Manu Setty (@SettyM) profile picture

Assistant Professor, @fredhutch @HutchBasicSci & Translational Data Science IRC. Computational biologist | Single-cell data analysis and modeling.

Seattle, WA
Joined January 2011
Incredibly easy, fast and efficient process. Thank you @KingCountyWA @kcelections !
1
183
16 years after arriving in the United States, I am honored and proud to cast my first vote in a Presidential election to a daugher of immigrants, @VP @KamalaHarris !
3
1
38
3,044
Finally, Mellon is blazing fast - density for 100k cells takes only a few minutes with a (near) linear dependence of computation time on number of cells.
1
1
389
The function is also differentiable, which means we can interpolate the change in density too! The measured time points are grey lines at the bottom - all other times are interpolated densities and changes.
1
2
443
Mellon framework can be extended to infer a density function that is continuous in both cell-state and time. We use this to interpolate densities between measured time points in time-series datasets. Thanks to @MarioniLab for the repo to download data: bioconductor.org/packages/re… !
1
1
440
We also developed a procedure to disentagle primed and lineage-specific peaks from single-cell ATAC data (rather than one gene score), and show that enhancer priming plays a role in driving low-density transitions.
1
2
433
We developed a gene change score procedure to identify genes with rapid changes in low-density regions. EBF1, B-cell master upregulator is top the gene during B-cell specification with upregulation exquisitely localized to the low-density transitory cells.
1
461
We applied Mellon to hematopoiesis and expectedly recovered the rich heterogeneity in density landscape. Using our Palantir algorithm, we show that first low-density regions correspond to fate specification in all hematopoietic lineages. These cells constitute <0.1% of data.
1
2
525
The result is a continuous density function that serves as a comprehensive description of the phenotypic landscape and can be used to evaluate densities at single-cell resolution!
1
1
541
We leverage the connection between nearest neighbor and density, and the observation that closely related cells share similar density in a Bayesian Model to robustly estimate densities.
1
1
542
The goal of Mellon is to estimate accurate high-resolution densities in single-cell data that remains robust in low-density areas. High-density regions likely correspond to major cell-types whereas the low-density regions are the biologically action-packed rare transitory cells.
1
1
2
1,526