The work in BrainGutAnalytics project was performed in three work packages. The first work package dealt with collection, preparation, processing and integration of multiomics data collected from various sites across human body in parent Bergen Brain-Gut study. This data included microbiomics (fecal and saliva microbiome), metabolomics (short chain fatty acids), functional connectivity maps of brain regions (functional MRI), patient questionnaires (IBS specific clinical questionnaires, psychological and cognitive questionnaires, sleep and fatigue questionnaires etc.) and physiology measurements (heart and respiratory rates, before and after lactulose intake). After necessary cleaning and pre-processing, all data was algorithmically readout from their source data bases and was integrated into a uniform database for downstream processing and analysis.
In the second work package, a variety of data science based analysis techniques e.g. exploratory data analysis, machine learning based modelling and statistical methods were applied with methodological variations (e.g. feature selection techniques, ensembles, etc.) on the collective multiomics data frame, developed in the first phase. The aim of these experiments was to search for any significant associations or differences in complex multiomics data, which could be used to identify any potential set of variables (or features) for effective classification between patients with IBS and healthy control subjects. For this purpose, we looked into abundance and diversity of microbiota species within faeces and saliva samples, metabolites collected from fecal samples, functional connectivity maps of various brain regions, heart and respiration rate before and after lactulose intake by the participants and a number of patient questionnaires on both individual basis as well as in various possible combinations with each other.
Unexpectedly, our extensive search in the given multiomics data did not yield any individual or combination of variables that could be exclusively associated with either patients with IBS or healthy control subjects and eventually account for IBS phenotypes. However, our experimentation revealed a new set of features that mainly comprised of anxiety, fatigue and gender related variables, which resulted in classification accuracy of about 93% on a hold out test dataset. This finding shows prevalence of a complex multivariate combination of psychological symptoms in patients with IBS and provides an opportunity to use the same feature set for psychological evaluation of IBS cases to potentially complement and improve existing IBS diagnosis practice in clinical setting.
The third work package in BrainGutAnalytics concerned the development of a clinical diagnosis system for IBS evaluation based on the psychological feature set identified and validated in the second work package. For this purpose, a free and open access online web application named www.checkibs.com was developed, which uses the machine learning model (developed in previous step) to provide a new decision support system to clinicians in making an effective diagnosis of IBS.
Throughout the project duration, the Fellow working in BrainGutAnalytics project participated in various conferences, seminars, scientific events and in-person meetings to disseminate information on his scientific research, the project and the Marie Curie Fellow program. He also setup two websites dedicated to the project to disseminate information on his research activities. His work on development of the web application based on psychological evaluation of IBS was awarded with National Scholar Award in United European Gastroenterology Week 2022, which one of the biggest scientific conference in the field of gastrointestinal health.
