Text Mining for Neuroscience

Text mining, sometimes alternately referred to as text analytics, refers to the process of extracting high‐quality knowledge from the analysis of textual data. Text mining has wide variety of applications in areas such as biomedical science, news analysis, and homeland security. In this paper, we describe an approach and some relatively small‐scale experiments which apply text mining to neuroscience research literature to find novel associations among a diverse set of entities. Neuroscience is a discipline which encompasses an exceptionally wide range of experimental approaches and rapidly growing interest. This combination results in an overwhelmingly large and often diffuse literature which makes a comprehensive synthesis difficult. Understanding the relations or associations among the entities appearing in the literature not only improves the researchers current understanding of recent advances in their field, but also provides an important computational tool to formulate novel hypotheses and thereby assist in scientific discoveries. We describe a methodology to automatically mine the literature and form novel associations through direct analysis of published texts. The method first retrieves a set of documents from databases such as PubMed using a set of relevant domain terms. In the current study these terms yielded a set of documents ranging from 160,909 to 367,214 documents. Each document is then represented in a numerical vector form from which an Association Graph is computed which represents relationships between all pairs of domain terms, based on co‐occurrence. Association graphs can then be subjected to various graph theoretic algorithms such as transitive closure and cycle (circuit) detection to derive additional information, and can also be visually presented to a human researcher for understanding. In this paper, we present three relatively small‐scale problem‐specific case studies to demonstrate that such an approach is very successful in replicating a neuroscience expert’s mental model of object‐object associations entirely by means of text mining. These preliminary results provide the confidence that this type of text mining based research approach provides an extremely powerful tool to better understand the literature and drive novel discovery for the neuroscience community.