ipSAE

[dunbrack]

Hello,

Would you be interested in implementing the ipSAE score in AF-cache? It is much better for PPI screens than ipTM. ipTM gives misleading results for some sequence pairs. ipTM can be too low when both proteins contain large amounts of non-binding disorder or folded domains (because residues are aligned on one chain followed by scoring the entirely of the other chain, including irrelevant non-binding regions). It can be too high when one protein is well folded but the other has lots of non-binding sequence (because the d0 parameter is calculated from the sum of the lengths of the two proteins; the more you lengthen one of the proteins, the ipTM score goes UP for no useful reason).

Let me know if you need help in implementing it. It's on github: https://github.com/DunbrackLab/IPSAE.

Thanks,
Roland

[arneelof]

Are we allowed to modify the AF3 code? I thought the license did not allow that.

[clami66]

Hi @dunbrack thanks for getting in touch, I think it would be great to include ipSAE

@arneelof the AF3 code should be fine, but I think we could just pull @dunbrack 's code and call the script from nextflow on the outputs as a post-processing step

@SarahND97 do you think you could take care of this?

[dunbrack]

Hi @dunbrack thanks for getting in touch, I think it would be great to include ipSAE

@arneelof the AF3 code should be fine, but I think we could just pull @dunbrack 's code and call the script from nextflow on the outputs as a post-processing step

@SarahND97 do you think you could take care of this?

That would be great. Let me know if you have any problems. The code could be simplified and made faster. It calculates other scores as well which could be removed.

[SarahND97]

Yes, this sounds like a good idea to implement! We can add an argument called run_ipsae or similar that if true will run ipSAE as a post-processing step. I will close this issue once it's implemented and tested.