Clustering colon cancer patients based on their gene expression

Colon cancer is the third death cause in women and the second in men with cancer-related diseases [1]. Many researchers are looking for new methods to cancer early diagnosis and recognition of cancer tumor characteristics [2]. Achieving to new cancer characteristics is very important for prediction, diagnosis, and treatment of cancer diseases. To recognize the cancer diseases, we need to analyze thousands of genes. Bioinformatics offers a great opportunity to comprehend and interpret a large number of genes. To address this problem, the first step is to use unsupervised learning algorithms to find significant hidden structures in the gene expression [3]. Here, we have analyzed the gene expression datasets of patients with colon adenocarcinoma; the dataset has been obtained from Genomic Data Commons (GDC) portal [4]. The number of analyzed patients was 456 and the number of genes available for each patient in the HTSeq - FPKM-UQ datasets was 60483. We have clustered the patients with colon cancer based on their primary tumor gene expression using a series of methods for sample-based clustering. The goal of sample-based clustering is to find the phenotypic signature of the patients in one cluster to obtain effective target therapies [5]. To use machine learning techniques, first, we need to normalize the dataset using an appropriate method. We modified the normalization method explained in [6] to normalize the dataset. We reduced its dimensions using dimensionality reduction techniques [7] and achieved to 99% variance with 273 dimensions. After dimension reduction, it is time to find a meaningful structure among cancer tumors. For this purpose, we clustered the colon cancer patients [8]. Finally, we determined the effective genes in each cluster using a new method. Definitely, the effective genes introduced in this research, have many applications in therapeutic analysis

Keywords:

Colon cancer , Clustering , Gene expression