Decoding the Molecular Landscape of Inflammatory Bowel Disease: A Mendelian Randomization and Machine Learning Analysis

Background and Aim Inflammatory bowel disease (IBD) is a complex condition with rising global incidence. Its development is influenced by genetic, environmental, and metabolic factors. This research seeks to explore the causal association between plasma proteins and IBD through Mendelian randomization (MR) and machine learning approaches. Methods A two‐phase MR approach was employed, using genetic variants as instrumental variables to assess the causal effect of plasma proteins on IBD. The discovery phase utilized summary‐data‐based MR with data from the Fenland study and FinnGen release 11, followed by machine learning prediction models. The replication phase validated findings using data from the Pan‐UKB and the IIBDGC. Two‐sample MR and SMR‐based Phenome‐Wide Association Studies (SMR‐PheWAS) were conducted to explore druggability and potential side effects. Results The study identified several proteins significantly associated with IBD risk, validated across independent cohorts. Machine learning models highlighted genes like FCGR3A, CD274, C4B, and TIMD4 in predicting CD, and MAPKAPK2 and TWSG1 in predicting UC. Potential therapeutic targets identified include glucokinase regulatory protein (GCKR), hepatocyte growth factor‐like protein (MSP), and low‐affinity immunoglobulin gamma Fc region receptor II‐a (FCG2A). SMR‐PheWAS indicated that MSP and GCKR exhibited positive causal associations with other endpoints in FinnGen release 10. Conclusion The study advances the understanding of IBD pathogenesis through an innovative approach, large diverse datasets, and advanced statistical tests. It contributes to precision medicine by improving diagnosis, treatment, and patient outcomes, though limitations include reliance on European populations and the need for validation in diverse populations.