Analysis of Novel Annotations in the Gene Ontology for Boosting the Selection of Negative Examples

Public repositories for genome and proteome annotations, such asthe Gene Ontology (GO), rarely stores negative annotations, i.e.proteins not possessing a given function. This leaves undefined orill defined the set of negative examples, which is crucial for trainingthe majority of machine learning methods inferring proteins func-tions. Automated techniques to choose reliable negative proteinsare thereby required to train accurate function prediction models.This study proposes the first extensive analysis of the temporalevolution of protein annotations in the GO repository. Novel an-notations registered through the years have been analyzed to ver-ify the presence of annotation patterns in the GO hierarchy. Ourresearch supplied fundamental clues about proteins likely to be un-reliable as negative examples, that has been verified into a novelalgorithm of our own construction, validated on two organismsin a genome wide fashion against approaches proposed to choosenegative examples in the context of functional prediction.