Abstract
Direct-coupling analysis is a statistical learning method for protein contact prediction based on sequence information alone. The maximum entropy principle leads to an effective inverse Potts model. Predictions on contacts are based on fitted local fields and couplings from an empirical multiple sequence alignment. Typically, the norm of the resulting two-body couplings is used for contact prediction. However, this procedure discards important information. In this paper we show that the usage of the full fields and coupling information improves prediction accuracy.
- Received 13 September 2020
- Revised 7 February 2021
- Accepted 27 March 2021
DOI:https://doi.org/10.1103/PhysRevE.103.042418
©2021 American Physical Society