The primary objective of this team is to develop a method for computers to extract information from published scientific works and populate an easily searchable database (Markerville) with this information. Using the natural language processing (NLP) and machine learning elements of DeepDive, this program will programmatically “read” through research papers.
The initial version of the software will simply extract biomarkers along with the diseases they are frequently associated with and will be limited to sentences with simple structures. As further development occurs however, the software will also collect information regarding the medium in which the biomarker is found, the method of data collection, drugs in which the biomarker is used, the type of biomarker, the stage of the disease during which a biomarker is present, etc. Following this, data about the clinical trials themselves (number of trials, number of patients sampled, levels of the biomarkers in the patients, and other patient background information) will be collected. By repeatedly training the software using DeepDive’s machine learning features, the accuracy of the software will be maximized.
The goal is to create a tool that can analyze thousands of papers and extract data vital to the accurate and efficient identification of biomarkers. This tool can then be incorporated into Markerville, thus making the process of manually input biomarker and patient data into Markerville completely automated.