Hi Isaac,

Not sure but from your QC plots there are indications of potentially needing to use more conservative denoising settings. I would probably suggest to try increasing the number of aCompCor components, using a more conservative outlier detection threshold, and/or using low pass filtering if not doing that already. I would also suggest running the QA plots "QA denoising: distribution of FC values" and "QA denoising: distribution of QC-FC associations" to get a better sense of how your data after denoising looks and evaluate the relative success of different denoising strategies (also see "Morfini, F., Whitfield-Gabrieli, S., & Nieto-Castañón, A. (2023). Functional connectivity MRI quality control procedures in CONN. Frontiers in Neuroscience, 17, 1092125" for a few more details)

Hope this helps

Alfonso

Originally posted by Isaac Treves:

I recently had a situation where I did melodic using 'dswau' files from conn, selecting networks that spatially correlated with the Yeo networks. To my surprise and my co-authors surprise, there were no anticorrelations between any of the networks. I'm now running a new project looking at task-based fMRI, and seeing the same thing, but using ROIs from CEN, DMN, SN, instead of ICs. I've attached connectivity plots (fisherz transformed) , as well as a couple example QCs. I'm doing filtering and despiking outside CONN for the new project, but I did it together in the first one - so I don't think that's the root of the findings. 

The one instance I got anticorrelations was when I ran GIFT on 'dswau' files, and the anterior DMN component was anticorrelated with other networks.