Recent advancements in functional genomics have provided an unprecedented ability to measure diverse molecular modalities, but learning causal regulatory relationships from observational data remains challenging. Here, we leverage pooled genetic screens and single cell sequencing (i.e. Perturb-seq) to systematically identify the targets of signaling regulators in diverse biological contexts. We demonstrate how Perturb-seq is compatible with recent and commercially available advances in combinatorial indexing and next-generation sequencing, and perform more than 1,500 perturbations split across six cell lines and five biological signaling contexts. We introduce an improved computational framework (Mixscale) to address cellular variation in perturbation efficiency, alongside optimized statistical methods to learn differentially expressed gene lists and conserved molecular signatures. Finally, we demonstrate how our Perturb-seq derived gene lists can be used to precisely infer changes in signaling pathway activation for in-vivo and in-situ samples. Our work enhances our understanding of signaling regulators and their targets, and lays a computational framework towards the data-driven inference of an ‘atlas’ of perturbation signatures.
False positive claims of differentially expressed genes (DEGs) in scRNA-seq studies are of substantial concern. We found that DEGs from individual Parkinson’s (PD), Huntington’s (HD), and COVID-19 datasets had moderate predictive power for case-control status of other datasets, but DEGs from Alzheimer’s (AD) and Schizophrenia (SCZ) datasets had poor predictive power. We developed a non-parametric meta-analysis method, SumRank, based on reproducibility of relative differential expression ranks across datasets, and found DEGs with improved predictive power. Specificity and sensitivity of these genes were substantially higher than those discovered by dataset merging and inverse variance weighted p-value aggregation methods. Up-regulated DEGs implicated chaperone-mediated protein processing in PD glia and lipid transport in AD and PD microglia, while down-regulated DEGs were in glutamatergic processes in AD astrocytes and excitatory neurons and synaptic functioning in HD FOXP2 neurons. Lastly, we evaluate factors influencing reproducibility of individual studies as a prospective guide for experimental design.
We assessed the reproducibility of differentially expressed genes (DEGs) in previously published Alzheimer’s (AD), Parkinson’s (PD), Huntington’s (HD), Schizophrenia (SCZ), and COVID-19 scRNA-seq studies. While transcriptional scores from DEGs of individual PD, HD, and COVID-19 datasets had moderate predictive power for case-control status of other datasets, genes from individual AD and SCZ datasets had poor predictive power. We developed a non-parametric meta-analysis method, SumRank, based on reproducibility of relative differential expression ranks across datasets, and found DEGs with improved predictive power. By multiple metrics, specificity and sensitivity of these genes were substantially higher than those discovered by dataset merging and inverse variance weighted p-value aggregation methods and had significant enrichment in snATAC-seq peaks and human disease gene associations. The DEGs revealed known and novel biological pathways, such as up-regulation of chaperone-mediated protein processing in PD glia and lipid transport in AD and PD microglia, and down-regulation of glutamatergic processes in AD astrocytes and glutamatergic neurons and synaptic processing and neuron projection genes in HD FOXP2 neurons. We find 56 DEGs shared amongst AD, PD, and HD, and validate BCAT1 as down-regulated in AD mouse oligodendrocytes. Lastly, we evaluate factors influencing reproducibility of individual studies as a prospective guide for experimental design.
Single-cell genomics is transformative for characterizing cellular heterogeneity, but many translational questions require comparing samples rather than individual cells. Typical analyses compare "case" and "control" groups, ignoring sample-level variation within them. Here we present scSLIDE, a framework that transforms each sample's single-cell data into a compact profile describing where its cells fall in high-dimensional space. By comparing these density profiles, scSLIDE calculates embeddings to characterization variation across samples. Applied to COVID-19 infection, Alzheimer's disease, and zebrafish embryogenesis, we show that scSLIDE can be used to cluster patients, reconstruct sample-level disease trajectories, and identify coordinated cellular programs across samples. We discover independent axes of infection and severity, a molecular disease progression that aligns with pathology-based estimates of neurodegeneration, and embryonic "pseudostages" varying across and within timepoints. In each case, we demonstrate how case-control analyses collapse rich biological variation into binary labels, and how sample-level embedding represents a powerful analytical alternative.