PCA percentage of variance explained is wrong

[lambdamoses]

In scRNA-seq and Visium data, the first 50 PCs usually only explain a small proportion of total variance, but I saw this code used to make the scree plot:

Giotto/R/dimension_reduction.R

Lines 649 to 652 in d2d746a

	eigs = slot(pca_obj, 'misc')$eigenvalues
	# variance explained
	var_expl = eigs/sum(eigs)*100

This is incorrect. Variance explained by each PC normally means the eigenvalue divided by the total variance, or the sum of all eigenvalues, but here the sum is of the eigenvalues computed instead of all eigenvalues. If I compute 50 eigenvalues, then the sum of the 50 eigenvalues would be used here, even if there should be thousands of eigenvalues and the sum of the first 50 is only about 25% of the total. Showing 100% cumulative variance explained as in the plot below while it should really be more like 25% is very misleading.

[RubD]

Hi @lambdamoses thanks for the feedback. We will update our documentation soon and provide additional clarification and suggestions to the users. I'm not sure if 50 PCs typically explain only a small proportion of total variance. That would be surprising but we will also empirically test this in the near future.