Akademik Veri Yönetim Sistemi
BIOMEDICAL SIGNAL PROCESSING AND CONTROL, cilt.100, 2025 (SCI-Expanded)
Over the past ten years, many machine learning methods emphasizing the relationship between the gut microbiome and gastrointestinal diseases have been proposed. These methods aim to minimize disadvantages such as high dimensionality and low sample size in metagenome datasets and attempt to find guiding biomarkers for early diagnosis of diseases. However, even if the same datasets are used in studies, reaching different results poses a critical obstacle in translating the findings into clinical practice. Therefore, by using deep learning models we adopted a novel approach (Disrobiom) to discover robust biomarkers on metagenomic datasets associated with Inflammatory bowel disease (IBD) and Colorectal cancer (CRC) for consistent and reproducible results. In the proposed study, we showed that the most important features selected on six different datasets with autoencoder, and U-net models resulted in superior classification performance compared to existing methods. We also evaluated the discovered biomarker candidates within the framework of consistency, compared them with existing methods, and discussed their contributions to the existing literature. Disrobiom has provided potential biomarkers that are more consistent than other methods for the non-invasive diagnosis and treatment of IBD and CRC diseases.