The Human Cell Atlas is the world’s largest single-cell reference atlas, and it’s still growing. It includes references to millions of cells from various tissues, organs, and stages of development. These resources assist clinicians in better understanding the effects of age, environment, and disease on cells, allowing them to better identify and treat patients.
However, single-cell datasets are complicated due to the batch effect and often contain measurement mistakes. Furthermore, learning from reference atlas is challenging due to the limited availability of computational resources and sharing restrictions on raw data.
Researchers from Helmholtz Zentrum München and the Technical University of Munich (TUM) have recently proposed a novel algorithm Single-Cell Architecture Surgery, named “scArches,” in short.
Previously, raw data was shared between clinics or research facilities using different techniques. Instead, this approach compares new datasets from single-cell genomics with existing references using transfer learning, ensuring privacy and anonymity. This makes annotating and understanding new data sets simple, and it substantially democratizes the use of single-cell reference atlases.
The team used their proposed scArches algorithm to study COVID-19 in several lung bronchial samples. Using single-cell transcriptomics, they compared COVID-19 patients’ cells to healthy references. For both mild and severe COVID-19 cases, the algorithm was able to distinguish diseased cells from controls, allowing the user to pinpoint the cells in need of therapy. The quality of the mapping technique was unaffected by biological differences between patients.
The team’s goal is to make cell references as simple to use as genome references. To put it differently, if we want to bake a cake, we usually don’t try to come up with our recipe; instead, we consult a cookbook. Similarly, the researchers plan to formalise and simplify this lookup process with scArches for cell references.