<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3303">
  <Title>Using the Gene Ontology for Subcellular Localization Prediction</Title>
  <Section position="6" start_page="22" end_page="23" type="concl">
    <SectionTitle>
5 Conclusion and Future work
</SectionTitle>
    <Paragraph position="0"> Our study has shown that using an external information source is beneficial when processing abstracts from biological journals. The GO can be used as a reference for both synonym resolution and term generalization for document classification and doing so significantly increases the F-measure of most subcellular localization classifiers for animal proteins.</Paragraph>
    <Paragraph position="1"> On average, our improvements are modest, but they indicate that further exploration of this technique is warranted.</Paragraph>
    <Paragraph position="2"> We are currently repeating our experiments for PA's other subcellular data sets and for function prediction. Though our previous work with PA is not  text based, our experience training protein classifiers has led us to believe that a technique that works well for one protein property often succeeds for others as well. For example our general function classifier has F-measure within one percent of the F-measure of our Animal subcellular classifier. Although we test the technique presented here on subcellular localization only, we see no reason why it could not be used to predict any protein property (general function, tissue specificity, relation to disease, etc.). Finally, although our results apply to text classification for molecular biology, the principle of using an ontology that encodes synonyms and hierarchical relationships may be applicable to other applications with domain specific terminology.</Paragraph>
    <Paragraph position="3"> The Data Sets used in these experiments are available at http://www.cs.ualberta.ca/ ~alona/bioNLP/.</Paragraph>
  </Section>
class="xml-element"></Paper>