Gruplama puanlama modelleme (G-S-M) ve geleneksel özellik seçim yaklaşımını kullanarak insan gastrointestinal kanser mikrobiyotalarındaki potansiyel taksonomik biyobelirteçlerin belirlenmesi

Analysis of microbial abundance values holds potential for cancer prediction. This study aims to identify shared microbial biomarkers among gastrointestinal (GI) cancer patients using both tissue and blood samples—an area not previously studied in parallel. This study analyzed blood and tissue samples, focusing on head and neck, esophagus, stomach, colon, and colorectal cancers, processing them individually. By performing decontamination steps, processing non-human genetic codes, determining microorganisms and their abundances at the species level, the TCMA data set was created from the "Cancer Genome Atlas", which collected tissue and blood samples from cancer patients. Traditional feature selection algorithms (CMIM, mRMR, FCBF, IG, XGB, and SKB) reduced the high-dimensional feature space. Classification performance was evaluated using a forest classifier with 100-fold Monte Carlo cross-validation. Moreover, the MicrobiomeGSM model, which was created to decrease the feature size and prediction time via a grouping method, was trained, and the generalizability of the MicrobiomeGSM model was showcased. Traditional feature selection methods and the biological data-based MicrobiomeGSM model were applied, and their performance was compared. In the future, common biomarker candidates may help to understand the possibility of metastasis, and medical doctors can decide their treatment path of patients.