Fig. 1

Systematic benchmarking of single-cell multi-omic analysis methods can involve experimental data (as per our hackathons), custom control datasets, where a known structure is imposed through the experimental design, or simulated data. The amount of biological signal and ground truth available varies considerably between these types of data. The generated multi-omics datasets are analyzed by competing methods and the results are compared using generic metrics (e.g., method stability, runtime, consistency via data splitting) or by considering the ground truth available (e.g., cell type labels or number of cell types)