dist_cutoff does not change output scores?

[DimaMolod]

Hello @dunbrack, and thanks for the very useful new metric and amazing tools! I was testing ipsae.py script with AlphaFold3 predictions and noticed that changing dist_cutoff argument doesn't change any of the scores (including ipSAE_d0chn / n0res). Sorry if my understanding is wrong, but shouldn't these scores depend on dist_cutoff via d0?
Here are the output tables I got for pae_cutoff of 20 and distance_cutoff of 0 and 100, respectively:

Chn1 Chn2  PAE Dist  Type   ipSAE    ipSAE_d0chn ipSAE_d0dom  ipTM_af  ipTM_d0chn     pDockQ     pDockQ2    LIS       n0res  n0chn  n0dom   d0res   d0chn   d0dom  nres1   nres2   dist1   dist2  Model
A    B     20   00   asym  0.003217    0.287639    0.049338    0.190    0.149263      0.1211     0.0094     0.0000      20   1175    118    1.00   11.23    4.01     96      22       0       0   O43439+Q9GZV8/O43439_and_Q9GZV8_model
B    A     20   00   asym  0.025211    0.521194    0.212523    0.190    0.141874      0.1211     0.0097     0.2281      65   1175    197    2.77   11.23    5.23    109      88       0       0   O43439+Q9GZV8/O43439_and_Q9GZV8_model
A    B     20   00   max   0.025211    0.521194    0.212523    0.190    0.149263      0.1211     0.0097     0.1141      65   1175    197    2.77   11.23    5.23     96     109       0       0   O43439+Q9GZV8/O43439_and_Q9GZV8_model


Chn1 Chn2  PAE Dist  Type   ipSAE    ipSAE_d0chn ipSAE_d0dom  ipTM_af  ipTM_d0chn     pDockQ     pDockQ2    LIS       n0res  n0chn  n0dom   d0res   d0chn   d0dom  nres1   nres2   dist1   dist2  Model
A    B     20   100  asym  0.003217    0.287639    0.049338    0.190    0.149263      0.1211     0.0094     0.0000      20   1175    118    1.00   11.23    4.01     96      22      96      22   O43439+Q9GZV8/O43439_and_Q9GZV8_model
B    A     20   100  asym  0.025211    0.521194    0.212523    0.190    0.141874      0.1211     0.0097     0.2281      65   1175    197    2.77   11.23    5.23    109      88     109      88   O43439+Q9GZV8/O43439_and_Q9GZV8_model
A    B     20   100  max   0.025211    0.521194    0.212523    0.190    0.149263      0.1211     0.0097     0.1141      65   1175    197    2.77   11.23    5.23     96     109      96     109   O43439+Q9GZV8/O43439_and_Q9GZV8_model

With dist_cutoff=0, I would have expected interface-related counts to collapse (or at least change), but they do not.
Is it an expected behaviour?
I also attach the full directory with predictions.

Thanks a lot again!

Dima

ipsae.zip

[dunbrack]

Hi Dima,

Happy to help. You're correct -- the dist_cutoff does not change the scores. The first cutoff (the PAE cutoff) is used to calculate the ipSAE scores and the nres1 and nres2 numbers reflect the number of residues in chain1 and chain2 respectively that have at least one residue pair better than the cutoff with the other chain.

The second cutoff (dist_cutoff) simply calculates how many residues in each chain pass the PAE_cutoff but also pass a CB-CB distance cutoff (so are the residues pairs are either in the interface or near it). These are the dist1 and dist2 numbers. It's helpful to know how big the interface is for the residues that have good PAE scores with the other chain. So a chain1 residue in dist1 has good PAE with a residue in chain2 that it is also within dist_cutoff of.

The reason is that there are often incidental interfaces between disordered regions and domains (or between domains) that are not actually part of the evolved interface. So calculating a score based on the interface is often misleading (which is why pDockQ and pDockQ2 do not work that well sometimes). And I think there is useful scoring information on the whole domain/domain interactions/scores, rather than just the interface.

Hope this helps. let me know if you have any questions.

Roland

[DimaMolod]

Thank you for the explanations and once again for your great tool! I have no more questions and going to close the issue now.

Best wishes,
Dima