Numbers and fraction of proteins where gplmDCA performs better than plmDCA.

In each row all proteins in the data set are included for which the PPV from both plmDCA and gplmDCA is larger than the cutoff value given in the first column. The full data set (last row) consists of 729 proteins for 522 (72%) of which gplmDCA performs better than plmDCA. In the most stringent selection (first row) there are 128 proteins where both plmDCA, plmDCA20 and gplmDCA have a PPV of at least 0.5. In this set gplmDCA performs better on 109 (85%) of the instances. By the same criteria, plmDCA20 performs slightly better than gplmDCA, outperforming plmDCA for 579 proteins (79% of all) and performing better in 117 cases (91%) out of 128 proteins highly amenable to contact prediction by these methods.Numbers and fraction of proteins where gplmDCA performs better than plmDCA.