Researchers analyze the expression of genes in individual cells to understand cellular functions and behaviors. However, one challenge researchers face with scRNA-seq data is the presence of many “zero” values—indicating either that a gene is not being expressed in a cell or that the data was not properly detected due to technical issues, like a “dropout.” Dropouts occur when the sequencing method misses detecting gene expression, but they are not an accurate reflection of whether the gene is truly unexpressed.

Existing methods for handling these zeros struggle to distinguish between these two possibilities—true zeros (unexpressed genes) and false zeros (dropout errors). This can lead to problems when trying to interpret gene expression data, as false zeros may be mistakenly treated as true zeros.

Researchers at Tsinghua University have developed a tool called scRecover that uses a statistical approach to accurately distinguish between dropout zeros and true zeros. The method is based on a zero-inflated negative binomial framework that models gene expression in each individual cell. This framework allows scRecover to estimate the probability that a zero value is due to a dropout event rather than a lack of expression.

Workflow of scRecover

After loading the expression matrix, scRecover first estimates dropout probabilities of genes and dropout zero number in each cell. Then it could identify true zeros and dropout zeros. scRecover docks with 3 intermediate imputation methods SAVER, scImpute, MAGIC respectively and alters dropout zeros with imputed values, which generates 3 imputed matrixes as the output.

In simple terms, scRecover improves upon previous imputation methods by focusing on fixing only the dropouts (false zeros), without mistakenly changing the true zeros (genuine lack of expression). It combines the strength of other imputation methods, such as scImpute, SAVER, and MAGIC, to enhance the accuracy of data recovery.

The effectiveness of scRecover was tested using simulations and real-world datasets. Results from these experiments show that scRecover significantly improves the accuracy of identifying and imputing dropout zeros compared to other methods. Importantly, it avoids the common problem of over-imputing true zeros, which can distort the analysis.

By providing more accurate data, scRecover enhances downstream analyses and helps scientists generate more reliable visualizations and interpretations of gene expression across individual cells. This breakthrough promises to make single-cell RNA sequencing even more powerful for understanding the complexities of biology, especially when it comes to analyzing rare or lowly expressed genes.

Availability – The R package scRecover is freely available at Bioconductor https://bioconductor.org/packages/scRecover and GitHub https://github.com/XuegongLab/scRecover