Deep Learning-Based Multi-Omics Integration Robustly Predicts Relapse in Prostate Cancer.

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Post-operative biochemical relapse (BCR) is common in prostate cancer (PCa) patients.
• Current methods inadequately stratify high-risk patients for localized PCa.
• Multi-omics data integration is needed for improved risk stratification.

Purpose of the Study:

• To develop a deep learning (DL) model for predicting BCR in PCa patients.
• To leverage multi-omics data for enhanced relapse prediction accuracy.
• To identify molecular markers associated with PCa relapse.

Main Methods:

• Utilized five-omics data from 417 PCa patients from The Cancer Genome Atlas (TCGA).
• Constructed a DL-based model using the H2O package for relapse prediction.
• Performed bioinformatics analyses including DEG, pathway, CNV, and immune cell infiltration analysis on relapse-associated subgroups.

Main Results:

• The DL model demonstrated robust predictive performance with a C-index of 0.767, validated externally.
• Identified 1530 differentially expressed genes (DEGs) between high-risk (S2) and low-risk (S1) subgroups.
• Found enrichment in DNA damage repair pathways and altered immune cell infiltration in the high-risk subgroup.

Conclusions:

• Developed a novel, robust DL-based classification for PCa relapse prediction.
• Validated the effectiveness of DL in cancer prognosis.
• The model holds potential for improving early detection and intervention strategies to prevent PCa relapse.