At first we knew that the UMAP plot for germ cells (scRNA-seq) always appeared continuous.
I have a single cell transcriptome dataset of fish testis (nearly 95% germ cells) , and I should set the PCs for UMAP to only 3 to ensure the plot appeared normally continuous. When the PCs come to 4 or higher, the topological structure will be twisted and cracked, so it can not demonstrate the developing stages of germ cells at all.
Why did you do this experiment? How many detectable cell types do you think exist in your testis sample? Why do you think continuity in a UMAP is a requirement of developing cell stages? Are you asking why is your data showing you more than you want to see? Rather than impose dimensionality on your data, as Mensur Dlakic suggests, determine the dimensionality from the data itself. That is, if you're using the data for discovery. On the other hand, if you've exhausted the data analysis and now want to tell a story involving a subset of PCs, that would be different. Why did you do this experiment?
Don't see why you think that 3 PCs are most suitable. All those plots look fine to me. I don't know much about the scientific problem you are studying, but in terms of visualization I don't think there is any rule that requires UMAP plots to be continuous. UMAP plots in general have nothing to do with biology - they are just low-dimensional representations of high-dimensional data. We try to extrapolate them to match what biology is telling us, which is probably why you think that 3 PCs are more suitable. Someone else may argue that 4 PCs are better because cluster 12 is more clearly separated from the rest.
Depending on the variance captured by 3 PCs versus a larger number, it is quite possible that 4 or 10 PCs are more appropriate because they include more variance. I suggest you try whatever number of PCs is required to capture at least 90-95% variance. Depending on the number of data dimensions, UMAP may be able to create this plot in a reasonable time from raw data.
Lastly, you may want to know that it is frowned upon when the same post is created at multiple websites (for reasons that are not always clear to me).
Why did you do this experiment? How many detectable cell types do you think exist in your testis sample? Why do you think continuity in a UMAP is a requirement of developing cell stages? Are you asking why is your data showing you more than you want to see? Rather than impose dimensionality on your data, as Mensur Dlakic suggests, determine the dimensionality from the data itself. That is, if you're using the data for discovery. On the other hand, if you've exhausted the data analysis and now want to tell a story involving a subset of PCs, that would be different. Why did you do this experiment?
You can upload pictures using the
imageicon (one that is to right of101010button).For pseudotime analysis part, you can use monocle.