Feature for future versions - FNC and Time properties evaluations

[jpprincich]

Dear developers, i was wondering if it is possible to allows NBSPredict to perform classification-predictions on FNC matrices..
Guess it is technically possible to test correlation matrices between brain networks but may need to be adapted for nodes grouping or networks atlas definition.
Additionally and more challenging but not less important .. could we expect NBSPredict to perform classification and predictions testing time varying properties of dynamic connectivity data in a near future? Lets say build ML models that can assess specific recurrent connectivity patterns extracted from networks timecurses along with their temporal properties or metastates such as Dwell time, number of transition states etc..

I think these are very sensitive measures that may capture additional properties of brain connectivity and remains largely unexplored.
Preprocessing methods are widely available, mostly based on ICA such GIFT toolbox.

Hope this sound interesting to you as well and that can be considered as new features of your extraordinary toolbox.
Congratulations.

Sincerely

JP

[eminSerin]

Hi!
Thanks a lot for your suggestions.

1. Technically, as long as the input connectivity matrices are symmetric, NBS‑Predict should work. However, note that the feature‑selection algorithm in NBS‑Predict inherently attempts to find connected subnetworks (please refer to Serin et al., 2021; Zalesky et al., 2010). If you do not want this behavior, then you should opt for simple ML algorithms using sklearn or PhotonAI.
2. Unfortunately, realistically, I don’t think this will be implemented in the near future. This is mainly because NBS‑Predict is an extension of NBS (Zalesky et al., 2010), which only works for simple functional connectivity, and I lack sufficient expertise in dynamic connectivity.
   Thanks!