Yes - I just had the models opened in the opposite order to you.
Ultimately, it seems to me that the problem here is that you’re asking for something impossible: a good overall structural alignment between two models that are fundamentally different. A rigid body alignment can only rotate and translate one model to overlay the other. If the two models are in very different conformations you have two choices: (a) get a bad overall all-atom alignment, or (b) try to find the biggest piece that *does* align well and just align that. By default MatchMaker takes the latter approach, which is why when you use it on your two models it ends up aligning just the antigen-binding region of the antibody fragment at the expense of everything else - that’s the biggest part of the two models that *can* be aligned well.
If you really want approach (a) you could use MatchMaker with the argument “cutOff None”, in which case it will try to find the best-aligning whole chain and align on that. You can also specify exactly which pair of chains you want to align on, as per the example in my previous email. Ultimately, though, nothing you do here is going to turn a bad AlphaFold prediction into a good one.
— Tristan
Hi Tristan Croll,
I presume that you meant "match #2/H to #1/D" and "match #2/L to #1/C" instead?
Regards,
Phung Cheng Fei
... huh. Looking at it again, it seems to me that AlphaFold has predicted the antibody heavy chain in the light chain conformation and vice versa. If you force it to align (predicted heavy) to (experimental light) with "match #2/D to #1/H" then
all four antibody domains align reasonably well. Seems like a failure on AlphaFold's part (and one I haven't seen before).