Look at the scatter of the dots in raw expression plots (figure 1e dots). Maybe that's normal for this field, but adjacent scores are all over the place, usually >50% different. That is: very noisy. Imagine the confidence intervals!
Maybe there's signal there, but to my eye, the basic acquisition technique is horrifyingly noisy.
This is only a short period and trivial N, but I also don't see anything systematic in the raw data related to sleep/wake phases.