Why you have to look at the raw data, not just the analysis.
Going over some 2019 data, looked at the students' notebooks, and they just have the quantification of 'puncate' intensity. This is generally a good proxy of whether our protein of interest is delocalized to the cytosol. It looks like the mutations doesn't do much, which is what the (undergrad) student concluded.
Then I looked at the actual micrographs, umm yeah...
This isn't my first rodeo, which is why I looked at the raw data.
It's just amazing to me how often an inexperienced trainee will apply a quantitative pipeline without actually looking at the images to ensure the pipeline will be valid for the samples.
In this case, there are notably fewer puncta, so the appropriate pipeline would count puncta per cell.
@MCDuncanLab agree that this is frustrating and that it's difficult to teach people... I used to worry about my trainees scrolling past interesting phenotypes on the scope and not noticing.
Our analyses these days tend to be pipelines which in theory makes the problem worse because trainees don't look at the data due to the automation. However, they're written in ways that analyse multiple things at once and would have detected this phenotype.
@steveroyle @MCDuncanLab I feel this is a problem in many areas and maybe it's me being old and grumpy, but I feel that many students nowadays are not really thinking about what they're doing and why.
I think in part this is due to due to a higher accessibility of science.
We have kits in the lab, we have R and Python packages that make very complicated analysis super easy, which is great, but it also means that you can do those analysis without thinking about what you *actually* are doing... it's a fine line to walk on
@nicolaromano in defence of "kids these days”... we're writing the analysis routines ourselves, so I feel that they can understand what the goals are as we develop a routine. We stay away from black box software solutions generally and try to roll our own. But obviously there's a limit and you have to rely on libraries and plug-ins at some point for efficiency. I share your concern though that trainees goals can be shallow: get a graph to put in my report even if it means nothing. @MCDuncanLab