Multi-Omics Feature Selection to Identify Biomarkers for Hepatocellular Carcinoma.

Area of Science:

• Biomarker discovery
• Liver cancer research
• Multi-omics data integration

Background:

• Hepatocellular carcinoma (HCC) is a leading cause of cancer mortality globally.
• Late-stage diagnosis and inadequate biomarkers hinder effective early detection and treatment of HCC.
• Novel biomarkers are crucial for improving patient prognosis and survival rates.

Purpose of the Study:

• To identify a panel of multi-omics features distinguishing HCC from liver cirrhosis using serum samples.
• To evaluate and compare various feature selection methods for multi-omics data integration.
• To develop and assess a novel deep learning approach for HCC biomarker discovery.

Main Methods:

• Untargeted and targeted mass spectrometry were used to generate multi-omics data from serum samples of HCC and cirrhotic patients.
• Recursive feature selection combined with a transformer-based deep learning model was employed for feature identification.
• Performance evaluation of different feature selection techniques was conducted to identify HCC-specific markers.

Main Results:

• Key molecules including leucine, isoleucine, and SERPINA1 were identified as significantly associated with liver cancer.
• SERPINA1 is implicated in LXR/RXR Activation and Acute Response signaling pathways.
• The novel recursive feature selection and transformer-based deep learning method outperformed other sequential deep learning approaches.

Conclusions:

• The study highlights the potential of integrating multi-omics data with advanced deep learning models for robust biomarker discovery in HCC.
• Adapting deep learning models for feature selection is essential to mitigate overfitting risks with limited sample sizes.
• Further validation of the discovered multi-omics features in larger, independent cohorts is recommended for robust HCC biomarker identification.