Integrative transcriptomic and CRISPR dependency analysis identifies hepatoblastoma-specific essential genes and...

Hepatoblastoma (HB) is the most common primary liver malignancy in childhood, yet its molecular determinants, functional dependencies, and therapeutic vulnerabilities remain incompletely characterized. Integrative analyses combining transcriptomic profiling with functional genomic datasets provide a strategy to identify essential genes, biomarkers predictive of tumor behavior and treatment response.

Differential expression analysis comparing HB tumors with normal liver was processed on training cohort. These genes were integrated with DepMap CRISPR-Cas9 dependency scores to prioritize HB-essential candidates. Elastic Net regression was used to derive a 16-gene predictive signature, which was validated in an external cohort. Single-cell RNA-seq datasets were analyzed to assess expression patterns across hepatic and tumor-associated cell populations. A supervised deep-learning classifier was trained on single-cell profiles to distinguish tumor cells from hepatocytes, and SHAP values were computed to interpret gene contributions. Drug-gene interactions were queried using curated repressive compounds from DGIdb, and approved drugs were screened for relevance in pediatric cancer clinical trials.

A total of 789 genes were found overexpressed in HB tumors from the training transcriptome cohort. Chronos DepMap analysis identified 73 HB-essential genes that were not essential in adult liver cancer cell lines (hepatocellular carcinoma and cholangiocarcinoma). Elastic-net tuning based on the expression of 16 HB-essential genes in the split training cohort enabled robust tumor-normal discrimination, with AUC = 0.88, specificity = 0.90, and sensitivity = 0.90 in internal validation. This performance was confirmed in an independent external cohort, achieving AUC = 0.99, specificity = 1.00, and sensitivity = 0.98. Single-cell validation further demonstrated tumor-specific enrichment of the signature. The deep-learning classifier (tumor cells vs. normal hepatocytes) reached high accuracy (AUC = 0.99; F1-score = 0.97), with SHAP analysis highlighting PEG10, GREB1, PLCB4, RHOBTB1, CRIM1, FSD1L, CORO2A, KIT, ANKRD50, HDAC11, ZNF233, SEMA7A, and FABP4 as major contributors. Six of these genes were confirmed to be absent or lowly expressed in the background liver microenvironment. Drug-gene interaction analysis identified HDAC11 as a potential therapeutic target of approved drugs used in pediatric oncology.

This integrative framework combining transcriptomics, CRISPR dependency mapping, machine learning, and pharmacogenomic annotation identifies clinically relevant HB-essential genes and predictive molecular signatures for tumor identity. The derived expression-based scores provide tools for patient stratification, while drug-gene mapping highlights actionable vulnerabilities on HDAC11 with pediatric approved drugs that support rational drug repurposing strategies in hepatoblastoma.