Application of Machine Learning and Large Language Model Module for Analyzing Gut Microbiota Data

In this paper, we aim to explore the applications of machine learning and large language model for analyzing gut microbiota data, particularly attempting to investigate large language model module to intelligently conduct data analysis just through prompts. The data of gut microbiota is from 16S rRNA sequencing result of obese mouse for obesity experimental research, and our primary task focus on identifying differentially expressed genes and uncovering biomarker microbiota associated with obesity. Statistical methods, including diversity analysis and principal component analysis, have been firstly conducted, revealing significant differences between experimental and control groups. Then different machine learning algorithms including random forest, SVM-RFE, Lasso, and XGBoost have been selected to discover distinguished genes through feature importance ranking. Several types of microbial genes have been identified by both of the machine learning methods, and these biomarkers exhibit significant abundance changes between groups. Some of them have been confirmed with related research literature, and the remains are worthy of attention for more investigation. Furthermore, we have developed a large language model module named Chat2GM, which is based on the LangChain framework, python Flask, and API of OpenAI’s GPT model, particularly for this data analysis task. Upon uploading the original microbiota data, data exploration and analysis can be conducted merely by using prompts. It has been demonstrated to achieve satisfactory performance in completing the aforementioned analysis tasks with precise natural language instructions, indicating that applications powered by large language models with more expertise knowledge have great potential in gut microbiota research field.