HepScope: CNN-based single-cell discrimination of malignant hepatocytes

Abstract

Background: Hepatocellular carcinoma (HCC) poses a significant global health challenge. This study aims to identify robust single-cell biomarkers specifically designed to distinguish between malignant and non-malignant hepatocytes, thereby advancing precision medicine strategies for HCC. Methods: At the single-cell level, we investigated the reliability of established marker genes for malignant hepatocytes in distinguishing them from their non-malignant counterparts. Leveraging one of the largest integrated single-cell transcriptomics (scRNA-seq) datasets of healthy and diseased liver tissues, alongside independent scRNA-seq, spatial transcriptomics (stRNA-seq), bulk RNA-seq, and bulk proteome datasets, we defined a signature gene set, denoted as HepScope. Five other publicly available gene sets related to HCC were used for comparison with HepScope. The development of machine learning (ML) and 1-dimensional convolutional neural network (1D-CNN) models involved utilizing normalized expression of signature genes from scRNA-seq datasets. The HepScope-1D-CNN model was developed to assess differentiation performance by considering the expression values of each signature gene across different tissues and within the same tissue. Finally, we calculated a risk score based on the signature genes to evaluate its prognostic capability. Result: We demonstrated the lack of specificity and consistency among current HCC marker genes in distinguishing malignant hepatocytes from non-malignant hepatocytes. The differential expression (DEG) analysis identified 113 genes upregulated in malignant hepatocytes, forming the HepScope. In contrast to HepScope, the expression values of the five other publicly available gene sets did not exhibit the ability to discriminate between malignant and non-malignant hepatocytes. Utilizing the genes from HepScope, the Gaussian Naive Bayes ML model proficiently distinguished between malignant and non-malignant hepatocytes. The HepScope-CNN model showed the highest performance and effciency compared to other 1D-CNN models trained with different gene sets. Conclusion: In conclusion, our study reveals a novel gene set designed to precisely identify malignant hepatocytes at the single-cell level. This research holds promise for advancing precision medicine in HCC, providing valuable insights into the development of targeted therapies and personalized treatment strategies. We acknowledge support provided by the National Research Foundation (NRF) of Korea (2020R1A6A1A03043539, 2020M3A9D8037604, and 2022R1C1C1004756) and the Korea Health Technology R&D Project of the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (HR22C1734). This is the full abstract presented at the American Physiology Summit 2024 meeting and is only available in HTML format. There are no additional versions or additional content available for this abstract. Physiology was not involved in the peer review process.