Machine-Learning-Based Clinical Biomarker Using Cell-Free DNA for Hepatocellular Carcinoma (HCC)

Abstract

(1) Background: Hepatocellular carcinoma (HCC) is one of the leading causes of cancer-related death worldwide. Although various serum enzymes have been utilized for the diagnosis and prognosis of HCC, the currently available biomarkers lack the sensitivity needed to detect HCC at early stages and accurately predict treatment responses. (2) Methods: We utilized our highly sensitive cell-free DNA (cfDNA) detection system, in combination with a machine learning algorithm, to provide a platform for improved diagnosis and prognosis of HCC. (3) Results: cfDNA, specifically alpha-fetoprotein (AFP) expression in captured cfDNA, demonstrated the highest accuracy for diagnosing malignancies among the serum/plasma biomarkers used in this study, including AFP, aspartate aminotransferase, alanine aminotransferase, albumin, alkaline phosphatase, and bilirubin. The diagnostic/prognostic capability of cfDNA was further improved by establishing a cfDNA score (cfD_HCC), which integrated the total plasma cfDNA levels and cfAFP-DNA expression into a single score using machine learning algorithms. (4) Conclusion: The cfD_HCC score demonstrated significantly improved accuracy in determining the pathological features of HCC and predicting patients' survival outcomes compared to the other biomarkers. The results presented herein reveal that our cfDNA capture/analysis platform is a promising approach to effectively utilize cfDNA as a biomarker for the diagnosis and prognosis of HCC.

Keywords: cell-free DNA (cfDNA); circulating tumor DNA (ctDNA); hepatocellular carcinoma (HCC); liquid biopsy; principal component analysis (PCA).

Figure 1

Enhanced detection of cfDNA for its use as a biomarker for the diagnosis and prognosis of HCC. (A) A schematic diagram illustrating the cfDNA detection and analysis methods using PDA-SiO₂ beads. (B) The copy numbers of plasma cfAFP-DNA isolated from human plasma samples using either PDA-SiO₂ beads (black) or QIAamp DNA mini-kit (gray).

Figure 2

The diagnostic capability of cfDNA for detecting HCC patients from non-cancer (NC) cohorts: (A) the expression profiles of serum enzymes, plasma cfDNA, and cfAFP-DNA quantified from a total of 152 HCC patients and 97 non-cancer cohorts, which include 43 patients diagnosed with liver cirrhosis (LC), 24 patients diagnosed with alcoholic liver hepatitis (LA), and 30 healthy donors (HD). (B,C) ROC curves for diagnosing HCC from NC, LC, LA, and HDs using the expression profiles of serum enzymes, plasma cfDNA, and cfAFP-DNA. (D) Diagnostic performance of serum enzymes, plasma cfDNA, and cfAFP-DNA for detecting HCC patients.

Figure 3

The clinical performance of cfDNA for determining the pathological features of HCC tumors: (A) ROC curves for determining the UICC stages, existence of LVI, tumor size, and multifocality of a tumor. (B) Sensitivity, specificity, and accuracy of each biomarker for determining the UICC stages, existence of LVI, tumor size, and multifocality of a tumor.

Figure 4

The clinical performance of cfDNA for predicting the survival outcomes: Kaplan–Meier survival analysis for (A) recurrence, (B) marginal recurrence, (C) multifocal recurrence, and (D) overall survival of TACE-treated patients. Univariate Cox regression analysis of the serum and plasma biomarkers for (E) TACE-treated and (F) non-treated patients.