Poster A136, Saturday, March 25, 5:00 – 7:00 pm, Pacific Concourse
Automated meta-analysis of event-related potentials and their correlates through text-mining
Thomas Donoghue1, Bradley Voytek1; 1University of California, San Diego
Event-related potentials (ERPs) have a rich history as a method to investigate the neural basis of cognition. Given the vastness of the literature, with over 400 000 ERP papers on Pubmed, there is the need for a systematic way to summarize and analyze the current status of ERP research. Here we present an automated text-mining approach, using the Pubmed e-utilities as a form of meta-analysis to examine the relationship between ERP terms, cognitive domains and disease states. We curated dictionaries of terms, including over 30 previously described ERP components, and determined co-occurrence probabilities in published papers between ERP components and cognitive and disease terms to investigate what different ERP components are associated with. We also extracted all content words from articles found using the same ERP dictionary, allowing us to build a data-driven profile for each ERP, including the terms with which they are most affiliated, and a topic modeling of the words used when discussing them. This database can be used to confirm and quantify known associations, such as how early ERPs, like the N100, typically relate to sensory and attentional processes, whereas later components, such as the N400, are associated with cognitive processes such as semantics. This data has been combined into an easily searchable database, allowing for efficient look-up of the ERP profiles, efficiently summarizing a large body of research. This database can be used both as a learning and teaching tool, and as a method of inquiry into the previously hidden structure of the existing literature.
Topic Area: OTHER