2024-04-24
Presentation by Izzy Grabski
Framework to deconvolute drug targets using CRISPR perturbation screens.
Statistical framework: 1. Estimate denoised perturbation effects from perturbation screen data. This can be confounded by cell cycle and other sources of baseline hetrogeneity. Multi-condition latent factor model. Non-targeting cells will have only baseline variation. Cells receiving perturbation will have an addition perturbation component. 2. Deconvolve drug-treated cells in terms of these perturbation effects. Scale, sparsity and uncertainty. Analogous to fine-mapping in GWAS. Use SuSiE. 3. Calibrate estimated probabilities within credible sets.
CPA-Perturb-Seq (Kowalski, 2023). Perturbation screens of cleavage and polyadenylation regulators in HEK cells.
K562 cells in Genome-wide Perturb-Seq (Replogle et al, 2022)
Application on HDACs.