Survival prediction in patients with colon adenocarcinoma via multi-omics data integration using a deep learning...

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Colon adenocarcinoma (COAD) poses a significant health challenge.
• Predicting patient survival is crucial for effective treatment strategies.
• Integrating multi-omics data offers a comprehensive approach to understanding COAD.

Purpose of the Study:

• To develop a deep learning (DL) algorithm for predicting survival in COAD patients.
• To identify survival-related features using multi-omics data integration.
• To validate the prognostic robustness of the developed model.

Main Methods:

• Utilized an autoencoder for DL implementation on The Cancer Genome Atlas (TCGA) COAD data.
• Compared autoencoder performance against PCA, NMF, t-SNE, and univariable Cox-PH models.
• Validated prognostic robustness using three independent confirmation cohorts.
• Performed differential expression, correlation, miRNA-target network, and pathway enrichment analyses.

Main Results:

• The autoencoder-based DL model identified two distinct risk groups with significant survival differences (log-rank p-value = 5.51e-07).
• The DL framework demonstrated superior performance over traditional methods based on C-index, log-rank p-value, and Brier score.
• Validated classification model robustness across three independent datasets.
• Identified 1271 differentially expressed genes, 10 miRNAs, and 12 hypermethylated genes.
• miR-133b and its target genes were implicated in ECM-receptor interaction, focal adhesion, PI3K-Akt signaling, and glucose metabolism pathways.

Conclusions:

• The DL algorithm provides a robust method for predicting COAD patient survival through multi-omics integration.
• miR-133b and its target genes represent potential diagnostic biomarkers for COAD.
• The findings contribute to understanding COAD pathogenesis and may aid in clinical decision-making.