In order to predict GO annotations of OMA orthologous groups, we employed the machine learning algorithm CLUS-HMC, based on decision trees for Hierarchical Multi-label Classification. The CLUS-HMC  decision trees are able to deal with multiple, hierarchically organized class labels. In our case, these are GO annotations of genes.

Clus-HMC can be extended to an ensemble (CLUS-HMC-Ens); in particular, we use a Random Forest - a large number of decision trees constructed from bootstrap samples of the data. An implementation of the Clus-HMC is available.

We compared the predictive performance of the CLUS-HMC Random Forest to a standard approach in analysis of gene functional interactions.

In particular, this means inferring the function of a gene by looking at its closest nodes in a network - here, the genes with most similar phylogenetic profiles. This is known as the k-nearest neighbors algorithm. The CLUS-HMC random forest considerably outperforms this baseline approach (see below).