# Integration of single-cell sequencing, transcriptome sequencing, and machine learning for constructing and validating histone acetylation-related prognostic risk models in hepatocellular carcinoma

## TL;DR

This study builds a risk model for liver cancer using histone acetylation data and machine learning, identifying NEU1 as a key biomarker and potential treatment target.

## Contribution

A novel histone acetylation-related risk model for hepatocellular carcinoma is developed using 101 machine learning algorithms and validated experimentally.

## Key findings

- An 11-gene risk model with strong prognostic accuracy was developed and validated across multiple LIHC subtypes.
- NEU1 was identified as a key risk factor, with high expression linked to tumor progression and validated through molecular and in vitro experiments.
- The model showed distinct immune and chemosensitivity profiles, with high-risk groups responding better to PD-L1/PD-L2 and certain drugs like Axitinib.

## Abstract

Liver hepatocellular carcinoma (LIHC), a prevalent gastrointestinal malignancy, continues to demonstrate poor prognosis despite therapeutic advances improving clinical outcomes. Histone acetylation, a key epigenetic modification, regulates critical processes including chromatin remodeling, gene expression and drives tumor progression in multiple cancers (e.g., lung, gastric) yet its systemic role in LIHC remains unclear.

This study integrated LIHC single-cell/RNA-seq data and histone acetylation-related gene sets to construct a LIHC risk prediction model based on histone acetylation-related genes using 101 machine learning combinatorial algorithms. The model’s comprehensive value was evaluated through prognostic analysis, pathway enrichment analysis, immune landscape analysis, chemosensitivity analysis, mutation analysis, ferroptosis, and m6A methylation analysis. NEU1’s functional role was investigated via cell communication networks and molecular docking. Experimental validation included in vitro assays (Cell Counting Kit-8, migration, invasion), and clinical sample verification (quantitative real-time PCR (qRT-PCR) and Western Blot (WB)); these were performed to validate the key findings.

Using 101 machine learning combinations, we constructed an 11-gene LIHC risk model (HLA-B, HEXB, CDK4, ACAT1, NAA10, B2M, HSPD1, NPM1, PON1, NEU1, CFB) demonstrating robust prognostic accuracy across training/validation cohorts and 10 LIHC subtypes. Immune landscape analysis revealed that the high-risk group exhibited higher tumor purity and lower immune infiltration, with better responses to PD-L1 and PD-L2 treatment. Chemosensitivity analysis showed that the high-risk group had increased sensitivity to four drugs, including Axitinib, but decreased sensitivity to 21 drugs, including Cisplatin. The risk model score significantly correlated with the expression levels of ferroptosis-related genes such as GPX4 and m6A methylation-related genes such as METTL3. NEU1 was identified as a key risk factor in this model, with the NEU1 high-expression group showing of intercellular communication in endothelial cells and other cell types. Pseudotime analysis suggested that NEU1 may promote LIHC progression by blocking normal differentiation of endothelial cells. Molecular docking revealed that five compounds, including Oseltamivir, could bind directly to NEU1. Knockdown of NEU1 significantly reduced proliferation, migration, and invasion of LIHC cells, and slowed LIHC tumor growth.

We constructed a histone acetylation-based risk model for LIHC diagnosis, prognosis, and therapy, identifying NEU1 as a key biomarker and potential therapeutic target.