Actually I also tried the method of Q. Liu et al and got one structure solved after merging 3 different datasets. The only thing different in my case is the Se-SAD phasing instead of Br. But the signal was also weak due to the disorder of incorporated Se in the protein crystals (I supposed so, because I used only 50% of Se-Met during the culture). So in my case, first I processed the data by XDS and made some resolution limitation according to the I/sigma and CCanom. I also did some limitation on the number of images because SCALA can only process not more than 10.000 images (remove the bad images according to the I/sigma and B-factor). And in fact, I did some combination between cut datasets and non-cut datasets and found that the cut datasets really gave me better merged datasets as some noises were removed and the resolution was limited, of course you also lose some information. But in our case, improving the weak anomalous signal to just enough for solving something is more important than losing some information of noise, so we need to be compromised. And any way, for facing with the weak anomalous signal issue, we usually collected such a high abundance of datasets, so losing some information of data is not the problem.