Machine learning algorithm and deep neural networks identified a novel subtype in hepatocellular carcinoma

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Hepatocellular carcinoma (HCC) is a prevalent malignancy often diagnosed at advanced stages due to subtle early symptoms.
• Lack of specific early indicators complicates timely diagnosis and intervention for HCC patients.

Purpose of the Study:

• To leverage machine learning for identifying critical genes in HCC progression.
• To develop a predictive model for assessing hepatocellular carcinoma patient survival risk.

Main Methods:

• Utilized TCGA and GEO transcriptome and clinical data.
• Employed differential expression analysis, COX models, K-Means, Random Forests, and LASSO for gene and subtype identification.
• Constructed a deep neural network (DNN) model for survival prediction and GSEA for pathway analysis.

Main Results:

• Identified two HCC subtypes with distinct survival rates (p<0.0001, AUC=0.720).
• Screened 17 key survival-related genes associated with HCC subtypes.
• Achieved >93.3% accuracy with the DNN survival prediction model.
• GSEA revealed significant enrichment of survival genes in hallmark pathways (MSigDB).

Conclusions:

• Machine learning effectively identified 17 genes crucial for HCC survival risk.
• A robust DNN model was developed to predict HCC patient survival.
• The identified genes are integral to cancer development and progression, offering potential therapeutic targets.