search

Rappsilber-Laboratory / AlphaLink2

AlphaLink2: Integrating crosslinking MS data into Uni-Fold-Multimer
Creative Commons Attribution 4.0 International
50 stars 16 forks source link

distance for crosslinks #14

Closed Samuel-gwb closed 11 months ago

Samuel-gwb commented 1 year ago

Great work ! I have a question that what is the distance for a crosslinking input like following: 5 A 5 A 0.1 Will it be 25 Ang ? How to define the distance for such a crosslinking data due to different crosslinking reagent?

grandrea commented 12 months ago

hello, alphalink2 is currently trained with 25 angstrom data. So that line indicates a residue pair between residue 5 of chain A and residue 5 of chain A with a false discovery rate of 0.1 (10%). The network is trained to consider this residue pair within 25 angstrom. Currently, the distance cannot be changed, but it is not a distance restraint in the traditional sense- it is the maximum crosslinker distance used in training the network.

Others have already used this with different crosslinkers, as AL2 weighs both experimental and MSA evidence, and indication of proximity is often enough to drive the prediction to the correct state.

disclaimer: I am not the AL2 developer, i was involved in developing AL1.

Samuel-gwb commented 12 months ago

Thanks for your clearance! I found an option in inference.py (not mentioned in README), --cutoff, with default 25. Does this refer to the 25 ang data? Will revision of the default 25 to 35, i.e., be more reasonable for dss crosslinking? Anyway, I agree that indication of proximity would be often enough.

grandrea commented 12 months ago

I have been using AL2 for bs3 and dsso with ok results (the lower restraint density makes things harder). I also would say that 35 is not a generally agreed cutoff for bs3 CA-CA, groups work with anything between 22 to 35 depending how much backbone flexibility is thought to play a role...

I don't know about the parser specification for cutoff, though i know the cutoff in the colab is about plotting crosslinks on structures only.

Samuel-gwb commented 12 months ago

Ok, I see. Thanks very much for your sharing!

lhatsk commented 12 months ago

The cutoff is just for the calculation of the crosslink satisfaction. It's by default 25 A because that's what the network was trained on. We also have another network which was trained on photoAA crosslinks (10 A).

Samuel-gwb commented 12 months ago

The cutoff is just for the calculation of the crosslink satisfaction. It's by default 25 A because that's what the network was trained on. We also have another network which was trained on photoAA crosslinks (10 A).

Yes! I've checked with different cutoff, 25 A, 35 A, 45 A. The resulted structure is simillar while output satisfaction ratio gradually increases. Thanks!