Hello. Referencing the following line of code, I was curious, "Throughout RoseTTAFold2's codebase, is the atom27 representation of a given protein structure precisely the same as the atom14 representation except for having 13 additional atoms appended to the end?". I ask because I was wondering whether it is correct to assume that one can slice the first 14 atoms from an atom27-shaped tensor to extract an equivalent tensor of atom14-shape (since such an operation would not be correct in different codebases such as that of AlphaFold 2 as the atom14 and atom37 tensors there have atoms in slightly different orders per residue).
Hello. Referencing the following line of code, I was curious, "Throughout RoseTTAFold2's codebase, is the
atom27
representation of a given protein structure precisely the same as theatom14
representation except for having 13 additional atoms appended to the end?". I ask because I was wondering whether it is correct to assume that one can slice the first 14 atoms from anatom27
-shaped tensor to extract an equivalent tensor ofatom14
-shape (since such an operation would not be correct in different codebases such as that of AlphaFold 2 as theatom14
andatom37
tensors there have atoms in slightly different orders per residue).https://github.com/uw-ipd/RoseTTAFold2/blob/a26f29bb92d4946ef8bf225edba9240c9e647747/network/train_multi_deep.py#L199