Fig. 5

Biological processes enriched by the signature genes in shaping the cell interaction networks. A, B For the two high-dimensional datasets of HDST, the biological processes involving the top-ranked genes were obtained through Gene Ontology Enrichment Analysis. C, D For two low-throughput datasets (seqFISH and MERFISH), due to the limitation of gene numbers, all the genes were sorted by their sensitivity scores, and gene set enrichment analysis was performed. E Prediction of cell interactions with genes annotated as ligands or receptors and the same number of top-ranked signature genes identified by DeepLinc in the seqFISH (left) and MERFISH (right) datasets. Simple machine learning models, including support vector classification (SVC), K-nearest neighbor (KNN), and gradient boosting decision tree (GBDT) models, were used for this test. The originally defined cell adjacency map was used as a positive set of cell interactions, and randomly selected nonadjacent cell pairs with the same numbers were used as a negative set. This process was repeated 30 times with different training and testing set assignments. P-value were obtained with Student’s t test