My group is working on method development, and I have an experiment where the samples are 0/100, 20/80, 40/60 60/40,80/20,100/0 mixes of two different tissues. I wanted to use DESeq and treat this like a dose response experiment, and then confirm that the two experiments return the same trend. But I realized that the size normalization assumptions are being violated, and when I look at the size normalization factors compared to the percentages, it makes a very nice slope, which is bad.

Any suggestions on how to normalize this data? Should I calculate size factors just based on simple library size?

Some groups use the controlGenes argument of estimateSizeFactors, using genes known a priori to be relatively stable expressed across conditions (here tissues).