Biomedical Named Entity Recognition (NER) for Chemical-Protein Interactions

Drug discovery is a strenuous manual effort. Relationships between chemicals and diseases (chemical-disease relations) play an important role in drug discovery, biocuration and drug safety. Identifying the chemical-disease relationships is critical and manually curating them is expensive, time-consuming, and inefficient considering the growth of the biomedical literature over the years. Several attempts have been made to assist curation using text-mining systems including the automatic extraction of chemical-disease relations in the past with limited success. The goal of this research is to build a model that can extract chemicals and gene mentions in natural language and further predict meaningful relationships between them. We have built a Bi-LSTM with a CRF layer to extract entities from biomedical text with 90% accuracy. The model can be used in the early stages of vaccine/drug development by the scientific community. This can help speed up the development process and reduce labor costs.