CRISPR-Cas9 screening enables systematic identification of gene dependencies across diverse cancer cell lines. However, distinguishing lineage-specific vulnerabilities from broadly essential genes remains a key challenge for precision oncology. In this study, we analyzed DepMap 22Q1 data to develop a selectivity-centered computational framework for prioritizing cancer-type-specific dependencies. A selectivity score was defined as the difference between the median dependency within a given cancer type and the global median across all cell lines. To enrich lineage-informative signals, the top 500 genes ranked by dependency variance were retained. Candidate targets were identified using thresholds of Chronos score < -1.0 and selectivity score < -0.3, followed by one-sided Wilcoxon rank-sum testing with Benjamini–Hochberg correction (q < 0.05). A total of 1,991 candidate gene–cancer pairs were identified across 43 cancer types, among which 87 were statistically significant. Hematological malignancies showed strong enrichment of selective dependencies. The top-ranked association, ABL1 in myeloproliferative neoplasms, is a clinically validated target, supporting the reliability of the framework. Functional enrichment analysis revealed a strong association with purine metabolism, nucleotide biosynthesis, and methylation-related processes. Notably, 9 of the top 10 targets were metabolic genes. Importantly, comparison between selectivity-based ranking and statistical significance demonstrated that integrating effect size with significance improves prioritization robustness relative to single-metric approaches. These findings suggest that selectivity-driven analysis provides a complementary perspective to conventional dependency ranking strategies. This study presents a reproducible workflow for identifying lineage-specific CRISPR dependencies and offers candidate targets for further experimental validation.