Single-cell RNA sequencing (scRNA-seq) has significantly advanced transcriptomics by enabling the analysis of gene expression at the resolution of individual cells. This technology is crucial for understanding cellular heterogeneity within tissues and identifying cell-type-specific gene expression patterns. A major application of scRNA-seq is the identification of differentially expressed genes (DEGs) between different conditions or groups of individuals. Additionally, scRNA-seq facilitates the study of expression quantitative trait loci (eQTLs), which are genetic variants influencing gene expression levels. While scRNA-seq offers powerful insights, the design of efficient experiments requires careful consideration of several factors. The statistical power of an scRNA-seq experiment, or its ability to detect true biological signals, depends on various parameters. These parameters include the number of samples (individuals), the number of cells sequenced per individual, the sequencing depth (reads per cell), the effect size (magnitude of the biological signal), and the variability of the data. Currently, methods for power analysis in multi-sample scRNA-seq studies are predominantly based on computationally expensive simulations, limiting the ability to explore a wide range of experimental designs and optimize for cost-effectiveness. This study addresses this limitation by proposing a novel analytical framework, named scPower, for efficient power analysis and experimental design in multi-sample scRNA-seq experiments, focusing specifically on inter-individual comparisons for DEGs and eQTLs. This framework offers a significantly faster and more scalable solution compared to simulation-based approaches, enabling researchers to effectively evaluate numerous experimental designs and optimize their experiments within budgetary constraints.

Existing literature highlights the challenges in designing efficient scRNA-seq experiments. Several studies have addressed power analysis for scRNA-seq, but many focus on single-sample comparisons or lack scalability for multi-sample studies. Simulation-based approaches are common, but they are time-consuming and computationally intensive, restricting the number of experimental designs that can be practically evaluated. Previous work on optimizing scRNA-seq experiments has primarily focused on maximizing the number of detectable genes or rare cell types. There is a need for a comprehensive framework that considers both the cost and statistical power of detecting DEGs and eQTLs in multi-sample scRNA-seq experiments. Studies have compared different scRNA-seq platforms, demonstrating variability in cost, sequencing depth, and overall performance. However, a unifying framework that integrates these factors into experimental design and power analysis remains lacking. This research builds upon existing literature by developing a novel analytical model that considers the interplay between various experimental factors and provides a flexible tool for researchers to optimize their study design.

The scPower framework models the relationship between various experimental parameters and the power to detect DEGs and eQTLs within specific cell types. The model decomposes the overall detection power into two components: (1) the expression probability, representing the likelihood of detecting a gene's expression in a given cell type; and (2) the DE/eQTL power, signifying the probability of identifying a gene as differentially expressed or associated with an eQTL given its expression. The expression probability is estimated using a negative binomial distribution, accounting for the sparsity and variability of scRNA-seq data. Cell type-specific expression priors are derived from pilot datasets, capturing the typical expression levels of genes in various cell populations. The model incorporates hyperparameters based on sequencing depth to account for variations in gene expression across different cell types. The DE/eQTL power is assessed using analytical power analysis methods, leveraging established tools for bulk RNA-seq data adapted to the pseudobulk approach. This approach aggregates gene expression counts across cells of the same type to perform DE/eQTL analyses efficiently, thereby circumventing the computational burden of analyzing individual cells directly. The model accommodates different statistical tests, multiple testing corrections (Bonferroni or FDR), and various effect size measures. The framework also includes cost modeling for different scRNA-seq platforms (e.g., 10X Genomics, Drop-seq, Smart-seq2) to enable budget-constrained optimization. The cost is modeled as a combination of library preparation and sequencing costs, depending on the number of samples, cells per sample, and read depth. Doublet rates are incorporated into the cost model, considering the trade-off between increasing cell numbers and introducing more doublets. The model is implemented as an R package and web application, providing a user-friendly interface for exploring different experimental designs and selecting optimal parameters.

scPower accurately predicts the number of detectable genes per cell type, considering sequencing depth and cell numbers. The model demonstrates excellent agreement with simulation-based methods (powsimR and muscat) for DE power calculations, confirming its accuracy and efficiency. For eQTL analysis, scPower shows good concordance with a custom simulation-based approach, providing a valuable tool for power analysis when simulation-based methods are computationally prohibitive. The analysis indicates that shallow sequencing of a large number of cells is more cost-effective than deep sequencing of fewer cells for maximizing overall detection power. The optimal experimental parameters vary depending on the budget, effect size, and expression level of the genes being studied. Optimal parameter combinations are determined for various scenarios using both simulated and observed priors from different tissues and scRNA-seq platforms (10X Genomics, Drop-seq, Smart-seq2). The study finds that the number of cells per individual is the most crucial factor influencing power in most scenarios, followed by sequencing depth and sample size. The optimal read depth is found to be around 10,000 reads per cell in many analyses. Doublet rates are modeled, and overloading lanes is shown to be beneficial despite a potential increase in doublets, due to the significant increase in detection power.

The scPower framework offers a significant advancement in experimental design for multi-sample scRNA-seq studies. The analytical approach significantly improves the efficiency and scalability compared to simulation-based methods, enabling a comprehensive exploration of various experimental parameters and optimization within budgetary constraints. The validation against simulation-based methods confirms the model's accuracy and reliability. The findings highlight the importance of balancing the number of cells sequenced, sequencing depth, and sample size to maximize power. The integration of cost modeling into the framework is particularly valuable for researchers facing budgetary limitations. The model's applicability across different tissues and scRNA-seq platforms demonstrates its versatility and potential to guide experimental design in diverse research settings. The study also emphasizes the crucial role of well-matched pilot data or established priors in accurately estimating power. The pseudobulk approach used in scPower leverages well-established statistical methods, simplifying the analytical process and making the framework readily accessible to researchers.

scPower provides a powerful and efficient tool for designing and analyzing multi-sample scRNA-seq experiments, focusing on inter-individual comparisons for DEGs and eQTLs. Its analytical approach offers significant advantages over simulation-based methods in terms of speed and scalability. The framework's versatility, cost modeling, and validation against simulation-based methods makes it a valuable resource for researchers seeking to optimize their experiments and maximize the biological discoveries. Future work could focus on extending the framework to accommodate more complex experimental designs, such as those involving time-course studies or multiple factors, and further refine the cost modeling for various scRNA-seq technologies. The increasing availability of large-scale single-cell atlases will also enhance the accuracy and applicability of the scPower framework.

The scPower framework relies on the use of prior information, which may not always be readily available for all studies. The accuracy of power estimation depends heavily on the quality and relevance of the chosen priors. The model currently focuses on inter-individual comparisons for DEGs and eQTLs; extending it to other types of analyses (e.g., co-expression analysis, variance QTLs) would broaden its applicability. The cost modeling is specific to certain scRNA-seq platforms, and may need modifications to accommodate other technologies. The model's assumptions about doublet detection and removal may also affect the accuracy of power estimations in experiments with high doublet rates.