HCCPred

HCCPred predicts platform-independent diagnostic and prognostic biomarkers for hepatocellular carcinoma using large-scale transcriptomic data.

Key Features:

• Platform-Independent Biomarker Identification: Compiles 2,306 HCC and 1,655 non-tumorous tissue samples from 29 studies and identifies biomarkers consistent across Affymetrix, Illumina, Agilent, and high-throughput sequencing platforms.
• Robust Biomarker Panels: Provides 3-gene (FCN3, CLEC1B, PRC1), 4-gene, and 5-gene prediction modules derived from differential expression analysis of 26 core HCC-associated genes.
• Machine Learning Techniques: Employs a suite of classifiers including simple threshold-based approach, Extra Trees, Support Vector Machine, Random Forest, K Neighbors Classifier, and Logistic Regression for biomarker selection and classification.
• High Predictive Accuracy: Reports classification accuracy of 93%–98% and AUROC of 0.97–1.00 for the three-gene panel across training and external validation datasets.
• Prognostic Evaluation: Assesses prognostic potential via univariate survival analysis on TCGA and GSE14520 cohorts for survival metrics OS, PFS, DFS/RFS, and DSS to stratify high- and low-risk patients.

Scientific Applications:

• Early detection of HCC: Enables development of diagnostic biomarkers applicable across microarray and sequencing platforms for earlier hepatocellular carcinoma detection.
• Prognostic stratification: Uses gene panels to stratify patients by overall and progression-related survival outcomes (OS, PFS, DFS/RFS, DSS).
• Cross-platform biomarker validation: Supports validation of candidate biomarkers across Affymetrix, Illumina, Agilent, and high-throughput sequencing datasets.
• Biomarker panel development: Facilitates derivation and external validation of compact gene panels (3-, 4-, 5-gene) for diagnostic and prognostic use.

Methodology:

Aggregated transcriptomic datasets (2,306 HCC, 1,655 non-tumorous samples) from 29 studies, performed differential expression analysis of 26 core genes, trained and evaluated classifiers (simple threshold, Extra Trees, SVM, Random Forest, KNN, Logistic Regression) for biomarker panels, and conducted univariate survival analysis on TCGA and GSE14520 cohorts.