I ran CNVkit on bulk RNA samples using the import-rna flag. I generated the TCGA CNA/expression correlation files as directed and they look very similar to the example. I used HTSeq to generate my count matrices. I used only samples that passed QC and simply calculated the mean expression for each gene. I ran CNVkit on both raw and normalized counts to see if that made a difference, but unfortunately although the program runs perfectly fine, my output shows the same weight of 1.0 for all genes and the exact same predicted copy number (0.044394...) for all genes as well.
I'm thinking there might be lots of reasons for why my runs didn't pan out: maybe my data is too sparse after all (RNAseq on formalin-fixed tissue with baits), maybe I can try to use RSEM for quantification. I was just wondering whether anyone has specific advice for what I can try next.