Figure 1
From: Consistent probabilistic outputs for protein function prediction
Overall approach. (1) A kernel matrix is computed for each available data type, and (2) these kernels are used to train one support vector machine (SVM) for each term and each data type. (3) The SVM predictions are combined and calibrated via a collection of logistic regressions. (4) Finally, the calibrated predictions are reconciled with respect to the Gene Ontology (GO) topology.