Question
Furthermore, the subsampled data does not only give us one valid solution to our question. If both cell types B and C decreased by 1,000 cells in the diseased case, we would obtain the same representative samples of 600 cells as above.
This above from chapter 18 (compositional analysis) doesn't seem to be true. To obtain similar sampling you would need all cell types to change proportionally right?
And I have a question about the main idea of the chapter, on the concern of the negative correlation between the number of cells under each cell type.
In DE analysis, normalized gene counts are also compositional — they sum to a constant (under whole library size normalization scheme), which induces negative correlations between genes. Yet this isn't treated as a fundamental problem; tools like edgeR and DESeq2 handle it routinely. Why is compositional bias considered a more serious issue in cell type abundance analysis? If you think of cell types as features (genes in DE are also features), they are fundamentally the same type of analysis. Is the concern qualitatively different, or is it primarily a matter of scale — fewer cell types meaning each one occupies a larger fraction of the total, combined with fewer biological replicates?
Question
This above from chapter 18 (compositional analysis) doesn't seem to be true. To obtain similar sampling you would need all cell types to change proportionally right?
And I have a question about the main idea of the chapter, on the concern of the negative correlation between the number of cells under each cell type.
In DE analysis, normalized gene counts are also compositional — they sum to a constant (under whole library size normalization scheme), which induces negative correlations between genes. Yet this isn't treated as a fundamental problem; tools like edgeR and DESeq2 handle it routinely. Why is compositional bias considered a more serious issue in cell type abundance analysis? If you think of cell types as features (genes in DE are also features), they are fundamentally the same type of analysis. Is the concern qualitatively different, or is it primarily a matter of scale — fewer cell types meaning each one occupies a larger fraction of the total, combined with fewer biological replicates?