Abstract
Rapid advances in single-cell-, spatial-, and multi-omics, allow us to profile cellular ecosystems in tissues at unprecedented resolution, scale, and depth. However, both technical limitations, such as low spatial resolution and biological variations, such as continuous spectra of cell states, often render these data imperfect representations of cellular systems, best captured as continuous mixtures over cells or molecules. Based on this conceptual insight, we build a versatile framework, TACCO (Transfer of Annotations to Cells and their COmbinations) that extends an Optimal Transport-based core by different wrappers or boosters to annotate a wide variety of data. We apply TACCO to identify cell types and states, decipher spatio-molecular tissue structure at the cell and molecular level, and resolve differentiation trajectories. TACCO excels in speed, scalability, and adaptability, while successfully outperforming benchmarks across diverse synthetic and biological datasets. Along with highly optimized visualization and analysis functions, TACCO forms a comprehensive integrated framework for studies of high-dimensional, high-resolution biology.
Competing Interest Statement
A.R. is a co-founder and equity holder of Celsius Therapeutics, an equity holder in Immunitas, and was an SAB member of ThermoFisher Scientific, Syros Pharmaceuticals, Neogene Therapeutics and Asimov until July 31, 2020. From August 1, 2020, A.R. is an employee of Genentech and has equity in Roche. F.C. is a founder and holds equity in Curio Biosciences. A.R. and O.R.-R. are co-inventors on patent applications filed by the Broad Institute for inventions related to single cell genomics. O.R.-R. has given numerous lectures on the subject of single cell genomics to a wide variety of audiences and in some cases, has received remuneration to cover time and costs. O.R.-R. is an employee of Genentech since October 19, 2020 and has equity in Roche.
Footnotes
↵# These authors jointly supervised this work
1 Setting n(tb) to p(tb) one can reproduce the result from Cable et al. by determining n(gb) from the t-marginal, ignoring the g-marginal i.e. the annotation result.
2 Taken from https://codeocean.com/capsule/6314882/tree/v1, code/analysis/Part1_Simulations/Step1_Simulate.ipynb