Reconstruction of gene-protein-reaction associations using biological big data and deep learning

Abstract

The field of biological sciences has seen a significant increase in the number of published papers, which serve as a crucial source for novel discoveries. However, despite the availability of advanced search engines, efficient gathering and processing of newly reported data from the vast collection of literature has become increasingly challenging. A critical area of information that requires systematic extraction from literature is gene-protein-reaction (GPR) associations. The availability of GPR associations plays an instrumental role in studying the connection between an organism’s genetic makeup and its observable characteristics. Additionally, they enable the development of computational models such as genome-scale metabolic models. This study introduces a Python-based text-mining framework, which facilitates the efficient and systematic extraction of GPR association information from literature. The system employs multiple deep learning-based language models, namely BioBERT, PubMedBERT, and BioGPT, to retrieve data on five entities: species, genes, proteins, chemicals, and metabolites. The extracted GPR associations are subsequently reconstructed in a Boolean logic. The text mining framework developed in this study holds promise in enhancing the efficient and comprehensive collection of biological information (i.e., GPR associations) from an extensive corpus of literature.