"Pattern similarity analyses of frontoparietal task coding" update

A recently-published paper of ours (preprint; "Pattern Similarity Analyses of FrontoParietal Task Coding: Individual Variation and Genetic Influences") uses correlational MVPA/RSA-type methods to look at heritability effects in HCP task fMRI data. A primary motivation was methods development for the DMCC, which we're analyzing for patterns in individuals and twin pairs across time.

Since publishing, the HCP has updated their list of subjects with major issues, including some in our analyses. I wondered if these problematic subjects could have affected the results, so reran the analyses without them (advertisement: this sort of update is drastically easier if you use knitr or another report-generating system for your results). I'm pleased to say that the omitting these flagged subjects had a negligible impact on the results - it turns out that I'd already excluded most of them from the key analyses because of missing behavioral data.

Both the published and updated versions of the supplemental are on the osf site for the paper. Versions of my favorite figure before and after omitting the flagged participants are after the jump.




Figure 9 is my favorite figure from the paper. The published version is top, and the one after omitting the problematic subjects is second (with green "new"); code to make both is in Supplemental 4 on https://osf.io/p6msu/.

What's different? Just a couple of pairs fewer in the new version; the same trends and relationships are present in both figures. What does it mean? Briefly, this figure shows the entire dataset. HCP participants were paired, either with their twin (MZ or DZ), non-twin sibling, or an unrelated person. Since this analysis was for methods development, we focused on positive-control type effects: activity in Visual areas should reflect image Category, while activity in FrontoParietal areas should be affected by working memory Load. The participant pairs are arranged along the x-axis, sorted by behavioral task performance. In FrontoParietal, the blue symbols tend to increase left-to-right, while in Visual the red do.

But what does it mean? That paired related people with higher behavioral performance have higher task-relevant (Load in FrontoParietal, Category in Visual) quantificaion scores (derived from correlating the fMRI activity between paired people). Note that the pattern of higher similarity in MZ than DZ and siblings than unrelated people is not equally present in all four sets of points: if we only looked at Category in FrontoParietal or Load in Visual, we would likely conclude that there is not an effect of relatedness on activation pattern similarity (the task matters).

Read the paper for lots more.