Testing spatial clustering by genetic cline
Comparisons between spatial and non-spatial models using conStruct reveal that the spatial model is always favored (Figure 5), as the accuracies of the spatial model are significantly higher than that of non-spatial model in K=1 to K=8. In addition, the analysis of layer contributions for K=1 to K=10 shows that additional layers contribute very little at K=5 and become negligible at K=7 (FigureS4 ). Viewed in combination with the results from sNMF, K=4 to K=6 are preferred models.