Scalable Transcriptional Measurements of Genetic Manipulations    

Oren, Bi,....Atray..., Aviv


By combining droplet based single cell transcriptomics with CRISPR-Cas based perturbations we demonstrate an approach that will allow researchers to perform thousands of genetic manipulations in a single pooled experiment. This pooled approach represents an approximate 10 fold improvement in cost over current methods. By increasing the number of cells per droplet we demonstrate that for a small reduction in sensitivity for differentially expressed genes that decrease in expression we gain an additional increase in sample size.  Finally, by randomly lentivirally integrating several sgRNAs in each cell, we show the ability to extend the method to perform combinatorial screens. The random structure of the experimental design matrix, so constructed, satisfies the incoherence properties in a compressed sensing framework. Together, these statistical and experimental methods will enable researchers to perform large scale screening of perturbations, including systematic dissection of epistatic effects, using RNA transcription as a phenotype.


One sentence 5-year old summary of intent:
It's hard to remove lots of genes and see their effect, even harder to pull out groups of genes. The field needs scalable ways to systematically dissect gene circuits and causal relations. 

Classical genetic approaches to mapping genotypes to phenotypes leveraging population diversity can be confounded by layers of molecular machinery and length scales between the two. Experiments measuring the direct transcriptional effect of introduced mutations scale linearly with number of experiments in a way that prohibits large scale testing (largest experiments to our knowledge assay hundreds of perturbations followed by RNA-seq...Oren's CRISPR paper or project Achilles). eQTL mapping, while effective, is not suited for examining how gene inactivation effects percolate through genetic networks. 
Figure 1: Overview of approach. Poisson loading of sgRNA followed by poisson loading of cells into droplets for scalable single cell transcriptomics yields an expression matrix of  cells. By matching polyadenylated RNA barcodes to the guide RNA a perturbation matrix is resolved. By fitting a coefficient matrix for each gene and each perturbation, the transcriptome wide effect of a knockout can be mapped with high confidence.  Additional covariates such as library complexity, cell cycle, and cell state effects can also be considered using intergenic/nontargeting guides as a control.