Integrated analysis of single cell transcriptomic data across conditions, technologies, and species

Abstract

Single cell RNA-seq (scRNA-seq) has emerged as a transformative tool to discover and define cellular phenotypes. While computational scRNA-seq methods are currently well suited for experiments representing a single condition, technology, or species, analyzing multiple datasets simultaneously raises new challenges. In particular, traditional analytical workflows struggle to align subpopulations that are present across datasets, limiting the possibility for integrated or comparative analysis. Here, we introduce a new computational strategy for scRNA-seq alignment, utilizing common sources of variation to identify shared subpopulations between datasets as part of our R toolkit Seurat. We demonstrate our approach by aligning scRNA-seq datasets of PBMCs under resting and stimulated conditions, hematopoietic progenitors sequenced across two profiling technologies, and pancreatic cell 'atlases' generated from human and mouse islets. In each case, we learn distinct or transitional cell states jointly across datasets, and can identify subpopulations that could not be detected by analyzing datasets independently. We anticipate that these methods will serve not only to correct for batch or technologydependent effects, but also to facilitate general comparisons of scRNA-seq datasets, potentially deepening our understanding of how distinct cell states respond to perturbation, disease, and evolution.

Benchmarking principal component analysis for large-scale single-cell RNA-sequencing

“…PCA is widely used for data visualization [39][40][41], data QC [42], feature selection [13,[43][44][45][46][47][48][49], de-noising [50,51], imputation [52][53][54], confirmation and removal of batch effects [55][56][57], confirmation and estimation of cell-cycle effects [58], rare cell type detection [59,60], cell type and cell state similarity search [61], pseudotime inference [13,[62][63][64][65][66], and spatial reconstruction [9].…”

“…In particular, such implementations are effective for large data matrices stored in 10X-HDF5 format using CSC format. Seurat2 [49] also introduces this approach by combining the matrix market format (R, Matrix ) and irlba function (R, irlba). When the data matrix is dense and cannot be loaded into memory space (e.g., GC ≥ 10 10 ), the out-of-core implementations, such as oocPCA CSV (R, oocRPCA), IncrementalPCA (Python, sklearn), and algorithm971 (Julia, OnlinePCA.jl ), are useful (dense matrix & out-of-core).…”

“…Although the down-stream analyses of PCA vary widely, and we could not examine all the topics of scRNA-seq analyses, such as rare cell-type detection [59,60] and pseudotime analysis [13,[62][63][64][65][66], differences among PCA algorithms might also affect the accuracy of such analyses. Butler et al showed batch effect removal can be formalized as canonical correlation analysis (CCA) [49], which is mathematically very similar to PCA. The optimization of CCA is also formalized in various ways, including randomized CCA [131] or SGD of CCA [132].…”

“…Scanpy [93] IRLBA PCA (Python, sklearn) SKL IncrementalPCA (Python, sklearn) Halko's method TruncatedSVD (Python, sklearn) Cell Ranger [22] IRLBA irlb (Python, from scratch) Seurat2 [49] IRLBA irlba (R, irlba)…”

Single cell immune profiling in transplantation research

“…In cases where data do not meet this assumption, manifold alignment attempts to find a transformation that aligns distributions across batches. 30 Normalization and transformation are also integral components of a preprocessing pipeline. Downstream analyses may make assumptions about the distribution and scale of the data, which necessitates transformation.…”