A deep learning approach based on multi-omics data integration to construct a risk stratification prediction model...

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Skin cutaneous melanoma (SKCM) is an aggressive cancer with complex etiology, making prognosis prediction challenging.
• High heterogeneity in SKCM necessitates advanced models for accurate risk stratification.

Purpose of the Study:

• To construct a novel risk subtype typing model for SKCM using a multi-omics deep learning approach.
• To improve the prediction of prognosis and treatment decisions for SKCM patients.

Main Methods:

• A deep learning framework combining early and late fusion autoencoders (AE) was developed.
• Integrated mRNA, miRNA, and DNA methylation data from The Cancer Genome Atlas (TCGA).
• Utilized differential expression analysis, LASSO regression, and SVM classification for subtype identification and validation.

Main Results:

• The proposed deep learning framework identified two distinct SKCM risk subtypes with significant survival differences (C-index = 0.748, log-rank P < 10^-8).
• Differentially expressed genes and miRNAs provided biological insights into SKCM subtypes.
• The model demonstrated robust predictive power on independent datasets (C-index > 0.680).

Conclusions:

• The multi-omics deep learning model effectively identifies SKCM risk subtypes.
• This approach enhances understanding of SKCM molecular mechanisms.
• The identified subtypes can assist clinicians in making informed treatment decisions.