Can we source localize the interested ICA components?
Remove all the components that you don't want (it is OK to keep only a handful), then estimate the sources for these selected ICs.
If you compute a noise covariance matrix from the recordings, the same IC decomposition and selection must be applied to the recordings.
I think we may be using similar methods for data processing. I am keen on using ICA on brainstorm & to do source localisation after cleaning.
I also recommend the EEG and epilepsy tutorial, very helpful! It may also help to consider the type of data you are working on, i.e. meg or scalp eeg.
I also recommend the following link for practise in detecting ICA components:
If you are not familiar with these concepts, I recommend you start by following all the introduction tutorials (section "Get started") using the example dataset provided, before trying to process your own recordings. https://neuroimage.usc.edu/brainstorm/Tutorials