Uncovering the Effects of Genes, Proteins, and Medications on Functions of Wound Healing: A Dependency Rule-Based Text Mining Approach Leveraging GPT-4

Jayati Halder Jui; Milos Hauskrecht

Abstract

Wound healing is a complex biological process characterized by intricate cellular and molecular interactions. Understanding the underlying mechanisms and the effects of different biological entities, such as genes, proteins, and medications, on the cellular and biological functions of wound healing is of paramount importance for developing effective therapeutic interventions. In this paper, we present a text-mining approach aimed to explore and unravel the complex regulatory relationships of genes, proteins, and medications with the biological mechanisms of wound healing. Our approach relies on a set of predefined dependency rules to identify and capture the relationships between biological entities and their target functions. By leveraging advanced AI technology like Generative Pre-trained Transformer 4 (GPT-4), also known as ChatGPT, we evaluate the accuracy and quality of the extracted relations. We also present a thorough discussion about the encouraging preliminary results that validate the efficacy of our model. Our dependency rule-based text-mining approach, combined with the capabilities of GPT-4, presents a promising avenue for unraveling the complex web of interactions involved in wound healing. The study underscores the future potential of incorporating multi-word concept embedding of complex functional entities and exploiting synthetic data from GPT-4 for enhanced relation identification. This research offers a new contribution to aid computational biology research by exploiting the power of large language models to facilitate biological text analysis.

Keywords: Wound Healing; Relation Extraction; GPT-4; Medline; Biological Function

Links
[Full text PDF][Bibtex]