Fig. 1

scEvoNet scheme. A scEvoNet takes a list of clusters and a matrix of expression for each sample as input. For each sample, it generates an object with cell type classifiers and top important features for each cluster from the provided set of clusters. In the final step, the tool builds a confusion matrix and a network of genes associated with each cell type. B We use the LGBM algorithm to produce a classifier for each cell type. To smooth the data in order to deal with the batch effects we apply the sigmoid function and use only the top important features to create the final cell type classifiers