Clustering Effect on Cancer Molecular Subtype Classification

Deep learning(DL) is a branch of artificial intelligence that emulates human brain functions through computational processes. It has demonstrated its effectiveness across various domains; healthcare is no exception. Encouraging outcomes have been achieved in multiple healthcare applications, which include the classification of cancer, its prognosis, diagnosis, and classifying different molecular subtypes of cancer. Molecular subtyping using gene expression data may provide biological insights into cancer heterogeneity, which is instrumental in developing personalized medicine. The samples' scarcity relative to the high dimensional feature space remains a challenge in implementing deep learning models. This research investigates the effectiveness of clustering for reducing the dimensionality of the transcriptomic data and its subsequent influence on classification accuracy. The proposed method clusters the features and leverages the cluster centroids to train the classification model to predict the cancer molecular subtypes of colorectal cancer. The result comparison of the model with and without clustering reveals improved performance, in our proposed framework, while achieving parity with accuracy levels in others.