IntegrIBS: Towards Building a Robust IBS Classifier with Integrated Microbiome Data

Abstract

Irritable Bowel Syndrome (IBS) is a condition that is quite complicated and shares its symptoms with other related diseases, making it difficult to diagnose. In this study, we initially trained machine learning models on individual microbiome datasets and tested their performance on other datasets, observing variability and low precision among them. To mitigate this, we hypothesised that integrating multiple publicly available microbiome datasets will capture a wide spectrum of microbiome variations across different geographies and demographics. Utilizing this integrated dataset, the XGBoost model achieved a mean accuracy of 0.75 with a standard deviation of 0.04 in 10-fold cross-validation, demonstrating its potential for robust IBS prediction. Explainability analysis identified key bacterial taxa influencing predictions, aligning with existing literature. However, the models performance declined significantly when using a leave-one-dataset-out approach, where the model was trained on all but one dataset and tested on the excluded dataset. The results highlight the challenges of generalizing across diverse datasets due to biological and technical variability. These findings present a cautionary tale regarding the integration of datasets and interpretation of results, emphasizing the need for more comprehensive approaches to develop reliable diagnostic tools for IBS.