I work on a non-model system, a small Caribbean lizard Anolis, and I am trying to address the hypothesis that distinct patches of skin (with distinct pigmentation) show different expression patterns. We have a rough genome and transcriptome that I am using in my analyses. I have three populations that I study: one with patches of skin that are completely yellow, a second with patches that are completely orange, and a third with patches with yellow margins and an orange center. I reasoned that to reduce the amount of false positives I could address my question in two ways. First, I would compare the expression between pure orange and yellow populations, and the second would be to compare yellow and orange portions of animals that show both pigments. I would then compare the results from both experimental designs, and highlight those found across both as the best candidates.

My question is the following, when running 'DESeqDataSetFromTximport' to compare the expression between yellow margins and orange center of a given specimen, am I correct to use the following design: ddsyointra <- DESeqDataSetFromTximport(txiyointra, colData = designyointra, design = ~ specimen + condition)​

Where the design is as follows: sample condition specimen 10839o.10840.carolinensis.clean orange 10839 10839y.10840.carolinensis.clean yellow 10839 10840o.10840.carolinensis.clean orange 10840 10840y.10840.carolinensis.clean yellow 10840 10844o.10840.carolinensis.clean orange 10844 10844y.10840.carolinensis.clean yellow 10844

And then use the following design for comparing skin patches from orange and yellow populations: ddsyointer <- DESeqDataSetFromTximport(txiyointer, colData = designyointer, design = ~ condition)

Is that valid? I am not entirely sure if I am correct in accounting for specimen ID in this way.

Thank you for your time.

The choice of design is up to you, I’d recommend discussing with someone familiar with linear models on the difference between adding a species effect or other choices in setting up the design. Usually if there is a term that one thinks may affect the counts it is included. But its an analysis decision not a software question.