Multi-Dimensional Scaling Analysis of Key Regulatory Genes in Prostate Cancer Using the TCGA Database

Abstract

Prostate cancer (PC) is a polygenic disease with multiple gene interactions. Therefore, a detailed analysis of its epidemiology and evaluation of risk factors can help to identify more accurate predictors of aggressive disease. We used the transcriptome data from a cohort of 243 patients from the Cancer Genome Atlas (TCGA) database. Key regulatory genes involved in proliferation activity, in the regulation of stress, and in the regulation of inflammation processes of the tumor microenvironment were selected to test a priori multi-dimensional scaling (MDS) models and create a combined score to better predict the patients’ survival and disease-free intervals. Survival was positively correlated with cortisol expression and negatively with Mini-Chromosome Maintenance 7 (MCM7) and Breast-Related Cancer Antigen2 (BRCA2) expression. The disease-free interval was negatively related to the expression of enhancer of zeste homolog 2 (EZH2), MCM7, BRCA2, and programmed cell death 1 ligand 1 (PD-L1). MDS suggested two separate pathways of activation in PC. Within these two dimensions three separate clusters emerged: (1) cortisol and brain-derived neurotrophic factor BDNF, (2) PD-L1 and cytotoxic-T-lymphocyte-associated protein 4 (CTL4); (3) and finally EZH2, MCM7, BRCA2, and c-Myc. We entered the three clusters of association shown in the MDS in several Kaplan–Meier analyses. It was found that only Cluster 3 was significantly related to the interval-disease free, indicating that patients with an overall higher activity of regulatory genes of proliferation and DNA repair had a lower probability to have a longer disease-free time. In conclusion, our data study provided initial evidence that selecting patients with a high grade of proliferation and DNA repair activity could lead to an early identification of an aggressive PC with a potentials for metastatic development.

Assessment of using a gamma index analysis for patient‐specific quality assurance in Japan

“…Accordingly, this method reveals the nonlinear and underlying relationships among the data nodes. 15 , 16 , 17 , 18 , 19 , 20 The factors for analysis were all composed of the number of facilities; therefore, we did not apply standardization (subtracting mean and dividing by standard deviation; z ‐score) but normalization (dividing by 148 that is the total facility number using IMRT in this survey) for preprocessing. We used scikit‐learn (version 1.0.2) for the PCA and MDS calculations, 21 a widely used Python 3.8 module library for machine learning.…”

“…The MDS method effectively maps the similarity of each data node based on the Euclidean or geodesic distance. Accordingly, this method reveals the nonlinear and underlying relationships among the data nodes 15–20 . The factors for analysis were all composed of the number of facilities; therefore, we did not apply standardization (subtracting mean and dividing by standard deviation; z ‐score) but normalization (dividing by 148 that is the total facility number using IMRT in this survey) for preprocessing.…”