Eleven grand challenges in single-cell data science

Table 4 Approaches for data integration, highlighting their promises and challenges

	Integration	Example MT combination	Example AMs	Promises	Challenges
1S	None	scDNA-seq	Clustering/unsupervised	Discover new subclones, cell types, or cell states	Technical noise ↓; data sparsity ↓
+S	Within 1 MT, within 1 exp, across >1 smps	scRNA-seq	Differential analyses, time series, spatial sampling	Identify effects across sample groups, time, and space	Batch effects ↓; validate cell type assignments ↓
+X+S	Within 1 MT, across >1 exp, across >1 smps	merFISH	Map cells to stable reference (cell atlas)	Accelerate analyses, increase sample size, generalize observations	Standards across experimental centers
+M1C	Across >1 MTs, within 1 exp, within 1 cell	scM&T-seq (scRNA-seq + methylome)	MOFA, DIABLO, MINT	Holistic view of cell state; quantify dependency of MTs	Scaling cell throughput; MT combinations limited; dependency of MTs ↓
+M+C	Across >1 MTs, within 1 exp, across >1 cells, within 1 cell pop	scDNA-seq + scRNA-seq	Cardelino, Clonealign, MATCHER	Use existing datasets (faster than +M1C); flexible experimental design	Validate cell/data matching; test assumptions for integrating data
+all	Across >1 MTs, across >1 exps, across >1 smps, within cells	Hypothetical (any combination)	Hypothetical (map cells to multi-omic HCA, single-cell TCGA)	Holistic view of biological systems	All from approaches +X+S, +M1C, and +M+C

The labeling corresponds to Fig. 6. For each approach, one (combination of) measurement type(s) that is available is given, but more exist and several are discussed in the text. As example analysis methods, actual tool names are given where few tools exist to date; otherwise, broader categories or imaginable methodologies are described
Abbreviations: “↓” same challenge also applies to all approaches below, AM analysis method, exp(s) experiment(s), HCA human cell atlas, MT measurement type, smps samples, TCGA The Cancer Genome Atlas

ISSN: 1474-760X