Despite substantial progress in single-cell RNA-seq (scRNA-seq) data analysis methods, there is still little agreement on how to best normalize such data. Starting from the basic requirements that inferred expression states should correct for both biological and measurement sampling noise and that changes in expression should be measured in terms of fold changes, University of Basel researchers derive a Bayesian normalization procedure called Sanity (SAmpling-Noise-corrected Inference of Transcription activitY) from first principles. Sanity estimates expression values and associated error bars directly from raw unique molecular identifier (UMI) counts without any tunable parameters. Using simulated and real scRNA-seq datasets, the researchers show that Sanity outperforms other normalization methods on downstream tasks, such as finding nearest-neighbor cells and clustering cells into subtypes. Moreover, they show that by systematically overestimating the expression variability of genes with low expression and by introducing spurious correlations through mapping the data to a lower-dimensional representation, other methods yield severely distorted pictures of the data.
Summary of the Sanity approach
a, Cartoon of the flow of causality from the physical state of the cell to gene expression patterns. The concentrations of transcription factors (TFs), chromatin modifiers and other regulatory factors determine changes in chromatin state, three-dimensional (3D) organization of the chromosomes, binding and unbinding of TFs to promoters and enhancers, and so on. These determine the time-dependent rate λg(t) at which gene g was described a time t in the past. Similarly, the concentrations of microRNAs, RNases and other RNA-binding proteins determine the time-dependent rate μg(t) at which mRNAs of gene g decayed at time t in the past. b, The transcription activity ag of gene g is defined as the expected number of mRNAs and is a weighted average of its transcription and decay rates in the past. We define the expression state of the cell as the vector of relative transcription activities of all genes. c, Logical flow from expression state to observed UMI counts . The expression state and total transcription activity Ac determine the transcription activities agc. For each gene g, the probability P(mgc∣agc) of having mgc mRNAs is a Poisson distribution with mean agc. Assuming each mRNA in cell c has a probability pc of being captured and sequenced, the probability P(ngc∣pc, agc) of obtaining ngc UMIs is a Poisson distribution with mean pcagc. d, The probability of obtaining the UMI counts given the cell state is a product over genes of Poisson distributions with means Ncαgc, where Nc is the total UMI count in cell c.
Availability – Sanity was implemented in C and is freely available for download at https://github.com/jmbreda/Sanity.