Might be a dead-end, might be pointless, but is something that could be interesting to try.
Update: Jan Aerts suggested trying out a minimum spanning tree (MST) to start instead of Mapper. We can also try leiden clustering: https://github.com/vtraag/leidenalg
The two dimensions are the geometrical distance between residues (alpha carbons) and the correlation of movement: https://docs.scikit-tda.org/en/latest/libraries.html
Might be a dead-end, might be pointless, but is something that could be interesting to try.
Update: Jan Aerts suggested trying out a minimum spanning tree (MST) to start instead of Mapper. We can also try leiden clustering: https://github.com/vtraag/leidenalg