Thread by @ndosenbach

shorturl.at/kyCEU has lots of parts to it. I like #neuroscience Twitter for getting a quick overview of what's happening. @smarek0502 & @tervoclemmensb wrote detailed 🧵s with figs. I'll just give bullet pts of what the study is & isn't saying 👇

It's specifically about BWAS (brain-wide association studies) as defined in the 2nd sentence: ‘studies of the associations between common inter-individual variability in human brain structure/function and cognition or psychiatric symptomatology’. BWAS ≠ neuroimaging

as brilliantly detailed here:

https://twitter.com/ten_photos/status/1504801784199585796

the key distinction is between cross-sectional association (BWAS) and detection (classic task fMRI, lesion studies, within-participant treatment effects, etc). The large sample size requirements are for associations.

The title emphasizes sample size, but improving measurement reliability, where possible, is of course strongly recommended. Unfortunately it won't be a magic bullet, since structural MRI and behavioral measures already have high test, re-test reliability. The supplement has more

Of course, these findings cannot be applied to potential future BWAS that could be conducted with massively improved or entirely different technologies.

In general associations using BOLD (RSFC/task fMRI) > cortical thickness; NIH Toolbox > CBCL; multivariate > univariate. It is important to remember that while more powerful with the largest samples, multi-variate methods were more prone to overfitting with small samples.

Besides sample size there's a whole host of other important considerations such as processing standardization, covariates, etc. None of these are 'either/or', they're all 'and'.

Why is the future for neuroimaging research of behavior brighter than ever? Everything needed for even better human neuroscience is already happening: 👇

consortia: HCP, ABCD, UKB, Enigma, HBCD, etc
bottom-up data sharing: openneuro.org, etc
standardization: BIDS, fMRIprep, etc
more & higher reliability BOLD data per participant: PFM, etc
more & better behav data: EMA, passive sensing
improved analytics: @bttyeo cont.👇

and in contrast to genomics there's myriad non-BWAS ways to study human brains
neurosurgical, lesion pts
interventions
longitudinal
classical task fMRI, MEG, DOT, etc.
precision functional mapping

practically: wanna do BWAS - download the biggest data sets; share your data & code; participate in standardization efforts ... be more like GWAS. Or do non-BWAS human which we've done with N=1-3 (shorturl.at/mnyAL shorturl.at/gxAOS shorturl.at/rxAY4)