Protein Contact Prediction: Pseudolikelihoods and Potts Models

This lecture introduces the problem of predicting contacts in proteins based on coevolution of amino acids. It presents a new approach to Direct-Coupling Analysis (DCA) using Potts models. The Potts model is explained, along with the inverse Potts problem and the regularization techniques used to avoid overfitting. The lecture covers the pseudolikelihood method for maximizing the likelihood of the Potts model. Sequence reweighting and interaction scores are discussed, as well as the comparison between mean-field DCA and pseudolikelihood DCA. The lecture concludes with a comparison of different methods for contact prediction in proteins.