Cell type classification and discovery across diseases, technologies and tissues reveals conserved gene signatures and enables standardized single-cell readouts

Abstract

Autoimmune diseases are a major cause of mortality. Current treatments often yield severe insult to host tissue. It is hypothesized that improved therapies will target pathogenic cells selectively and thus reduce or eliminate severe side effects, and potentially induce robust immune tolerance. However, it remains challenging to systematically identify which cellular phenotypes are present in cellular ensembles. Here, we present a novel machine learning approach, Signac, which uses neural networks trained with flow-sorted gene expression data to classify cellular phenotypes in single cell RNA-sequencing data. We demonstrate that Signac accurately classified single cell RNA-sequencing data across diseases, technologies, species and tissues. Then we applied Signac to identify known and novel immune-relevant candidate drug targets (n = 12) in rheumatoid arthritis. A full release of this workflow can be found at our GitHub repository (https://github.com/mathewchamberlain/Signac).

Multi-Omics Characterization of Tumor Microenvironment Heterogeneity and Immunotherapy Resistance Through Cell States–Based Subtyping in Bladder Cancer

“…Then, integration of eight data sets was performed by the function “IntegrateData,” and batch effects were removed using the SCTransform algorithm in Seurat. Finally, the remaining 70,387 cells were applied to SignacX 2.2.4 ( Chamberlain et al, 2021 ), a neural network–based approach to identify cell states. Next, the cell states were further validated based on the CellMarker database ( Zhang et al, 2019 ), followed by subclustering to obtain 22 cell states, which were visualized with UMAP plots.…”

DiSiR: a software framework to identify ligand-receptor interactions at subunit level from single-cell RNA-sequencing data

“…Afterwards, we used SPRING [37], a web-based software tool to visualize high dimensional transcriptomic data, to visualize the data, with 60 (COVID) and 30 principal components (AMP RA Phase 1). We then annotated the cells with SignacX version 2.2.0 [38] – a neural-network-based cell annotation tool that uses expression data and is also aided by the edges from the SPRING plot. Finally, we transferred the resulting annotated expression data into log2 space for DiSiR analysis.…”

A novel cell type negatively associated with secondary autoimmunity in alemtuzumab-treated patients is revealed through single-cell longitudinal analysis of clinical trial samples

“…As described previously in our single-cell optimization study 4 , cryopreserved PBMCs were thawed (2 vials at a time) in a 37°C water bath for 1-2 minutes until a small crystal remained. The cryovial was removed from the water bath and cell solution was transferred to a sterile 2 mL Eppendorf tube using a wide bore pipet tip.…”

“…Single cell analysis was carried out as described previously 4 . Briefly, following BCL conversion, FASTqs were processed through CellRanger version v2.1.1 for demultiplexing, alignment, filtering, barcode counting, UMI counting, and generating of gene x barcode matrices.…”

“…1 ). To process this dataset, we developed a high-throughput cloud-based pipeline with large dataset capacity to dimensionally reduce, visualize, and automatically label the data with known cell-type and subtype labels based on transcriptional similarity with sorted bulk profiles 4 ( Fig. 1C ).…”