Nils Kroemer

From what I read the preprint has major discrepancies in resting state fMRI data preprocessing compared to drysdale. Signal frequencies examined do not overlap between the studies. Dinga et al. highpass filtered at .01 Hz, whereas drysdale et al bandpass filters: 0.008-0.09 Hz.

The two studies make completely different choices regarding motion censoring, spatial smoothing, and denoising approach. Until the preprocessing is revised, one cannot interpret the preprint as a replication.

Still, seems like an important existence proof that high CCA and discovery of clusters are obtainable under their respective nulls.

In terms of the analytic approach within the methods, sure. However, it cannot be interpreted with respect to drysdale et al. The timeseries data carry different artifacts and signal at high frequencies; the data were not denoised, so the analytic approach was applied to noise.

Agree re the availability of best practice pipelines for rsfMRI, but @dinga92 et al's work illustrates that critical comparisons for evaluating validity of taxa/clusters was not made. AFAICT, this is largely independent of data.

Unless efforts are made to clean the data, calling this a replication effort misleads readers unfamiliar with nuances of rs-fcMRI preprocessing. That doesn't mean drysdale is valid either. Rather, both papers have critical, addresable flaws that render interpretations meaningless

I think both papers provide useful information (i.e. I don't think all interpretations are meaningless), though admittedly I am still working through Drysdale et al to understand exactly what they did re the post-clustering classifiers.