Deep Learning-Based Multi-Omics Integration Robustly Predicts Survival in Liver Cancer

Area of Science:

• Oncology
• Bioinformatics
• Genomics
• Machine Learning

Background:

• Identifying distinct hepatocellular carcinoma (HCC) patient subgroups is crucial for improving treatment strategies and patient outcomes.
• Current methods for predicting HCC survival often lack the integration of multi-omics data across diverse patient cohorts.
• A gap exists in robustly predicting HCC survival by integrating multi-omics data from multiple patient cohorts.

Purpose of the Study:

• To develop and validate a deep learning (DL)-based model for differentiating HCC survival subpopulations.
• To integrate multi-omics data (RNA-Seq, miRNA-Seq, methylation) for enhanced HCC prognosis prediction.
• To identify molecular features associated with differential survival in HCC patients.

Main Methods:

• A deep learning model was constructed using RNA sequencing, miRNA sequencing, and methylation data from 360 HCC patients (TCGA cohort).
• The model was designed to be survival-sensitive, differentiating patient subgroups with significant survival differences.
• Model performance was validated on five independent external datasets comprising various omics types.

Main Results:

• The DL-based model identified two HCC patient subgroups with statistically significant survival differences (P = 7.13e-6) and good predictive accuracy (C-index = 0.68).
• The aggressive subtype was characterized by TP53 mutations, elevated stemness markers (KRT19, EPCAM), BIRC5 expression, and activated Wnt/Akt pathways.
• The model demonstrated robust performance across multiple validation cohorts, with concordance indices ranging from 0.67 to 0.77.

Conclusions:

• This study presents the first deep learning approach to identify multi-omics features linked to differential HCC patient survival.
• The developed DL model offers a robust and validated method for HCC prognosis prediction across diverse patient cohorts.
• The findings provide a foundation for personalized treatment strategies by stratifying HCC patients into distinct prognostic subgroups.