Feature Enhancement and Chaining of Deep Nueral Networks in Colorectal Cancer Classifaction based on Gut-Microbiome Data

Abstract

Colorectal Cancer (CRC) is among the top three cancers in world. The current clincal methods for CRC detection have several limitations which range from low accuracy, discomfort and high costs. Availability of next generation sequencing (NGS) technology has opend an opportunity for non invasive detection of CRC which uses gut-microbiome abundance in stool samples. The high dimension of sequence base microbiome data has prompted research interest in the application of machine learning (ML) in order to classify host disease based on microbial counts. However the classification performance of ML methods such data is still limited by factors shuch as high dimensionality and data imbalance. Therefore, in this paper, we propose a deep nueral network based method that combines feature extension and feature and chained execution of deep neural network to improve CRC classifaction based on gut microbiome in stool samples. The proposed method scored a mean area under the receiver operating characteristics curve (AUC) of approximately 95.4%, which is higher than state-of-the-art methods. The proposed method can positively contribute to the development of robust diagnostic and prognostic methods for CRC.