Facilitating Precision Medicine in HCC Patients by Deep Learning-Directed lncRNAs Classification and Ascertaining...

Hepatocellular carcinoma (HCC) is a deadly disease, ranking as the fifth most prevalent type of cancer worldwide. To advance precision medicine for HCC, it is imperative to identify novel biomarkers that can facilitate its clinical management. Notably, long noncoding RNAs (lncRNAs) have been demonstrated to play pivotal roles in regulating gene expression and driving tumor progression.

This study harnessed artificial intelligence (AI) to identify novel lncRNA biomarkers using patient data from The Cancer Genome Atlas. A hierarchical composite deep learning (DL) framework was constructed to classify patient samples according to the pathological stages. Explainable AI-based SHapley Additive exPlanations (SHAP) analysis, Kaplan-Meier survival analysis, and biological investigations were performed to identify potential lncRNA biomarkers.

The DL hierarchical composite framework employed three deep neural networks (DNNs) to classify tumorous liver tissue samples from normal samples (Model 1), distinguish between early and advanced disease stages (Model 2), and further assess the pathological stage of HCC progression (Model 3). Model 1 (AU-ROC = 1.000) and Model 2 (AU-ROC = 0.977) exhibited robust predictive capabilities on unseen data. Model 3 (AU-ROC = 0.774) showed lower performance, highlighting the inherent challenges in the data due to the close association of lncRNAs across advanced HCC stages. SHAP analysis revealed the key lncRNAs driving these classifications, providing insights into the molecular mechanisms of HCC progression. Further investigations identified AC010280.2, AC118754.1, DIO3OS, LINC01748, and LINC00659 as potential biomarkers for HCC.

This comprehensive study integrated cutting-edge technologies that pave the way for biomarker discovery, facilitating precision medicine in HCC.