Steven is right, this is what I tried to do during my rotation a few years ago, and I should have just used the C++ applications. I wasted a decent amount of time and a potential paper…. But, I learned python and parts of pyrosetta, so that was useful.
Anyhow, I am not sure if Abrelax works in PyRosetta. After some digging, there is an AbrelaxApplication object exposed. You can make one using x = AbrelaxApplication(). However, I know nothing about it, and wouldn’t recommend trying to get it to work. Best bet is to use Abinitio in main Rosetta, cluster the 10-30 thousand models using the program calibur, and then take the top cluster or two and loading that into pyrosetta or in main rosetta to do loop modeling / further repacking / design/ etc.