As our next test case we take an ABC transporter responsible for development of multidrug resistance was analyzed through a Bortezomib mutational scan of transmembrane domain 11 of mouse orrthologue, by Hannah et al. The related groups of orthologues used are ABCB4 and ABCB5. Compared to the LacI case, the size of the study was small. Both TP and TN sets might be incomplete here. However comparing the ability of different functions to pick up the confirmed true positives from confirmed true negatives shows the ability of determinant model to enrich the top scoring portion of the residues with confirmed TP cases. The E. coli methyltransferase RsmC was studied by Sunita et al. Charged residues, demonstrated therein through alanine mutagenesis to be involved in catalysis, are used as the true positive set. The paralogous family consists of bacterial RlmG proteins, with different substrate specificity. The nonspecific residues were not explicitly tested in the study. The sequences used in the alignments, as well as the set of functional residues can be found in Materials S1. Residues conserved across all groups were never considered to be a part of “positive” set of specificity conferring residues. In all cases the performance of related earlier methods GroupSim, SPEER, and SDP is shown on the same graph. These methods have on their own been successfully compared with other, earlier approaches. GroupSim, uses Jensen-Shannon divergence, Eq. 6, as the conservation, and squared difference, Eq. 10, as an overlap measure, combined linearly into a single score. The two quantities are not scaled to ½0,1 interval as we do here, and additional conservation filter is imposed on the neighboring residues. SDPpred is an elaboration on the mutual information approach, Eq. 13, that additionally estimates the statistical significance of the assigned score. The exchangeability of the residue types is incorporated into the significance calculation. SPEER uses rate4site, a phylogeny based method that on its own uses exchangeability in estimating prior mutational probabilities, to estimate difference in evolutionary rates among groups, and linearly combines it with Euclidean distances based on amino acids’ physico-chemical properties, and Kullback-Leibler, Eq. 5, type of conservation score. All implementations were used with their default choice of parameters. The problem that is encountered in discussion of these methods is their compounding of conservation and overlap measures, and at times fuzzy correction for residue type exchangeability, all of which make difficult tracing the sources of their failure and success alike. In Fig. 2 we show one particular choice of conservation and overlap methods discussed in the Methods section. However, other choices are possible, and indeed perform on the level within the noise bracket of the data. This is illustrated in Fig. 3, for the LacI test case. The remaining cases are relegated to supporting material. In the figure, all possible scores that can be obtained by combining the scoring and residue conservation – from literature, as well as proposed here – are listed on the x-axis in the order of decreasing area under the ROC curve.