One thing that is always tricky to dissect: multiple aliquots for a patient and different aliquots have different assays.

Sometimes patients don't have all of the aliquots, which makes things a headache when you're trying to track down discrepancies in patients btw assays.

I'm mentioning this because this actually happens a lot, and it takes a lot of detective work to find these discrepancies.

In an ideal world, all patients would have all aliquots. But they don't.

This is also why you can't just hand off the data and metadata to an analyst and expect an analysis - these discrepancies have to be noted and explained.

You need a feedback loop between the analysts and the experimentalists. Otherwise, there will be a lot of discrepancies.

This is also why I don't take any chances on assuming I understand what the metadata means when it is handed to me. That way lies hubris.

I guess my point is that analysts need to have conversations with subject matter experts/data collectors and not assume anything about how the data is collected.

This is probably the difference between a great analyst and an okay one - they actively wonder about these things and their curiosity makes the analysis much stronger.

The post-hoc analysis, where you try and answer why you see a pattern in the data is actually way more important than the actual statistical analysis.

It establishes that you understand what patterns your collaborators are looking for and whether they can believe them or not.

Anyway, sorry this is blathering. I mostly write such things to remind myself rather than be preachy.

But it's also why I spend so much time teaching Exploratory Data Analysis - it makes you curious about the data and it makes you agile and nimble in your approaches to the data.