To produce the best flats, the mean ADU of the flats should equal the mean ADU of the Light frames. This result comes about because the camera (sensor + electronics) is not linear.
EXPLANATION:
I've seen a few recommendations for the proper ADU level (mean ADU) to use for the flats. However, when I tried the recommended values the flats didn't correct very well. I'm using a Gain of 140 and an Offset of 50, and, the Light frames used had a mean ADU of approximately 3600.
Because the flats didn't correct well I tried an experiment. I created Master Flats with mean ADU levels of 3000, 5000, 7000, 10000, 15000 and 25000. The maximum ADU value for this camera is 65535 (16 bits). If the camera (sensor + electronics) is linear, then, all of the Master Flats should produce equal corrections when used to calibrate the Light frames. The picture below shows the results of six individual calibrations. Each calibration used a different one of the Master Flats mentioned above. Their are 6 individual images in the picture below. Each individual image shows the same small portion of the original image. Counting from left to right, the 6th image (far right) shows the result of calibration with a Master Flat created using a mean ADU value of 25000. The 5th image shows the result of calibration with a Master Flat created using a mean ADU value of 15000 and so on. If you look at the extreme right edge and bottom of the 6th image you will see that it is slightly darker than the rest of the image, i.e., the Master Flat didn't fully correct the edge of the image. As you look at the 5th image then 4th image and so on you will see that the correction is slowly getting better. Finally the 1st image (the one on the left) looks best of all with the mean ADU of the Master Flat being 3000 which is close to the the Light frames mean ADU of 3600. A Master Flat with mean ADU of 3600 would be the best.
ASI1600 Flats by STEPHEN KING, on Flickr
Because the camera is not linear, taking the Flats with the same mean ADU as the Light frame ensures that any small change in brightness of the Light frame (caused by vignetting or a dust mote or both) will have the same change in brightness of the Flat, because, both are at the same nonlinear portion of the sensors response (same mean ADU). If the camera were truly linear this wouldn't be necessary.
I don't know if all of the ASI1600MM-Pro sensors have the same degree of non-linearity that mine does, but, this is now a requirement for me.
Hope this makes sense, and, is useful for you.
Steve