Negative Control - HADDOCK3+PRODIGY

PRODIGY
1

Hi there! I have successfully run HADDOCK3 for a few antibodies and antigen. As a follow up to the docking results, I also ran PRODIGY to assess Kd and free energy of the complexes. So far, all of my complexes have acceptable binding affinity (negative free energy, small Kd) and reasonable docking metrics (iRMSD, DOCKQ, etc.).

I am interested in running some sort of computational validation. I tried a negative control (used an unrelated antibody) and kept the same antigen PDB. I followed all steps strictly similarly to how I did with my positive control. This methodology however also generated acceptable docking metrics and binding affinity.

Could you advise on best practices to validate HADDOCK3 + PRODIGY results when explicit non-interactors are unavailable? Ideally I wanted to have a control group to compare it with.

I appreciate your time and help!

2

Hi there

Note that docking software were never developed to predict if two proteins interact, but rather, given the knowledge they interact, how they interact.
As such scores are not representative of bonding affinities. Docking software will always produce some model of the interacting molecules.

As for PRODIGY, it was trained on crystal structures for which binding affinity data were known. No negative data included.
For this I would rather first use some classifier methods (binding or non-binding) and then for the binding ones, predict the interaction.
As such you finding are not surprising to me.

Some people have used instead PRODIGY-CRYSTAL to predict if two proteins might interact (not benchmarking on our side).

Some relevant old papers on this topic:

And a more recent one:

  • H. Schweke, Q. Xu, G. Tauriello, L. Pantolini, T. Schwede, F. Cazals, A. Lhéritier, J. Fernandez-Recio, L.A. Rodríguez-Lumbreras, O. Schueler-Furman, J.K. Varga, B. Jiménez-García, M.F. Réau, A.M.J.J. Bonvin, C. Savojardo, P.-L. Martelli, R. Casadio, J. Tubiana, H. Wolfson, R. Oliva, D. Barradas-Bautista, T. Ricciardelli, L. Cavallo, Č. Venclovas, K. Olechnovič, R. Guerois, J. Andreani, J. Martin, X. Wang, D. Kihara , A. Marchand, B. Correia, X. Zou, S. Dey, R. Dunbrack, E. Levy, S. Wodak. Discriminating physiological from non-physiological interfaces in structures of protein complexes: a community-wide study. Proteomics 23, 1-21 (2023).