Single-cell RNA sequencing, or scRNA-seq, is a cutting-edge technique that lets scientists study gene activity in individual cells. This technology helps researchers uncover how different cell types function and interact in health and disease. One crucial step in analyzing scRNA-seq data is identifying which cell types are present in the sample. This process is called cell-type identification.

To make cell-type identification more accurate and efficient, scientists often use supervised methods. These methods rely on reference data, essentially, examples of cells whose types are already known, to help label new data. Think of it as having a labeled guidebook that helps you recognize and categorize unknown cells based on their gene activity patterns.

However, a big challenge arises because the quality of cell-type identification depends heavily on how well the reference data matches the new target dataset. Differences in the types of cells present, or variations in how the data was collected, can cause errors. For example, if the reference data doesn’t include certain rare cell types found in the target data, or if the overall makeup of cell types differs significantly, the identification process may struggle.

To solve this, a team led by researchers at Brown University have developed a new strategy called Target-Oriented Reference Construction, or TORC. TORC creates a customized reference dataset that is specifically designed to match the characteristics of the target data. This means it carefully selects and adjusts reference cells to better reflect the distribution and composition of the new dataset.

By doing so, TORC reduces the differences between reference and target data, allowing supervised methods to work more accurately. The team tested TORC extensively on both simulated datasets, which are computer-generated to mimic real data, and actual biological samples. In all cases, TORC consistently improved the accuracy of cell-type identification.

Overview of TORC framework

The main goal of TORC is to construct a reference that closely resembles the target. TORC employs a two-round prediction strategy. In the first round, a classifier is trained on a reference with known cell labels to predict the cell-type composition of the target. Using the predicted probability matrix from this initial prediction, entropies are calculated for each target cell. Cells with relatively low entropies are then added to expand the reference pool. A target-oriented reference is subsequently created by sampling from the expanded reference pool according to the estimated cell-type composition. The constructed reference is then used to retrain the classifier and update the cell labels of the target

This advancement is important because better cell-type identification helps researchers draw more reliable conclusions from scRNA-seq studies. It can enhance our understanding of complex biological systems, disease mechanisms, and potentially lead to improved diagnostics and therapies.

Availability– https://github.com/weix21/TORC