Prediction of human functional genetic networks from heterogeneous data using RVM-based ensemble learning

Abstract

Motivation: Three major problems confront the construction of a human genetic network from heterogeneous genomics data using kernel-based approaches: definition of a robust gold-standard negative set, large-scale learning and massive missing data values.

Results: The proposed graph-based approach generates a robust GSN for the training process of genetic network construction. The RVM-based ensemble model that combines AdaBoost and reduced-feature yields improved performance on large-scale learning problems with massive missing values in comparison to Naïve Bayes.

Contact: dargenio@bmsr.usc.edu

Supplementary information: Supplementary material is available at Bioinformatics online.

Fig. 1.

Two genes are linked together with solid lines if they function together in the present KEGG pathways (positive examples) while those connected with dotted lines do not function together. (a) illustrates potential positive examples. It would be more likely that genes 7 and 2 function together in some unknown pathways than genes 1 and 2, because the former pair shares more pathway partners. The width of the dotted line reflects the probability that a linkage exists. (b) Illustrates ideal negative examples. It would be less likely for genes 2 and 5 to function together in some unknown pathways than genes 2 and 3 or genes 2 and 4.

Fig. 2.

(a) RVM-AdaBoost. (b) RVM-based double ensemble model.

Fig. 3.

(a) Proportion of gene pairs with specific topologic distances in the old KEGG network that function together in the new KEGG pathways. (b) GO scores of gene pairs with specific topologic distances in the old KEGG network.

Fig. 4.

Precision–recall curves of models with different kernel combinations based on 10-fold cross-validation testing.

Fig. 5.

Precision–recall curves of models with different ways for dealing with missing values based 10-fold cross-validation testing.