Identifying novel transcript biomarkers for hepatocellular carcinoma (HCC) using RNA-Seq datasets and machine...

Area of Science:

• Biochemistry
• Genomics
• Computational Biology

Background:

• Hepatocellular carcinoma (HCC) poses a significant global health challenge due to limited prognostic accuracy.
• Current diagnostic methods, including imaging and serum biomarkers, are insufficient for early detection.
• Transcriptomics offers a promising avenue for identifying more accurate early biomarkers.

Purpose of the Study:

• To leverage RNA-Seq data and machine learning (ML) for identifying novel, high-accuracy transcript biomarkers for HCC.
• To evaluate the efficacy of ML algorithms in pinpointing key molecular signatures indicative of HCC.

Main Methods:

• RNA-Seq data from healthy and HCC cell models were analyzed using five ML algorithms (Random Forest, KNN, Naïve Bayes, SVM, Neural Networks).
• Performance metrics including sensitivity, specificity, and AUC-ROC were evaluated.
• Recursive feature elimination was employed to identify the minimal set of transcripts with maximum discriminatory power.

Main Results:

• Transcriptomics data demonstrated superior efficiency over traditional protein biomarkers for HCC prognosis.
• Random Forest and Support Vector Machine algorithms exhibited the best performance.
• Three transcripts (PARP2-202, SPON2-203, CYREN-211) were identified with 0.97 accuracy and 0.93 kappa, significantly outperforming random selections.

Conclusions:

• RNA-Seq combined with ML effectively identifies novel transcript biomarkers for HCC.
• The identified biomarkers PARP2-202, SPON2-203, and CYREN-211 show high potential for early HCC detection.
• The developed ML pipeline is adaptable for discovering biomarkers in other RNA-Seq datasets.