GSE60450 PCA: three gene filters, PCA by numpy SVD, no extra packages
CPM uses library size = column sum of the original full matrix (all genes).
Values: log2(CPM+1); genes centered across the 12 samples; no variance scaling.
SVD on samples x genes centered matrix; variance = S^2/(n-1), n=12.
PC1/PC2 signs of B and C aligned to A by positive dot product; no rotation applied.
Distance check: Euclidean distance in top-2 PC space, all 66 sample pairs, Pearson r vs A.
No significance testing performed.
