Uma ferramenta multiagente baseada em conhecimento para anotação de proteínas : um estudo de caso para o Fungo Saccharomyces cerevisiae

Identifying biological function of sequences is a key activity in genome projects. This task is done in the annotation step, which has two phases. In the manual phase, biologists use their knowledge and experience to determine the function for each sequence, based on the results produced by the automatic phase, where tools and data bases are used to predict functional annotation. This dissertation presents BioAgents-Prot, a knowledge based multiagent tool, which simulates biologists expertise to annotate proteins. BioAgents-Prot is defined with an approach of cooperative agents, where specialized intelligent agents work together to suggest proper manual annotation. The proposed three-layer architecture was implemented with Java Agent DEvelopment Framework-JADE and Drools (a rule-based inference engine). To assess performance, transcript annotations of the Saccharomyces cerevisiae fungus were compared to the annotations suggested by BioAgents-Prot. Using basic rules that represents the annotation reasoning, we obtained 95.84% of sensitivity, 93.22% of specificity, 98.40% of F1-score and 0.80 of MCC, which shows the usefulness of BioAgents-Prot in annotation step of transcriptome projects.