![]() |
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ЦЭМИ РАН |
||
The stages of clustering and typing are the most difficult in terms of interpreting the results. Their main task is to divide sample cells into clusters containing cells similar in expression pattern. Existing methods for analyzing scRNA-seq data make it possible to cluster and type cells of nervous tissue and blood samples well. Cell typing of connective tissue samples is a difficult task, since the cells in the sample can be in different states and do not have specific markers that can be used to successfully identify them. A good result is the additional use of machine learning methods and neural networks. In addition to using the Cell Ranger pipeline and the Seurat1 R package, which have become classic, it is necessary to use machine learning and neural network methods to analyze connective tissue cells. So, for clustering, an R-package based on scDeepCluster2 neural networks is used. ScCapsNet3 (CapsNet) is used for cell typing. When using scDeepCluster to analyze a sample of MSCs in adipogenic differentiation, 10 clusters were obtained instead of 8 when using Seurat. The clusters obtained using these two methods contain cells whose gene expression lists are very different, which indicates fundamentally different clustering algorithms. Using scCapsNet, small differences in biochemical processes between clusters were revealed, which makes it possible to more accurately characterize and identify the analyzed cells. The use of standard scRNA-seq data analysis algorithms does not allow obtaining the maximum amount of information. This requires additional use of machine learning methods and neural networks.