Abstract
Liver cancer is one of the most prevalent forms of cancer worldwide. A significant proportion of patients with hepatocellular carcinoma (HCC) are diagnosed at advanced stages, leading to unfavorable treatment outcomes. Generally, the development of HCC occurs in distinct stages. However, the diagnostic and intervention markers for each stage remain unclear. Therefore, there is an urgent need to explore precise grading methods for HCC. Machine learning has emerged as an effective technique for studying precise tumor diagnosis. In this research, we employed random forest and LightGBM machine learning algorithms for the first time to construct diagnostic models for HCC at various stages of progression. We categorized 118 samples from GSE114564 into three groups: normal liver, precancerous lesion (including chronic hepatitis, liver cirrhosis, dysplastic nodule), and HCC (including early stage HCC and advanced HCC). The LightGBM model exhibited outstanding performance (accuracy = 0.96, precision = 0.96, recall = 0.96, F1-score = 0.95). Similarly, the random forest model also demonstrated good performance (accuracy = 0.83, precision = 0.83, recall = 0.83, F1-score = 0.83). When the progression of HCC was categorized into the most refined six stages: normal liver, chronic hepatitis, liver cirrhosis, dysplastic nodule, early stage HCC, and advanced HCC, the diagnostic model still exhibited high efficacy. Among them, the LightGBM model exhibited good performance (accuracy = 0.71, precision = 0.71, recall = 0.71, F1-score = 0.72). Also, performance of the LightGBM model was superior to that of the random forest model. Overall, we have constructed a diagnostic model for the progression of HCC and identified potential diagnostic characteristic gene for the progression of HCC.
Introduction
Hepatocellular carcinoma (HCC) is a leading cause of cancer-related mortality globally. Its insidious progression—often starting from chronic hepatitis and advancing through cirrhosis and dysplastic nodules before reaching malignancy—makes early detection challenging. Traditional diagnostic methods, such as imaging and standard biopsies, often struggle to differentiate between these subtle stages, particularly the transition from precancerous lesions to early-stage HCC. This study addresses the critical need for precise molecular markers to facilitate early intervention.
Methodology
The research utilized a supervised machine learning workflow to analyze transcriptomic data from the GSE114564 dataset, which includes 118 samples across the full spectrum of HCC progression. Two primary algorithms were evaluated: Random Forest (RF) and Light Gradient Boosting Machine (LightGBM).
The workflow involved several key stages:
- Feature Selection: Initial filtering of genes based on the highest overall variance (top 1,000 genes).
- Model Training: Data was split into training and verification sets (8:2 ratio), with 5-fold cross-validation applied to ensure robustness.
- Identification of "Characteristic Genes": The top 50 genes by importance were extracted from both models. The intersection of these sets defined the "characteristic genes" for HCC progression.
- Downstream Analysis: Functional enrichment (GO) and survival analysis (GEPIA2) were performed to validate the biological and clinical relevance of these markers.
Results
The models were evaluated on two classification tasks: a broad 3-group classification (Normal, Precancerous, HCC) and a granular 6-group classification representing the full progression spectrum.
3-Group Classification
The LightGBM model demonstrated exceptional performance in distinguishing between normal liver, precancerous lesions, and HCC, achieving an accuracy of 96%. The Random Forest model also performed well, though with lower accuracy (83%).
6-Group Classification
In the more challenging 6-group task—which includes chronic hepatitis, liver cirrhosis, and dysplastic nodules—the LightGBM model maintained a respectable accuracy of 71%, significantly outperforming the Random Forest model (63%).
Discussion and Insights
The superiority of LightGBM over Random Forest suggests that gradient boosting is particularly effective at capturing the non-linear patterns inherent in cancer progression data. By focusing on the "intersection" of genes identified by both models, the study isolated a robust signature of 12–16 characteristic genes, including CLEC3B, IGFALS, and VIPR1.
Functional analysis revealed that these genes are primarily involved in complement activation and immune response pathways, highlighting the significant shifts in the liver's immune environment during tumor development. Furthermore, survival analysis indicated that markers like CLEC3B and MASP2 have prognostic value, with low expression correlating with poorer patient outcomes.
Conclusion
This study successfully constructed high-performance diagnostic models for HCC progression using machine learning. The identification of characteristic genes provides a potential molecular roadmap for earlier detection and personalized treatment strategies. Future work should focus on validating these models in larger cohorts and exploring the feasibility of non-invasive detection methods, such as liquid biopsies.
Summary updated