GO Function prediction by True Path Rule

SPETALE, FLAVIO; BULACIO, PILAR; ANGELONE, LAURA; TAPIA, ELIZABETH

Oro Verde, Entre Ríos

Congreso; III Congreso Argentino de Bioinformática y Biología Computacional; 2012

Facultad de Ingeniería de la Universidad Nacional de Entre Ríos (FIUNER) y la Asociación Argentina de Bioinformática y Biología Computacional (A2B2C).

Protein function prediction is an important problem in bioinformatics research. Useful tools have been developed to identify similar sequences regarding their corresponding annotation database. But when no similar sequences can be found, carefully designed data mining techniques may provide an important clue to protein function prediction. In particular, hierarchical classification methods like the True Path Rule (TPR) can take into account the relationships among protein functions defined on Gene Ontology (GO). This GO structure influences in two points: i) In the design of the training datasets for machine-learning classifiers, and ii) In the global function prediction since a sequence may belong to multiple classes. Fig.1 and Fig.2 show a simplified TPR analysis on GO with Arabidopsis data. Consensus probability p′ represents the membership of a given sample to GO nodes. Favourable probabilities p′ after TPR are shown in blue. The starting point to apply TPR is the set of local probabilities p. Favourable probabilities p at the beginning of TPR are shown in bold.