I have 4 groups of cells that were separated via FACS into No expression (negative) Low expression (Low), Medium expression (medium), and Hi expression, should the design matrix look like this or have another row for the negative, or 0 for the Low, medium, and Hi in the base line when using MAGeCK?
cat > design_matrix.txt <<'EOM'
Samples baseline Low Medium Hi
Negative_rep1 1 0 0 0
Negative_rep2 1 0 0 0
Negative_rep3 1 0 0 0
Low_rep1 1 1 0 0
Low_rep2 1 1 0 0
Low_rep3 1 1 0 0
Medium_rep1 1 0 1 0
Medium_rep2 1 0 1 0
Medium_rep3 1 0 1 0
Hi_rep1 1 0 0 1
Hi_rep2 1 0 0 1
MPH_Hi_rep3 1 0 0 1
•
written
by
Matthew •
1 answer
Your matrix is right as it stands. Don't add a column for Negative -- it's already the reference, absorbed by the baseline column. If you gave it its own column the design would be rank deficient (baseline would equal Negative + Low + Medium + Hi) and MAGeCK would either error out or hand you nonsense. The beta scores for Low/Medium/Hi are each relative to Negative, which is what you're after.
One thing worth checking: your last row is MPH_Hi_rep3 while the others are Hi_repN. If that doesn't match the column header in your count table exactly, MAGeCK will drop that sample or fail on it. Might be deliberate, but it looks odd next to the others.
Separately, since your four groups are an expression gradient rather than unrelated conditions, you could collapse the three columns into one ordinal column (0/1/2/3 across Negative to Hi) if what you actually care about is a monotonic dose-response. That gives you more power and one beta per gene instead of three. Comes down to whether the question is "does knocking this out shift cells along the gradient" or "which specific bins differ from negative".
Log in to answer this question.
-
Downregulated Genes with GFOLD
written by jth
