Single-cell RNA sequencing (scRNA-seq) has become an essential tool for understanding the diversity of cells in development, disease, and tissue organization. It lets researchers examine gene expression in individual cells rather than looking at average signals from bulk tissue. However, a major challenge in scRNA-seq is the occurrence of “dropouts”, instances where genes appear to be inactive not because they truly are, but because the sequencing method missed them. These missing values can make the data harder to interpret and can skew downstream analyses like clustering or trajectory inference.

To solve this problem, researchers have developed various imputation methods that attempt to fill in the missing gene expression values. But many of these tools risk introducing new errors or altering values that were already accurate. Researchers at Xiamen University have introduced a new two-step approach called scTsI that offers a more reliable way to handle missing data in single-cell RNA-seq datasets.

Flow chart of scTsI

In the first stage, scTsI uses neighboring cells and genes for initial imputation through KNN. In the second stage, scTsI leverages bulk RNA-seq data as a constraint and uses ridge regression for adjusting the initial imputed values.

In the first step, scTsI fills in missing values by borrowing information from similar cells and genes. Then, it further adjusts the data by converting the matrix into a vector, applying a transformation, and refining the imputed values through ridge regression. Importantly, this step uses bulk RNA sequencing data to guide the corrections, which helps maintain realistic biological patterns and avoids overwriting truly high gene expression values.

scTsI is also computationally efficient because it can handle sparse matrices directly, which is useful for large datasets. The team tested scTsI on a wide variety of real and simulated data and showed that it performs well even when dropout rates are high. It improves gene recovery, preserves relationships between cells, and enhances the results of downstream analyses like data visualization and clustering.

Availability – scTsI is available at: https://github.com/JGuan-lab/scTsI.