Forum:Reasons to refuse to give support

Question

[Benn]

I would like to start a discussion about giving bioinformatics support to other researchers (biologists), and when to stop giving support or end collaborations. Small background here, I am a bioinformatician at a small research institute for almost 5 years, and do the analysis for many projects mostly RNA-Seq data. Most of the experiments are with sound statistical design (biological replicates), and sometime researchers ask my advice about proper designs before the start the experiment (all good). However, I have encountered some researchers that come with data with poor designs, hence n=1 experiments. They disagree with that fact, and claim they have biological replicates, and say that it is a matter of opinion.

For example, they took cells from 5 (blood) donors, pooled it and then took 2 samples from this same pool to do some treatment. Then they claim they have n=2, but I say no this is a technical replicate here with n=1. No biological replicates, so I cannot do sound statistics here. At the end, after a ghastly discussion, (always with the "mice from the same strain are also genetically identical so who cares?" argument), the only thing I can do is to say that my collaboration stops here, I am not wasting time in n=1 experiments. I had this twice now, and I am wondering how other bioinformaticians deal with these situations? Please share your experiences or your manner to deal with these cases. Thanks.

[h.mon]

As a contract bioinformatician (employed by a company), I perform the analysis, but tell the customers beforehand they will get crap results, and ask if they want to proceed anyway.

Side comment: I think an n=1 experiment is poor design, but at least it is honest. Pooling 5 samples, taking 2 technical samples from this pool and calling them biological replicates is, at best, an "honest mistake".

[WouterDeCoster]

I guess this depends on your job description...

Answer 1

[GenoMax]

You could always do the analysis. Provide the standard disclaimers (results provided AS IS use at your own risk) and ask them not to include you in the publication. That way no one is likely to find the publication in future and it would not badly reflect on you.

Since you are an employee at the said institution I doubt that you can refuse to do the analysis (however bad a design, it may be). Your performance eval may be negatively affected. But if you are in a position to refuse to do the analysis, document your reasons to do so in an email. Make sure your supervisor is informed of this first before you actually refuse to do the analysis so (s)he can back you up, if needed. (as an aside if this person IS your supervisor then you have less choice but to document your reservations and carry on).

Answer 2

[Devon Ryan]

For actual collaborations your reasons to refuse can be anything you want. If your potential collaborators have nonsense designs like you're describing then tell them that (in nicer words) and let them go. Likewise, if you can't agree early on about authorship then you should decline a collaboration.

This is obviously different if you're not actually collaborating but instead providing a service. In that case declining an analysis on a nonsense experiment is totally reasonable because you have limited time and need to prioritize for experiments that are likely to go somewhere. You need to be able to take the fall out from that, but if "you suck at experimental design" is the harshest thing your collaborators have heard then they should count themselves lucky.

[Benn]

Thanks for sharing guys (all of you), this helps a lot! I was hired by the institute to get NGS experiments to a higher level, so I am not a contract bioinformatician that has to do what ever they want me to do. Maybe I can send them to you then h.mon (I am kidding). The only reward I get is co-authorship, so I would like to focus indeed on sound experiments (which I agree on).

[Devon Ryan]

Particularly if you were hired explicitly to up their game you can use that as a defense if they complain.

Answer 3

[Dawe]

Here’s what we do: in a true collaboration we take part to the conception of the experiment from the very beginning, we discuss which conditions should be met, which should be avoided. Of course there may be some borderline cases (e.g. unique samples in the sense they are extremely rare), we discuss them and we explain the limits of any analysis. Sometimes a collaborator comes with the experiment performed already. Again, we warn about the limits and the fallacies that may come from poorly designed experiments but, in general, we provide a bird’s eye view on data. From that point on, the collaboration can only continue if some requirements are met.

Answer 4

[Kristoffer Vitting-Seerup]

I have started to refere to this paper for a discussion of technical vs biological replicates - I especially like the definition of science as something that can be replicated.

What usually brings home the message when a discussion of what the current experimental setup would allow generalisation to (in this case no generalisation). I have actually been discussing it so much that I added a section to my vignette about it recently.

With regards to the number of samples needed there are plenty of benchmark studies showing that to get trustworthy result from RNASeq data at least 3 replicates are needed - enough so that all recent best practice papers refere to it. See fx Box 1 in this paper.

In summary I think you have a good cause for saying no due to your scientific integrity, because as a community we want research to be reproducible and because there probably are other project you could do. To be honest I would take it up with your boss asking whether they think it make sense you use time on analysing datasets that will not give reproducible results.