Mapping Machine Learning Trends in Chemistry Research using LLM with Multi-Turn Prompting

Abstract

Listen

Translate article

A review of research in the field of chemistry that incorporates machine learning is essential to identify recent developments and explore its potential applications. Published research articles provide an opportunity to analyze emerging research trends. The use of natural language processing (NLP) technology not only accelerates text data analysis but also enhances accuracy in understanding the content and context of scientific articles. Previously, trend analysis in ophthalmology research had been conducted using Zero-Shot Learning. In this study, an analysis of chemistry-related articles focusing on machine learning was carried out using a multi-turn prompting technique. The process began with data collection through web scraping of abstracts containing the keywords "machine learning" and "chemistry." The retrieved data was then tabulated and analyzed using a Large Language Model (LLM) with a Multi-Turn Prompting approach, where general prompts were initially used, followed by deeper exploration based on previous responses. Additionally, statistical descriptive analysis was performed using targeted prompts. Analysis of 200 article abstracts identified seven key terms related to the use of machine learning in chemistry: chemical (138 articles), protein (119 articles), drug (107 articles), structure (100 articles), molecular (96 articles), chemistry (91 articles), and quantum (84 articles). Furthermore, three dominant research topics were found in the intersection of chemistry and machine learning: protein and molecular structure, quantum chemistry, and drug discovery. The number of articles on machine learning in chemistry began to rise in 2012 and saw a significant increase in 2019. The findings suggest that there are still many opportunities for developing machine learning applications in chemistry, particularly in quantum chemistry. This field only began to gain attention in 2013, and the number of published articles remains relatively low each year, indicating that it is still in the early stages of exploration.