Amine Heddad



Evolving Classifiers for Protein Nuclear Localisation Using Genetic Programming

Abstract

Being able to predict the location of a protein in the cell is one of the steps toward knowing its role and activity. With that information one could also conclude possible effects on the organism carrying those proteins. The number of putative, unclassified proteins is constantly growing due to the continuous genome sequencing projects. Hence, the need for fast and cheap methods to classify proteins beside the classical in vivo/vitro experimental approaches is more actual than ever. Here we present a system of classifiers to predict if a protein is nuclear or not. Genetic programming was used to evolve classifiers according to three grammatical strategies with increased complexity. The classifiers where then combined using majority voting into a final classifying system called NucPred and made available as a free web service.


Artificiell evolution av proteinklassificerare med genetisk

programmering

Sammanfattning

Att veta var proteiner befinner sig inom cellen är viktigt för att förstå deras roll. Med denna kunskap kan man även dra slutsatser om proteinens roll i den bärande organismen. Här presenteras ett klassificerings system för nukleära proteiner. Genetisk programmering med hjälp av PerlGP har använts för att evolvera flera klassificerare enligt tre grammatiska strategier. Vi har sedan kombinerat dessa klassificerare till en slutlig klassificerings system kallad NucPred.