Makine öğrenmesi bazlı sınıflandırma kullanarak nörogelişimsel hastalıklar için ayırt edici özelliklerin bulunması

Major Depressive Disorder (MDD) is one of the leading neurodevelopmental disorders worldwide and its heterogeneous and complex nature remains a significant challenge for scientists to fully understand and unravel. In recent years, the importance of gut microbiome through the human gut-brain axis has gathered the attention of scientists to analyze and model the bacterial components of the human gut and its effects on the human brain, especially in the case of MDD. In this study, we analyzed the American Gut Project (AGP) dataset using fecal samples of 361 controls and 23 MDD patients. After retrieving the Qiita bioinformatics analysis, the cohort was analyzed for its characteristics and for alpha and beta diversity indexes which did not reveal any statistical significance except for the age of patients. Various differential abundance analysis (DAA) methods were conducted to find potential biomarkers and these results were fed into our machine learning models as an alternative DAA-filtered dataset compared to the raw dataset to find important features. Our best two models, which were Random Forest and XGBoost, at their intersection, have found that despite some inconsistencies in the literature, species Bifidobacterium adolescentis and genera Odoribacter, Ruminococcus, and Adlercreutzia could be potential biomarkers for MFD. Our models found that species B. adolescentis decreased in MDD patients whereas the rest increased in MDD patients. Despite supporting evidence for these potential biomarkers, factors such as modeling and filtering choices, as well as external influences like sex, stress and diet should also be taken into account when analyzing gut microbiome datasets.