COUSCOus

Contact prediction using shrinked covariance (COUSCOus). COUSCOus is a residue-residue contact detecting method approaching the contact inference using the glassofast implementation of Matyas and Sustik (2012, The University of Texas at Austin UTCS Technical Report 2012:1-3. TR-12-29.) that solves the L_1 regularised Gaussian maximum likelihood estimation of the inverse of a covariance matrix. Prior to the inverse covariance matrix estimation we utilise a covariance matrix shrinkage approach, the empirical Bayes covariance estimator, which has been shown by Haff (1980) doi:10.1214/aos/1176345010 to be the best estimator in a Bayesian framework, especially dominating estimators of the form aS, such as the smoothed covariance estimator applied in a related contact inference technique PSICOV.