Testing and Estimation for Multi-Modality Single Cell Genomics

This project aims to develop new statistical analysis methodologies for data produced by a new technology known as single cell CRISPR screens, which promise to help unravel the molecular mechanisms of human disease and guide drug development. This technology discovers the functions of crucial but poorly understood genomic regions by knocking them out and then measuring the impact on the cell.

Given these possibilities, this technology has attracted enormous interest in both academia and industry, leading to its rapidly growing adoption. However, the statistical analysis of the data produced by CRISPR screens presents a major challenge to realizing their promise. The PI will tackle several of the most important statistical analysis challenges presented by this technology and implement the resulting methodologies in software tools.

These tools will help scientists draw reliable conclusions from their single cell CRISPR screen data and therefore accelerate progress in the prognosis, diagnosis, and treatment of human disease. The project will provide training opportunities to graduate students by involving them in the research.

The PI recently leveraged the distilled conditional randomization test (dCRT) methodology to design SCEPTRE, a computationally efficient, well-calibrated, and powerful analysis method for single cell CRISPR screens. In this project, the PI will expand his recent work on association testing for single cell CRISPR screens to more challenging problem settings (non-interventional screens) and more challenging inferential tasks (estimation and interaction detection).

First, the PI will extend the ideas underlying SCEPTRE to address the problem of association testing in multi-omics screens, the observational counterparts of CRISPR screens. Second, the PI will develop a procedure to estimate the effect of knocking out a regulatory element on the expression of a gene based on single cell CRISPR screens. This will require more carefully accounting for the indirect measurement mechanism of CRISPR perturbations.

Third, the PI will design a test of interaction between two regulatory elements, also based on CRISPR screens. This problem is even harder but critical to capturing the complexity of gene regulation. Finally, the resulting methods will be implemented in an integrated software package designed for practical application to real single cell data.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.