We are excited to share Genotype-Neighborhood Associations, GeNA, a new tool adapting our CNA framework to detect cell states associated in abundance with genetic variants at genome-wide scale in high-dimensional single-cell data w/ @soumya_boston 🧵.

6/6 You can find and try GeNA at Thank you to all the GeNA team members—@saorisakaue, Y Reshef, @joycebkang, P-R Loh and more—and to the donors and analysts of the published datasets that made this work possible!.

5/6 csaQTL analyses may help illuminate the convergence of effects from distinct risk loci for the same disease on shared functional pathways. We find evidence for increased IFN-a signaling, a hallmark of SLE, associated with total SLE genetic risk in the absence of disease.

4/6 For example, rs3003-T associates with increased fractional abundance of NK cells expressing TNF-α response programs out of all NK cells. This csaQTL colocalizes with increased risk for psoriasis, an autoimmune disease that responds to anti-TNF treatments.

3/6 In the OneK1K dataset (PBMC scRNA-seq, N~1000), GeNA reveals novel cell state abundance QTLs that replicate in independent datasets and colocalize with immune-mediated disease risk loci, offering insight into tissue composition changes that may contribute to disease.

2/6 In genetic studies, well calibrated statistics are essential to avoid false associations when testing many variants. This issue might increase when considering many cell states, too, but we demonstrate that GeNA sustains well-controlled type 1 error in permutation analyses.

1/6 It is essential but difficult to map how disease risk alleles affect the composition of cells in the body. Single-cell datasets offer unbiased, granular profiling of cell states in diverse tissues, but we lack tools to flexibly identify variant-associated states in such data.

Bonus 8/7💫 An updated schematic with the a-e neighborhood labels corrected.

7/7 You can find and try the CNA Python package at: A huge thank you to our awesome co-authors @joycebkang @aparnanathan @megan_b_murray and Branch Moody!.

6/7 We apply CNA to three datasets published with cluster-based analyses to illustrate that CNA uncovers important structure in these data and improves characterization of immune states altered in rheumatoid arthritis, sepsis, and patients who progress to active tuberculosis.

5/7 CNA can be used to study diverse sample attributes (categorical and continuous): clinical, environmental, genetic, and demographic factors. CNA is also scalable; running CNA on a KNN graph for a single-cell dataset of 271 samples and 500k cells takes <1min.

4/7 We confirm in simulation that CNA has more statistical power to find associations and greater accuracy in localizing driving cell states than cluster-based analysis across a range of signal types (e.g. clusters, gene expression programs)

3/7 CNA does this by identifying groups of very small regions in transcriptional space—termed neighborhoods—that covary in abundance across samples, suggesting shared function or regulation, and modeling sample-level attributes as determined by abundances of these groups.

2/7 The goal of CNA is to identify associated cell states with more flexibility and to achieve this quickly, without parameter tuning, and with strong statistical power.

1/7 Characterizing cell states that vary across samples and associate with sample attributes can improve our understanding of disease, treatment and basic physiology, but most current approaches only examine changes in the abundance of clusters.

Excited to share covarying neighborhood analysis (CNA), a more powerful method for identifying fine-grained cell states in single-cell data associated with sample attributes (e.g. disease status) than cluster-based analyses, w/ Y. Reshef @soumya_boston: �.