Causal modelling and adaptive experimental design for single-cell perturbation screens

点击翻译按钮获取中文摘要

Gene expression is controlled by a complex network of transcription factors and epigenetic regulators and defines cellular, tissue and organismal phenotypes. Gene regulatory networks (GRNs) capture thee complex interactions and can be used to describe gene expression patterns in health and disease. A causal understanding of the underlying mechanisms within these networks is required to identify the factors and processes that regulate development and disease and predict druggable targets. However, functional data resolving these causal interactions is not available systematically and modelling approaches that provide quantitative predictions of perturbation effects across different GRNs and cell types are still in their infancy. In this project, we develop and validate an adaptive experimental design methodology, based on causal generative modelling, and apply it to dissect GRNs and their causal interactions in a systematic iterative fashion in human organoid model systems. Our project will apply the developed methods to increase our understanding of GRNs governing neural and cardiomyocyte differentiation, and their interaction with known risk-genes for psychiatric and cardiovascular disease. Thereby, we aim to learn more about the gene regulatory logic required for normal organ development and function, how gene regulation is changed during disease, and which genetic elements need to be targeted to achieve a maximum therapeutic effect. To these ends, we will systematically perturb GRNs by performing CRISPR/Cas9 pooled genetic screens in stem cell derived heart and brain organoid model systems. These will allow us to partially recreate the architecture and physiology of these human organs, which are associated with the highest global disease burden. Therefore, we will dissect GRNs relevant for cardiomyocyte and neural differentiation, which interact with disease associated genetic variants for a more precise prediction of therapeutic intervention points. To readout the effects of CRISPR perturbations, we will use targeted perturbation sequencing (TAP-seq, developed in the Steinmetz lab). The perturbation screens will be performed in an iterative fashion, and will employ an adaptive experimental design that utilises combinatorial gene-repressing (CRISPRi) and/or gene-activating (CRISPRa) perturbations guided by causal generative modelling. Thereby, we will leverage probabilistic estimates of causal relationships between genes to quantify the expected information gain associated with potential perturbations, and select a set of maximally informative interventions for the next experiment. This approach will allow us to efficiently probe GRNs in a causal fashion, provide a detailed understanding of gene regulatory mechanisms in health and disease and provide novel tools to predict the outcome of therapeutic interventions within those GRNs.