| Popis: |
Computational modeling of protein structures that adopt multiple conformations is a challenging task. We propose a procedure for generalizing the knowledge gained from AlphaFold2 models to represent such conformational landscapes. The procedure was applied to the HAMP domain, a short helical bundle that transduces the signal from sensors to effectors in two-component signaling proteins such as EnvZ or Tar. The results obtained correctly reproduce the experimentally confirmed conformations and support the universality of the mechanism by which HAMP domains transduce signal by axial rotation of their helices. Furthermore, the AlphaFold2 models were used to train a deep learning tool, HAMPpred, which can predict the resting conformation of a given HAMP domain within seconds. Using HAMPpred, we searched 60,000 HAMP domains for cases where two HAMP sequences differing by one amino acid showed a significant change in predicted conformation. We found that these likely adaptive mutations occur in proteins involved in processes such as antibiotic production and cell wall damage sensing, or in proteins containing rapidly evolving poly-HAMP arrays. All code associated with this work is deposited athttps://github.com/labstructbioinf/HAMPpred. |
