Skip to main content

Table 2 Experimental results on the DNA-protein sequence data set. Experimental results with Naive Bayes (NB) and Logistic Regression (LR) models, and Mixture of Experts (ME) models on the non-redundant DNA-protein sequence data set, where the identity cutoffs are 30% and 90%. The results are shown for default threshold θ = 0.5. ME-NB-global and ME-LR-global use NB and LR at the leaves and exploits the global sequence similarity to construct the hierarchical structure. ME-NB-local exploits the local sequence similarity to construct the hierarchical structure. ME-NB-random randomizes the global similarity matrix and constructs the hierarchical structure based on the randomized matrix.

From: Mixture of experts models to exploit global sequence similarity on biomolecular sequence labeling

Classifier DNA-protein 30% DNA-protein 90%
Precision Recall CC FM AUC Precision Recall CC FM AUC
NB 0.59 0.05 0.16 0.10 0.75 0.56 0.07 0.18 0.13 0.75
ME-NB-global 0.62 0.12 0.25 0.20 0.77 0.65 0.15 0.29 0.25 0.78
ME-NB-local 0.65 0.06 0.18 0.12 0.76 0.64 0.08 0.21 0.15 0.76
ME-NB-random 0.58 0.05 0.15 0.09 0.75 0.56 0.07 0.18 0.13 0.75
LR 0.57 0.07 0.18 0.12 0.79 0.57 0.08 0.18 0.14 0.79
ME-LR-global 0.57 0.14 0.26 0.23 0.80 0.63 0.17 0.29 0.26 0.81