PlantConnectome: knowledge networks encompassing >100,000 plant article abstracts

Abstract

AbstractPredicting gene function is indispensable to our understanding of biology. However, these predictions hinge on large collections of experimentally characterized genes, the compilation of which is not only labor-intensive and time-consuming but rendered near-impossible given the volume and diversity of scientific literature. Here, we tackle this challenge by deploying the text-mining capacities of Generative Pre-trained Transformer (GPT) to process over 100,000 plant biology abstracts. Our approach unveiled nearly 400,000 functional relationships between a wide array of biological entities—genes, metabolites, tissues, and others—with a remarkable accuracy of over 85%. We encapsulated these findings in PlantConnectome, a user-friendly database, and demonstrated its diverse utility by providing insights into gene regulatory networks, protein-protein interactions, as well as developmental and stress responses. We believe that this innovative use of AI in the life sciences will significantly accelerate and direct research, drive powerful gene function prediction methods and help us keep up to date with the rapidly growing corpus of scientific literature.