The discrepancy between the KI- and 10x Genomics-derived cell types could be a consequence of a lower sequencing depth in the 10x Genomics dataset (~18,500 mapped reads per cell) than in the KI dataset (~1.2 million mapped reads per cell). Notably, the minimum sequencing depth is generally considered to be between 25,000 and 50,000 mapped reads per cell [34]. This suggests that the relatively low sequencing depth of the 10x Genomics dataset led to overlapping cell clusters. Additionally, although k-means clustering is commonly used for single-cell data, selecting the right value of k is challenging [34]. PCA-based clustering methods would be particularly well-suited for low sequencing depth [35], and for instance, could be expanded to initially select significant principal components with PCA and use these for subsequent clustering [36].
