Fig. 3

Learning of RF00059 at equilibrium. a The evolution over the iterations of \(q_{ext}\), \(q_{int1}\) and \(q_{int2}\) used to tune the waiting time \(T_{wait}\). b We plot the fitting errors (red, orange and brown markers) \(\varepsilon _{c}\), \(\varepsilon _{f}\) and \(\varepsilon _{s}\), and the Pearson correlation coefficient between the two-site connected statistics of the natural sequences and of the configurations sampled during training (blue markers). c Depicts the projections of the natural sequences into the space of the first two principal components (PC1, PC2) of the covariance matrix associated with the natural sequences; in d we show the projection the re-sampled configurations, obtained from the converged model, into PC1 and PC2 associated with the natural sequences. e We show the behavior of the PPV versus the number of non-trivial contact predictions, i.e. those associated with site indices \(|i-j|>4\), for adabmDCA, plmDCA [33] and bl-dca [34]. f Displays the contact maps used as ground truth (gray markers) for the TPP riboswitch and those obtained by the DCA scores larger than 0.20 associated with the three compared methods