Proper integration of feature subsets boosts GO subcellular localization predictions

FLAVIO E. SPETALE; TAPIA ELIZABETH; MURILLO JAVIER; KRSTICEVIC FLAVIA

BIOINGENIERIA Y FISICA MEDICA CUBANA

Sociedad Argentina de Bioingeniería

Lugar: San Miguel de Tucumán; Año: 2018 vol. 22 p. 3 - 6

1606-0563

Prediction of multiple subcellular localizations in proteins brings relevant information for biological function discovery. The use of computational methods based on knowledge can be a helpful starting point forguiding the costly experimental validation. In this work, we present a multilabel classifier framework to perform Gene Ontology - Cellular Component prediction focused on the improvement of two design aspects: i) the protein sequence characterization, regarding biological knowledge with experimental evidence, and ii) the error evaluation by considering a noise model inherent in real prediction frameworks. Our proposal is validated against sets of well-known protein sequences of four model organisms D. rerio, A. thaliana, S. cerevisiae and D. melanogaster