<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-3105">
  <Title>Mining MEDLINE: Postulating a Beneficial Role for Curcumin Longa in Retinal Diseases</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Consider a bioscientist who is studying a particular disease. Assume that she is already well familiar with the pathophysiology and accepted therapeutic options for treating this condition and wishes to determine if there are other, yet unrecognized, substances that may have therapeutic potential. She begins by searching for documents on the disease mechanism(s) and related disorders. Very soon she finds herself immersed in a morass of pathways and possible directions that need to be further explored. It will come as no surprise if even our most determined user quickly becomes overwhelmed and discouraged. The challenge of searching for a novel therapeutic substance is at best like looking for the proverbial &amp;quot;needle in a haystack&amp;quot;. However, in reality the challenge is greater since there is no assurance that there indeed is a needle in the haystack. Consequently, the goal of text mining (also known as literature mining) systems and algorithms is to assist users find such needles, if these exist at all in the literature &amp;quot;haystacks&amp;quot; (Hearst 1999).</Paragraph>
    <Paragraph position="1"> In general, as shown in Figure 1, a user may start with any type of topic (A), be it a disease, a pharmacological substance, or a specific gene. As he navigates the literature and follows connections through appropriate intermediate topics (B1, B2 etc.), the user hopes to reach terminal topics (C1, C2 etc.) that are both relevant and novel, in the sense of shedding new information on topic (A). This text mining approach commonly referred to as 'open' discovery was pioneered by Swanson in the mid 80s. A classic example discovery is one where starting with Raynaud's disease (A) Swanson identified fish oils (C) as a substance that may have therapeutic potential (Swanson, 1986). Intermediate connections (B) such as 'blood viscosity', 'platelet aggregation' were observed. Swanson also proposed a variation called 'closed' discovery wherein starting with a pair of topics (A and C) one explores possible connections (B links) between them that are not yet recognized. In collaboration with Smalheiser, Swanson used his open and closed discovery methods on MEDLINE and proposed a number of hypotheses (eg. Swanson, 1990; Smalheiser &amp; Swanson1996a; Smalheiser &amp; Swanson1996b; Smalheiser &amp; Swanson1998). The hypotheses they proposed were subsequently corroborated in clinical studies.</Paragraph>
    <Paragraph position="2"> The text mining framework established by Swanson and Smalheiser has attracted the attention of several researchers (Gordon and Lindsay, 1996; Lindsay and Gordon, 1999; Weeber et al., 2001) besides us (Srinivasan, 2004). A key goal in these follow-up efforts has been to reduce the amount of manual effort and intervention required during the discovery process. In previous work  using algorithms for MEDLINE which we developed, we replicated the eight open and closed discoveries made by Swanson and Smalheiser. In comparison with other replication studies these algorithms were the most effective (Srinivasan, 2004). They also require the least amount of manual input and analyses. For example, in open discovery, our methods expect the user to specify only the type of B terms of interest. Following this our algorithm selects B terms automatically. In contrast the other methods rely more on user input for selecting B terms. Our current research demonstrates that our open discovery algorithm can be used to generate new hypotheses for disease treatment that could be tested. In particular, we apply our open discovery procedure to explore the therapeutic potential of curcumin/turmeric (Curcumin Longa) a dietary substance commonly used in Asia. We show that our automatic discovery algorithm identifies retinal diseases as the novel context for research on curcumin. We review genetic and biochemical evidence to indicate that curcumin may be beneficial for treating retinal diseases.</Paragraph>
    <Paragraph position="3"> We first describe our open discovery algorithm. Next we show its application with curcumin as the starting point (topic A). We then present an analysis of the curcumin - retinal diseases connection. The next section is on related research. The final section presents our conclusions and plans for the next phase of this research.</Paragraph>
  </Section>
class="xml-element"></Paper>