How to get a IPSAE score for protein complexes?

[susiegriggo]

Thanks for your work on this and the easy-to-use script!

I want to compute a single IPSAE score for homomeric protein complexes; however, I notice that the ipsae scores are either per residue or comparing pairs of chains. Would it be possible to do something similar to equation 13 in the preprint with the per-residue ipsae scores to get a score for an entire complex?

Alternatively, I thought it might be possible to get a score for a complex by taking the maximum per-chain ipsae for each chain in a complex and then averaging them. This here assumes that the maximum ipsae score for each chain is the one it is most connected to in the complex so would encourage all chains to to have an interface with at least one other chain.

[dunbrack]

Hi Susie,

I'll have to think about that. It's a question of what the d0 value should be. There could be a single value (for the aligned residue i, the total number of residues with respect to that chain that have PAE<cutoff). For residue i, the d0 could vary by the chain of residue j (the number of residues that have PAE<cutoff for that chain).

It depends on what a single ipSAE (or ipTM) score should convey. I think it should convey that all the chains are connected by at least one link to the other chains. If there is a spare chain that is not connected (by an interface that is well-predicted by AlphaFold), then the whole complex should get a poor score.

My code outputs the number of residues that are in each pairwise chain-chain interface within the input distance cutoff AND the input PAE cutoff (it had to be both, because some interfaces only contain poorly predicted PAE scores; and even some chains not in contact have good PAE scores, e.g. in a homomultimer). These are listed under the dist1 and dist2 columns for each pair of chains. The quality of the complex might be calculated as:

1. for each chain, pick the maximum ipSAE score for which dist1 and dist2 are >0
2. make an edge for each of these connections
3. make sure the graph is fully connected (every chain has at least one edge)
4. take the minimum over all these ipSAE scores.

But I will think about this some more before adding to the code.

Happy to hear your thoughts on this.

Roland

[Xingyue-Feng]

Hi,

Thanks for sharing. It's very helpful to read.

I just wanted to add a small observation from my side: in some complexes I've checked, not every chain pair ends up being fully connected under the distance and PAE cutoffs. So incomplete connectivity does happen.

In such situations, what would be the recommended way to handle these partially connected complexes?

Thank you,
Xingyue