Exploring the Potential of Large Language Models in Differential Abundance Analysis

Abstract

The rapid development of Large Language Models (LLMs) has opened new possibilities in various fields, including microbiome research. This dissertation explores the application of LLMs for differential abundance analysis, a crucial method for understanding the relationship between microbial communities and health conditions. Through a series of experiments, we assessed the capabilities and limitations of LLMs in automating and enhancing the differential abundance analysis process. Our findings reveal that while LLMs can effectively extract relevant information from scientific literature and assist in generating reports, they also face significant challenges, including data inaccuracies, hallucinations, and issues with reference generation. These challenges highlight the importance of integrating LLMs with human oversight to ensure scientific rigor. The study suggests that while LLMs can increase research efficiency, they are not yet reliable enough to replace human expertise in complex scientific tasks. This research contributes to the ongoing dialogue on the role of AI in scientific research, emphasizing the need for further development of LLMs and the exploration of hybrid models that combine AI capabilities with human judgment. The dissertation concludes with recommendations for improving the accuracy and consistency of LLMs, and expanding their applications.