Inference Methods for High-Throughput CRISPR Screens - CGSI 2022

Explore inference methods for high-throughput CRISPR screens in this 46-minute conference talk from the Computational Genomics Summer Institute (CGSI) 2022. Delve into the complexities of gene contribution to complex traits and learn how upstream regulators can be inferred through perturbations. Examine the combination of Fluorescence Activated Cell Sorting (FACS) and CRISPR techniques for high-throughput gene expression screening. Analyze the computational aspects of the setup, starting with the research question and considering unique data carefully. Understand the correlation between multiple guides targeting the same gene and investigate the sampling distribution. Study the model as a form of density estimation with overdispersion, linking unobserved reports to the sampling distribution. Discover how the updated model incorporates sparsity at the gene level and how the hierarchical model enables accurate inference with few samples. Gain insights from related papers on inferring expression changes in sorting-based CRISPR screens and the systematic discovery and perturbation of regulatory genes in human T cells.

Intro
How can so many genes contribute to complex traits?
Upstream regulators can be inferred by perturbations
Fluorescence activated cell sorting (FACS) + CRISPR enable high-throughput gene expression screening
Review of setup from the computational side
Start with the question
We have unique data - let's think carefully
Multiple guides target the same gene and thus should be correlated
What is my sampling distribution?
The model is a form of density estimation with overdispersion
Our updated model links the unobserved reported to the sampling distribution
Our new model incorporates sparsity at the gene- level
Hierarchical model enables accurate inference with few samples