Development and Application of Analytical Methods for Genetic Interaction Mapping from Combinatorial CRISPR Screens in Human Cancer Cells

Juihsuan (Rosalind) Chou, MS (Advisory Professor: Traver Hart, PhD)

Genetic interactions, quantitative deviations of double-mutant phenotypes from expected independent effects, reveal functional relationships among genes that are invisible to single-gene perturbation studies and hold direct therapeutic relevance through the principle of synthetic lethality. The development of combinatorial CRISPR technologies has made it feasible to systematically map pairwise genetic interactions in human cancer cells; however, the field lacks standardized analytical methods and objective benchmarking resources. This dissertation addresses these gaps through three interconnected aims: comparative evaluation of scoring methods for paralog synthetic lethality screens, development of a regression-based analytical framework and simulation-based benchmarking platform, and application of optimized methods to discover biologically and clinically relevant genetic interactions within the DNA damage response (DDR) and receptor tyrosine kinase (RTK) signaling networks.

The first part of this dissertation addresses the question of optimal analytical methodology for paralog synthetic lethality screens. Using data from four independent enCas12a-based in4mer combinatorial knockout screens conducted across cancer cell lines, three delta log fold change (dLFC)-based methods were compared based on cross-screen hit consistency. This analysis confirmed the importance of normalizing each dataset against a data-derived null model. Re-analysis of three previously published paralog screens confirmed the generalizability of the optimal method across both Cas9- and Cas12a-based platforms. Approximately 50% overlap between independent screens was established as a reasonable benchmark, with the non-overlapping fraction comprising an inseparable mixture of false positives, false negatives, and true background-specific interactions.

Second, GRAPE (Genetic interaction Regression Analysis of Pairwise Effects) was developed, a regression-based framework that estimates single-gene fitness effects from shared structure across dual-knockout measurements and identifies statistically significant deviations from additive expectation. The companion simulation framework Synulator enables objective benchmarking by generating realistic screen data with known ground-truth interactions. GRAPE outperformed existing methods and generalized across published screens spanning three CRISPR modalities.

Third, GRAPE was applied to two complementary experimental designs targeting the DNA damage response (DDR) and receptor tyrosine kinase (RTK) modules across up to 12 cancer cell lines. These screens identified hundreds of synthetic lethal and suppressor interactions, including structured synthetic lethal architecture centered on the PCNA-loading RFC complex and a previously uncharacterized interaction network within the ER-localized protein glycosylation pathway. The most robust interactions were validated in three-dimensional organoid and in vivo xenograft models, demonstrating that genetic interactions discovered in monolayer culture persist in physiologically relevant contexts.

Together, this dissertation provides validated analytical tools, an objective benchmarking framework, and biologically relevant genetic interaction discoveries that advance the systematic mapping of genetic interaction networks in human cancer cells.

Advisory Committee:

• Traver Hart, PhD, Chair
• Swathi Arur, PhD
• Paul Scheet, PhD
• John Paul Shen, MD
• Han Xu, PhD