Feature set prioritization of metagenomes for predicting colorectalcancer based on multiple human gut microbiome datasets

Colorectal cancer is one of the most prevalent cancers in the world and based on various studies, this canceris associated with gut microbiome dysbiosis. One step in developing diagnostic methods is understanding can cer mechanisms by finding the key features changing during cancer. However, independent studies have reporteddifferent results in finding the best feature sets for classifying samples into groups, due to differences in the quallity of datasets, sampling method, sample size, and analyzing methodologies. To address this issue, we defineda combined classification-robustness score and then employed a genetic algorithm to find the feature set thatmaximizes this score. This feature set is considered robust because it gives a high classification score whenapplied to new, unseen datasets. The classification score measures the strength of classifying groups by calcu lating separability measures, and the robustness score reflects the level of consistency of selected features acrossvarious input datasets. In addition to taxonomic profiling, which provides insight into the cancer-associatedspecies, we also achieved a functional perspective by extracting features from seven public whole-metagenomesequencing datasets of colorectal cancer at both the species and functional levels (i.e., KEGG Orthology group,Enzyme Commission number, and reaction). We used the genetic algorithm to prioritize feature sets and foundthat exploiting multiple datasets and different feature types resulted in higher classification and robustnessscores compared to using a single dataset or feature type. In conclusion, our study emphasizes the importanceof combining datasets and feature types for a more robust and effective classification of colorectal cancer

کلیدواژه ها:

Colorectal Cancer ، Microbiome ، Genetic Algorithm ، Robust Features