Another weight values for kinectome?

[pedrocau1]

Hi Karolina and Robbin,
so far we are using the pearson correlation values for the weights in our kinectome. While this is a good predictor for relationships between the nodes, it has some downsides: it spots mainly linear correlation and it has also negative values. The negatives values may affect the values of some graph invariants and obscure the interpretation( like clustering coefficient and maybe some of the modularity results maybe affected by it).

it just occurred to me that we could use the distance correlation (https://en.wikipedia.org/wiki/Distance_correlation). It ranges between [0,1] and can also spot non linear relations.

Maybe we could test it and see if it helps with results, before thinking of other fancy ways of defining another weighting function for the kinectome graph? What do you think? We would still be considering a correlation metric but more robust (in the sense it may spot non linear associations) and better suitable for some graph analysis. I've push a small func to do it in kinectome.py file/ top_centrality branch.

[karsae]

Hey Pedro,

thanks for your input. I have calculated the kinectome using distance correlation for one subject and plotted it together with pearson's kinectome and indeed the results are similar, but not the same. I'll be back to work next week and will calculate the kinectomes for all subjects using distance correlation, then we can compare the two kinectomes (pearson's vs. distance corr) on a bigger level.

Thanks again :)