If users want, they should be able to denoise the optimally combined data produced by fMRIPrep in whichever space they want. The tedana pipeline would still use the native-space echo-wise data, of course, but the bad components could be regressed out of any target space's data.