Precise identification of cell states altered in disease with healthy single-cell references

Abstract

Single cell genomics is a powerful tool to distinguish altered cell states in disease tissue samples, through joint analysis with healthy reference datasets. Collections of data from healthy individuals are being integrated in cell atlases that provide a comprehensive view of cellular phenotypes in a tissue. However, it remains unclear whether atlas datasets are suitable references for disease-state identification, or whether matched control samples should be employed, to minimise false discoveries driven by biological and technical confounders. Here we quantitatively compare the use of atlas and control datasets as references for identification of disease-associated cell states, on simulations and real disease scRNA-seq datasets. We find that reliance on a single type of reference dataset introduces false positives. Conversely, using an atlas dataset as reference for latent space learning followed by differential analysis against a matched control dataset leads to precise identification of disease-associated cell states. We show that, when an atlas dataset is available, it is possible to reduce the number of control samples without increasing the rate of false discoveries. Using a cell atlas of blood cells from 12 studies to contextualise data from a case-control COVID-19 cohort, we sensitively detect cell states associated with infection, and distinguish heterogeneous pathological cell states associated with distinct clinical severities. Our analysis provides guiding principles for design of disease cohort studies and efficient use of cell atlases within the Human Cell Atlas.

Establishing a human bone marrow single cell reference atlas to study ageing and diseases

“…However, a comprehensive healthy tissue atlas is still valuable to disease studies. As recently demonstrated by Dann et al (41), using a separate healthy reference as a reference scaffold to map disease samples and matched healthy controls can improve the identification of disease-associated cell states and reduce the number of control samples while preserving the rate of false discoveries. Here, our constructed healthy BM atlas encompassing a large number of donor samples from multiple studies and across a wide age range, is a comprehensive healthy reference that can serve as a baseline for comparative studies with diseased samples.…”

Identification of cell types, states and programs by learning gene set representations

“…We then tested for the differential activity of the 40 selected gene sets in severe COVID-19 versus healthy in each neighborhood group using the testDiffExp() function in miloR, which calls limma under the hood akin to differential gene expression. These analyses were performed within query data, on the integrated embedding, that is post reference mapping, which is found to result in optimal detection of disease-specific cell states using healthy cell atlases 47 . To summarize differential pathway activity over all the neighborhood groups in a cell type, we reported the most significant t-statistic (smallest adjusted P-Value from limma’s moderated t-statistic) in neighborhood groups of a cell type.…”

Scvi-hub: an actionable repository for model-driven single cell analysis

“…At present, transfer learning is primarily used for projecting cells onto a common low-dimensional representation that is employed for tasks such as annotation or trajectory inference (Lotfollahi et al 2022; Kang et al 2021; Hao et al 2022). However, moving forward, additional applications will become more prevalent, such as interpretation of spatial data (Lopez et al 2022) prediction of perturbation outcome (Roohani, Huang, and Leskovec 2023), prediction of multi-modal information from single modality data (Ashuach et al 2023), detection of anomalous cellular subsets (Dann et al 2023), or more robust analysis of differential expression (Boyeau et al 2023).…”