Validation pipeline for computational prediction of genomics annotation

Controlled biomolecular annotations are key concepts in computational genomics and proteomics, since they can describe the functional features of genes and their products in both a simple and computational way. Despite the importance of these annotations, many of them are missing, and the available ones contain errors and inconsistencies; furthermore, the discovery and validation of new annotations are very time-consuming tasks. For these reasons, recently many computer scientists developed several machine-learning algorithms able to computationally predict new gene-function relationships. While several of these methods have been easily adapted from different domains to bioinformatics, their validation remains a challenging aspect of a computational pipeline. Here, we propose a validation procedure based upon three different sub-phases, which is able to assess the precision of any algorithm predictions with a reliable degree of accuracy. We show some validation results obtained for Gene Ontology annotations of Homo sapiens genes that demonstrate the effectiveness of our validation approach.