Fig. 8
Infrastructure of decision-tree based meta-predictor. Query miRNA:mRNA sequences are firstly fed into miRanda, miRDB, PITA, and TargetScan to get individual predictions. These individual predictions may be scored or unscored (null output). Based on the scored individual predictions, a specific module of decision tree based artificial neural networks (DANN) will be selected. For example, if only miRanda and miRDB have scored predictions, module DANN-2-1 will be selected. There are eleven modules in the pipeline, each module corresponds to one of the eleven datatsets and uses scores different predictors as follows, DANN-4: miRanda, miRDB, PITA, and TargetScan; DANN-3-1: miRanda, miRDB, and PITA; DANN-3-2: miRanda, miRDB, and TargetScan; DANN-3-3: miRanda, PITA, and TargetScan; DANN-3-4: miRDB, PITA, and TargetScan; DANN-2-1: miRanda and miRDB; DANN-2-2: miRanda and PITA; DANN-2-3: miRanda and TargetScan; DANN-2-4: miRDB and PITA; DANN-2-5: miRDB and TargetScan; DANN-2-6: PITA and TargetScan. “Y” above an arrow and “N” along an arrow represent “Yes” and “No”. “T/F” inside a circle stands for true (T) or false (F) prediction. “NT1” is the number of predictors that make true prediction using the 1st-level true threshold values, and so on so forth for NF1, NT2, NF2. “b1” and “b2” are the differences of the predictions score from their corresponding 1st-level threshold values. “c1” and “c2” are the differences of the predictions score from their corresponding 2nd-level threshold values. “dT2” and “dF2” are the Euclidean distances of prediction scores from their corresponding 2nd-level threshold values for true (T) predictions and false (F) predictions, respectively. The infrastructure of the 2-hidden-layer ANN is described in the text. There are in total eleven DANNs