This is a test version of Biostars. For the public version, visit https://www.biostars.org.
R Model Matrix has extra columns

I am trying to do differential expression analysis on some RNA microarray data. I was setting up my model matrix for limma from a csv file which has info on the samples, specifically how they were to be grouped (cre/flox status). Some example data is below:

geo_name,cre_lox,cell_type,treatment1,replicate_num
sample1,flox,c1,no,1
sample2,flox,c1,no,2
sample3,flox,c1,no,3
sample4,cre,c1,no,1
sample5,cre,c1,no,2
sample6,cre,c1,no,3
sample7,wt,c2,no,1
sample8,wt,c2,yes,1

The subset of the data I want (where cell_type=c1) has only "cre" and "flox" in the "cre_lox" column.
I selected for it using:

q1a_selected_col_data = col_data[(col_data$cell_type == 'c1'),]

However, when I used the function model.matrix(~q1a_selected_col_data$cre_lox) it results in a matrix like this:

(Intercept) q1a_selected_col_data$cre_loxflox q1a_selected_col_data$cre_loxwt
1         1                                 1                               0
2         1                                 1                               0
3         1                                 1                               0
4         1                                 0                               0
5         1                                 0                               0
6         1                                 0                               0

How did it "know" to add a column for "wt" status even though the data I passed to it does not have "wt" in it? Is there a way I can prevent things like this without having to modify the csv or remove columns from the model matrix by hand?

r rna-seq limma
• 3,471 views • link
written by tpham2654 • tpham2654

1 answer

your crelox column is stored as a factor. model.matrix will automatically put in a column for all non-reference levels of a factor variable, even if there isn't a sample with a given factor level.

To mitigate against problems like this, you could use droplevels on your original dataframe

• 1 views • link
written by russhh russhh

Thanks. I found that the design matrix is the inverse of what I want. Basically the 1 and 0 in the q1a_selected_col_data$cre_loxflox column should be switched since cre is the experimental group.

I tried model.matrix(~q1a_selected_col_data$cre_lox-1) after uing droplevels to invert it and now I get this:

 q1a_selected_col_data$cre_loxflox q1a_selected_col_data$cre_loxcre
1                                1                                0
2                                1                                0
3                                1                                0
4                                0                                1
5                                0                                1
6                                0                                1

I want the last column. Is there a way I can select the label model.matrix should mark as "1"?

• 1 views • link
written by tpham2654 • tpham2654

it doesn't really make any difference (although it might make things simpler to reason about) since you specify the experimental comparisons in your contrasts matrix, not your design matrix

• 1 views • link
written by russhh russhh

Log in to answer this question.

More posts like this