Fig. 7

Naive Bayes classifier analysis of Pfam domains. a Naive Bayes classifier analysis of the training set evaluating thresholds for positive (cyan dots) and negative (orange dots) hits discrimination of the Pfam domain “bacteriorhodopsin-like protein” (PF01036). The space of positive hits is coloured blue, and the space of negative hits is the complementary one, coloured from green to dark red (see colour scaling). From dark red to blue, each coloured area is characterised by a different probability for a sequence to be a positive sequence. The space of solutions for all CCMs confounded (left) and the SCM (pHMM; right) are described. The red dots correspond to the MG sequences identified by the models in the Equatorial Pacific (EPAC) MT dataset discussed below [57]. The green vertical line on the SCM plot (right) corresponds to hmmscan GA threshold (= 24, for this domain). Note that the grid is discrete and that rectangular regions are coloured with respect to probability intervals. This means that two adjacent regions with the same colour might have associated two different probabilities. b The probability space of CCMs and SCMs generated for MetaCLADE; all domains included. Each point of the plot is the average of the corresponding points in all spaces of solutions computed for CCMs and SCMs, respectively. Examples of such spaces of solutions are reported in a for the Pfam domain PF01036. The colour scale is defined with respect to probability values associated to the regions. The green vertical line in the SCM plot (right) corresponds to the average hmmscan GA threshold; all domain confounded. Note that if we take the bit-score only as a threshold in our database, we obtain that an average bit-score at 25 in a is statistically meaningful, independently of the mean-bit-score. This value is used as a default threshold in http://www.ebi.ac.uk/Tools/hmmer/search/phmmer