It makes no sense whatsoever to first screwup the relations between R-noise, G-noise, and B-noise, and then let a color noise reduction application try to figure out how to reduce noise, when it's known that that is not how the model was trained.
What does make sense is to let the application remove things like Chromatic Aberrations and RGB noise, then desaturate or use another method to produce a monochrome image.
Cheers,
Bart