The average product correction (APC) and entropy bias correction (EntC) term correlate strongly.

A For protein 1c9oA in the PSICOV data set the correction defined by the APC correlates well with the correction determined by our entropy bias correction strategy when learning a Markov random field (MRF) model with persistent contrastive divergence (PCD). B Also the factors appearing in the APC and EntC corrections, and ., correlate well. C,D The Pearson correlation is similarly large when the MRF model is learnt from synthetic alignments generated with CCMgen using binary tree topologies (C) or using star tree topologies (D), both in the example of protein 1c9oA shown here and also across all 150 proteins in the Pfam dataset (S4 Fig).