Abstract 230P
Background
Hepatocellular carcinoma (HCC) is a prevalent digestive tract tumor and a leading global cause of death, posing significant challenges for physicians and patients alike. Incorporating prior knowledge of gene interactions, we extract more informative features from structural data to enhance patient prognosis characterization. By constructing a specific edge network of gene interactions, we establish a robust HCC prognostic model, yielding an average C-index of 0.76.
Methods
We collected public datasets from TCGA, GEO, and CLCA, retaining 1308 samples with complete follow-up information. The dataset comprised TCGA-LIHC, ICGC-LIRI , GSE10141, GSE116174, GSE14520, GSE54236, GSE76427, and CLCA. Utilizing TCGA-LIHC, GSE10141, GSE116174, GSE14520, and CLCA as the training set, and others as the test set, we constructed a single-sample gene interaction network. By integrating various machine learning methods and hyperparameter search strategies, we identified the best-performing model in the test set, designated as the Gene Interaction Predictor (GIP).
Results
GIP effectively stratifies the training cohort into high-risk and low-risk groups, showing significant differences in overall survival rates. Despite variations in time distribution among cohorts, GIP maintains robust performance. In the test cohort, GIP also demonstrates superior performance, outperforming 20 published HCC prognostic models compatible with our dataset.
Conclusions
In this study, we established a superior and robust HCC prognostic model by constructing a patient-specific edge network based on gene interactions, which outperforms existing ones.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Anhui Medical University.
Disclosure
All authors have declared no conflicts of interest.