Five-Gene Expression Formula Accurately Detects Hepatocellular Carcinoma Tumors

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Hepatocellular carcinoma (HCC) is a major global cause of cancer mortality.
• Current diagnostic methods like imaging and Alpha-Fetoprotein (AFP) testing are limited to later disease stages.
• Transcriptomic analysis offers potential for early HCC detection, but machine learning applications face challenges in generalizability, interpretability, and complexity.

Purpose of the Study:

• To develop a novel, interpretable, and efficient predictive formula for early hepatocellular carcinoma (HCC) detection.
• To address the limitations of existing machine learning approaches in transcriptomic data analysis for cancer detection.

Main Methods:

• Utilized the Kolmogorov-Arnold Network (KAN) to create a predictive formula for HCC detection.
• The formula is based on the expression levels of five specific genes: VIPR1, CYP1A2, FCN3, ECM1, and LIFR.
• The model was trained on the GSE25097 dataset and validated on six independent datasets.

Main Results:

• The developed KAN-based formula achieved 99% accuracy on the GSE25097 test set.
• The formula demonstrated robust performance across six independent datasets, with accuracies consistently above 90%.
• Identified VIPR1, CYP1A2, FCN3, ECM1, and LIFR as critical gene biomarkers for HCC detection.

Conclusions:

• The novel KAN-based predictive formula provides a simple, interpretable, and efficient approach for early HCC identification.
• The identified five genes serve as promising biomarkers for improving HCC diagnostic strategies.
• This research lays the groundwork for future clinical applications to enhance early liver cancer detection and management.