Analyzing single-cell RNA sequencing (scRNA-seq) data is crucial for understanding complex biological processes and disease development, but identifying individual cell types within these vast datasets has been a significant bottleneck. An international research group led by The University of Osaka has developed a new computational tool, scODIN (Optimized Detection and Inference of Names in scRNA-seq data), to automate and streamline this complex process, paving the way for faster discoveries with high-accuracy classification in biomedicine.

scODIN tackles the challenge of cell type identification by employing a tiered system that allows users to define specific cell subsets at different levels of detail. The tool first identifies major cell clusters automatically, like T cells, B cells, and monocytes. Then, researchers can specify more granular subsets, like regulatory T cells or helper T cells, within those broader categories. This flexible approach accommodates varying levels of resolution depending on the research question. Importantly, scODIN recognizes cells with intermediate phenotypes or transitional states, assigning “double labels” which capture the complexity often missed by traditional methods. Furthermore, it employs a k-nearest neighbor algorithm to recover cell identities affected by dropout events, a common issue in scRNA-seq data where some genes fail to be detected.

Optimized Detection and Inference of Names in scRNA-seq data (scODIN) overview

scODIN consists of three main steps and additional detail phenotyping. Above is a schematic of each step, in the middle, the UMAPs of identified cell types for each step, and below is an example of time used for the indicated cell numbers. a, Step one is a fast and accurate automatic cell type identification at the top level, to identify clusters of e.g. CD4 T cells (CD4-T), B cells, and monocytes (Mono) for further subset phenotyping. At the top level, the cell type identities are made on a cluster level. The remaining steps in scODIN is truly single-cell determination of cell identities. b, A user-defined tier-based system is implemented to identify cells of different detail levels. In the example, specific subsets such as Tregs and T helper cells are identified in the first tier and more generic cell types such as central memory (CM) and effector memory (EM) are identified in the second tier. c, In the same calculation as b) it is possible to add accepted double labels, such as mixed T helper phenotypes or intermediate phenotyping, to identify cells that have an in-between gene expression of naïve and effector cells. d, Finally, the last step uses k-nearest neighbor (kNN) machine learning to expand cell identification in case crucial core genes have undergone single-cell dropout events. A user-defined cutoff of stringency is used to either give every single cell a category or leave some unidentified. e, Cell numbers and running times for each step of scODIN. 

The automation and streamlined approach offered by scODIN promises to accelerate biomedical research. By alleviating the laborious manual annotation process, researchers can focus on interpreting the biological meaning of their data and pursue new therapeutic avenues more efficiently. This improved efficiency can lead to faster discoveries, particularly in areas like immunology and cancer research, where understanding cellular heterogeneity is crucial for developing personalized therapies.

“scODIN empowers researchers to easily navigate and analyze complex scRNA-seq datasets,” says Dr. James Wing, co-senior author of the study published in The Journal of Immunology. “Its automated and flexible approach not only saves time but also reveals intricate details about cellular populations, opening new doors to understanding disease mechanisms and developing effective treatments.”

Source – The University of Osaka

Availability – The scODIN R package and bioinformatics data analysis pipeline have been deposited to GitHub (https://github.com/jonasns/scODIN).