Microbiome Classification in Colon Cancer Patients Using Chaos Game Representations and CNN Models

Emerging evidence links Colorectal Cancer (CRC) risk to antibiotic use, yet it remains uncertain if CRC correlates with the gut microbiota’s resistome. The human gut resistome encompasses all genes in the microbiome that confer antibiotic resistance, termed Antibiotic Resistance Genes (ARG). The diversity and genome location of ARGs complicate their detection in the complex gut microbiome. Variability in sequences, dispersion in larger genetic structures, and background noise in microbiome data challenge traditional string matching methods. Advanced bioinformatics tools, like machine learning (ML) models, are needed to detect and analyze ARG dynamics in the gut microbiome. This paper proposes a ML approach for the identification of ARGs in gut microbiome data with the aim of serving as a tool for analyzing the association between ARG abundance and CRC development. Chaos Game Representation (CGR) was used as a data preprocessing and feature extraction technique for representing microbiome sequences and ARGs as images. Convolutional Neural Networks (CNNs) were used to define the classifier to identify ARGs in the gut microbiomes of CRC patients and healthy individuals. The results show that by exploring different features of these algorithms, including the resolution of the CGR images and different CNN structures, it is possible to build classifiers with accuracy and precision of the class of interest of 98.7%-99.2% and 95.2%-98.3%, respectively. However, it is necessary to analyze a larger volume of data from patients and healthy individuals in order to conclude whether CRC development is indeed linked to the gut microbiota’s resistome.