Machine learning identifies exosome features related to hepatocellular carcinoma

Area of Science:

• Oncology
• Biomarkers
• Genomics

Background:

• Hepatocellular carcinoma (HCC) is a highly malignant cancer with a poor prognosis.
• Effective biomarkers for predicting HCC prognosis are lacking.
• Exosomes play a crucial role in cancer development and progression through intercellular communication.

Purpose of the Study:

• To develop and validate a novel prognostic signature for hepatocellular carcinoma (HCC).
• To identify exosome-related genes (ERGs) that can predict HCC patient survival.
• To assess the clinical utility of a random forest (RF) based signature for HCC prognosis.

Main Methods:

• Machine learning algorithms, including univariate feature selection and random forest (RF), were employed to select 13 ERGs and construct a prognostic signature.
• A novel RF signature was generated based on ERG signature and pathway scores.
• Expression of BSG and SFN was validated using RT-qPCR and immunohistochemistry; functional assays (CCK-8) assessed their impact on cell proliferation.

Main Results:

• The developed RF signature demonstrated excellent prognostic ability for HCC, with AUC values of 0.845 (1 year), 0.811 (2 years), and 0.801 (3 years).
• The RF signature outperformed a previously developed ERG signature in predicting HCC prognosis.
• Elevated BSG and SFN expression in HCC tissues correlated with suppressed cell proliferation upon inhibition, suggesting their oncogenic roles.

Conclusions:

• The novel RF signature accurately predicts the prognosis of hepatocellular carcinoma (HCC) patients.
• This RF signature holds significant potential for clinical application in HCC patient management.
• BSG and SFN are potential therapeutic targets in HCC, as their inhibition suppressed proliferation.