deposit yourself, so a naming scheme exists before anyone has to invent one under pressure. Run the reproduce-one-figure drill on your own most recent paper, and fix whatever it exposes. Name the person outside your reporting line that a worried trainee can go to, and say that name aloud rather than burying it in a document. None of it takes a month, and all of it is easier now than after the first thing goes wrong.

More: Directive Academy · Publication ethics