<p>Prostate cancer is among the most prevalent cancers in men, affecting the prostate gland, a crucial part of the male reproductive system. It is associated with higher mortality rates and generally develops in older age, although it can also affect younger men. In this study, we integrate multi-omics data, including mRNA expression, DNA methylation, and copy number alterations, to identify potential molecular biomarkers that influence mortality rates and treatment responses across different age groups. First, we used Kaplan–Meier survival analysis to determine the optimal diagnostic age cutoff point at which mortality rates increase and treatment responses decline. The analysis identified 68 years as the optimal cutoff <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="521_2025_11386_Article_IEq1.gif" Format="GIF" Height="19" Rendition="HTML" Resolution="72" Type="Linedraw" Width="73" /> </InlineMediaObject> <EquationSource Format="TEX">$(p < 0.05)$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mo stretchy="false">(</mo> <mi>p</mi> <mo><</mo> <mn>0.05</mn> <mo stretchy="false">)</mo> </mrow> </math></EquationSource> </InlineEquation>, stratifying patients into two groups <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="521_2025_11386_Article_IEq2.gif" Format="GIF" Height="15" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">$\le$</EquationSource> <EquationSource Format="MATHML"><math> <mo>≤</mo> </math></EquationSource> </InlineEquation> 68 and > 68. We then applied multiple classification models using machine learning and deep learning algorithms based on multi-omics data to classify patients. After preprocessing the multi-omics data, we applied nonnegative matrix factorization (NMF) combined with a genetic algorithm for dimensionality reduction, transforming the data into a latent space. This transformation facilitates the seamless integration of diverse omics data types into the classification framework. We then classified the age groups using various standard classification techniques, including random forest, XGBoost, and the convolutional neural network (CNN). Among these, the random forest classifier outperformed the others, achieving an accuracy of 89% and an AUC of 0.88. The results demonstrate that classification models effectively identify low-risk patients, offering valuable insight to improve personalized treatment strategies in the management of prostate cancer.</p>

Integrative multi-omics approach for identifying and classifying high-risk patients in prostate cancer management

期刊论文
发表日期：

2025年6月23日

查看全文

Mohammad Azzeh,
Noor Afeshat,
Hazem Qattous,
Abedrahman Alkhateeb

摘要

Prostate cancer is among the most prevalent cancers in men, affecting the prostate gland, a crucial part of the male reproductive system. It is associated with higher mortality rates and generally develops in older age, although it can also affect younger men. In this study, we integrate multi-omics data, including mRNA expression, DNA methylation, and copy number alterations, to identify potential molecular biomarkers that influence mortality rates and treatment responses across different age groups. First, we used Kaplan–Meier survival analysis to determine the optimal diagnostic age cutoff point at which mortality rates increase and treatment responses decline. The analysis identified 68 years as the optimal cutoff $(p < 0.05)$ $(p < 0.05)$ , stratifying patients into two groups $\le$ $\leq$ 68 and > 68. We then applied multiple classification models using machine learning and deep learning algorithms based on multi-omics data to classify patients. After preprocessing the multi-omics data, we applied nonnegative matrix factorization (NMF) combined with a genetic algorithm for dimensionality reduction, transforming the data into a latent space. This transformation facilitates the seamless integration of diverse omics data types into the classification framework. We then classified the age groups using various standard classification techniques, including random forest, XGBoost, and the convolutional neural network (CNN). Among these, the random forest classifier outperformed the others, achieving an accuracy of 89% and an AUC of 0.88. The results demonstrate that classification models effectively identify low-risk patients, offering valuable insight to improve personalized treatment strategies in the management of prostate cancer.