Thread by @damadam

Our paper is live! Can fMRI measure well-being? Can the "positive-negative axis" from @fmrib_steve et al. be seen in a larger, independent dataset w/younger subs collected across sites and scanners? Want it as a preregistered replication? Read on! doi.org/10.1098/rsos.2… 1/14

Using code made publicly available by @fmrib_steve, we first did an exact computational replications of their 2015 Nature Neuroscience paper. It worked! You can read about that at BioRxiv: doi.org/10.1101/2020.0… 2/14

Next, we wondered if this same relationship would show up in a totally independent, younger, cross-scanner dataset like the ABCD study funded by @NIAAAnews @NIDAnews @NIMHgov @NICHD_NIH @NINDSfunding @NIHOBSSR @NIMHD @NIH 3/14

We sent our analysis plan to @chrisdc77's RoyalSoc Open Science journal and the reviewers were excellent! They pointed out many issues and had great suggestions. The final plan was pre-registered at @OSFPrereg doi.org/10.17605/OSF.I… @RSocPublishing @BrianNosek @openscience 4/14

At last to the analysis! we preprocessed the ABCD resting state fMRI and subject measure data to prepare for canonical correlation analysis (CCA). After exclusions, the final inputs to the CCA consisted of 5013 subjects, 73 subject measures, and 200x200 connectivity matrix. 5/14

As in the original paper, we found a statistically significant positive correlation between functional connectivity and subject measure canonical weights, suggesting a robust multidimensional relationship exists. 6/14

When looking at the top 20 modes of the CCA, we found that Mode 2 was the “one mode to rule them all” - that is, the mode that met our criteria as being the primary mode. 7/14

Now, here’s the exciting part: CCA weights for Mode 2 and subject measure scores were arranged along a positive-negative axis, just like what was previously shown in HCP! Positively-valanced subject measures = positive weights, negatively-valence measures = negative weights. 8/14

Obviously, the brain is more complex than a single dimension, but we were surprised how well this finding replicated! 9/14

Finally we looked at the relationship between CCA weights and functional connectivity edges. As was shown in HCP, the default mode network is represented in the edges most strongly related to Mode 2. However, … 10/14

… we found that Default Mode Network (DMN) nodes were not clustered together, but rather distributed among multiple clusters, which makes sense since the DMN in adolescents undergoes protracted development. 11/14

Co-first authors @nikhil_r_goyal & @dmoracze spearheaded this ambitious replication w/guidance from @esfinn @fmritoday & @damadam. 12/14

Huge tnx to auths of the replicated paper for making data+code avail @fmrib_steve @ten_photos @neurovidaurre @AndersonWinkler @behrenstimb @KamilUgurbil @BarchDbarch @fmrib_karla David Van Essen,Matt Glasser and to @chrisdc77 for his work advancing prereg'd replications! 13/14

See the paper for all the details! royalsocietypublishing.org/doi/10.1098/rs…
We count this as a #preregistered #openscience #replication win! Check out @ABCD_ReproNim for more about reproducibility w/ABCD @phyzang @satra_ @dnkennedy44 @jbpoline @DrDamienFair @HGaravan @yarikoptic 14/14