Hacker News Reader

Comment by greenflag | Hacker News Reader

Interesting discussion on twitter regarding the relatively small sample size (n=15) given the heterogeneity of the underlying condition - https://twitter.com/WiringTheBrain/status/112928482114080358...

View on HN

Finding an indicative difference in a tiny, yet random sample appears to have accomplished more than the guy yelling "small sample size" - and, which I find ironic - "waste!" here. They managed to add a tiny bit of contrast to a picture that's totally blank, and provide 15 pixels of data, please explain how that's supposed to be worthless?

By the same logic, if I manage to cure an extremely rare type of cancer under certain circumstances when it's detected early on, I get fired for having too little success? I mean, what? Seriously?

By all means, prove them wrong, but stop the yelling. That's even more worthless than any form of pioneering research.

It hasn't been established whether this has "accomplished" anything at all or if it's a false positive. That's what's being discussed in the Twitter thread.

It has opened an avenue for future research that has better than "monkeys throwing darts" odds, which is an accomplishment.

Because you can easily find "a tiny bit of contrast" by comparing any two things; there will always be differences. this doesn't demonstrate that it isn't just noise and this difference is biologically and clinically meaningful.

Also, "the yelling" is called peer review. It's the entire basis of science.

I'd imagine "peer review" would be along the lines of taking another set of samples and comparing results, but maybe that's just me.

Peer review does not involve trying to replicated the results as far as I'm aware... it's just commentary, hopefully by qualified people, on the perceived standard of the work at hand.

Also, this technology is brand new. I mean, brand new. Single-cell RNA seq is only a few years old; single-nuclear is even newer. There are significant problems with analyzing this data, mostly related to the necessity to correct statistically between samples to remove batch effects before you can cross-compare. This is very, very hard and in general we are bad at it. I would say we can't really do it yet, and so experiments purporting to show differences between patients are likely to be the result of statistical artifacts, and not actual biological differences.

Reply on news.ycombinator.com