Precise diagnosis and prognostic prediction of HBV associated hepatocellular carcinoma by AI assisted multi-omics models: A multi-center prospective study

This repository contains the source code and data analysis workflows associated with the study titled "Precise diagnosis and prognostic prediction of HBV associated hepatocellular carcinoma by AI assisted multi-omics models: A multi-center prospective study". Study Abstract: Hepatitis B virus (HBV) infection progresses through chronic hepatitis B (CHB) and liver cirrhosis (LC) to hepatocellular carcinoma (HCC). Existing non-invasive biomarkers often lack sufficient performance for early diagnosis and risk stratification. This study aimed to establish lipid and peptide fingerprint-based machine learning models for the stepwise identification of HBV-associated liver diseases and HCC prognosis prediction. Methodology & Code Content: The scripts in this repository cover the analysis of serum samples from 1,052 individuals processed via high-throughput mass spectrometry Lab-on-a-Chip (MS LOC). The workflow includes: Data Preprocessing: Extraction and processing of peptide and lipid fingerprints. Feature Selection: Identification of key biomarkers (9 peptides and 10 lipids) revealing disruptions in biological processes and metabolism. Model Development: Implementation of ensemble machine learning techniques to construct diagnostic and prognostic models. Validation: Code for evaluating model performance (AUC, sensitivity, specificity) and conducting stepwise identification (HC, CHB, LC, HCC). Key Results: The AI-assisted HCC diagnostic model implemented herein achieved an AUC of 0.960 (93.3% sensitivity, 97.6% specificity) in the test cohort, significantly outperforming α-fetoprotein (AFP). The stepwise models successfully stratified disease stages in a blinded multi-center prospective test cohort. Additionally, the code includes the analysis for identifying 5 independent prognostic biomarkers. Clinical Trial Registration: ChiCTR2300073462