Making sense of unstructured biological metadata

Abstract

Ribonucleic Acid sequencing (RNA-seq) has become a cornerstone of modern biological research, offering deep insights into gene expression and function. However, RNA-seq analysis is often challenged by the presence of unlabelled or unstructured tissue-type information in the metadata, making the annotation process both laborious and computationally intensive. Accurate tissue annotations are crucial for interpreting gene expression profiles, as they provide essential context for understanding the biological significance of the data. This paper aims to enhance the efficiency of metadata processing by introducing an improved annotation pipeline that leverages Generative Pre-trained Transformer (GPT) technology, specifically GPT-4o, to automate the annotation of tissue types in metadata. Preliminary results indicate that the automated approach significantly reduces the time and computational resources required while maintaining high accuracy (based on F1 score) in tissue-type annotations. This automated approach addresses key bottlenecks in metadata annotation, highlighting the potential of Natural Language Processing (NLP) tools in enhancing the process of RNA-seq analysis. Offering a solution for managing large volumes of RNA-seq metadata and serves as a proof-of-concept for large-scale annotation efforts in other types of biological data. This advancement not only streamlines the annotation process but also facilitates accelerated and more efficient biological research, paving the way for deeper insights into gene functions and their implications across various fields.