Coef Bug and EdgeR Design Inquiries

[Marwansha]

Hi,

thanks for developing such useful tool.

i have some questions regarding the edgeR pipline as am not very familiar with it and it will be very helpful if you can answer my 4 rather naive questions

i have this design CMV+AGE+Batch+sex and i have some specific Questions

Testing Covariates:

1)How do I specify which covariates to test in glmLRT(fit,)? Attempting glmLRT(fit, coef="AGE") results in an error., same error for any of the covarites put in coef="" , the error dont rise if i use the index number glmLRT(fit, coef=[2]), or not indicating any.

2. based on what we put the group_col in DGEList , ** for my design am not sure what should i put as group_col?**

in the description it says "group_col : str the name of the column containing the group information in :code:samples.
only used if :code:group is not :code:None"

3. does the processing steps include a similar function to calcNormFactors? or its not necessary for edgeR?

example code and error below

                design = dmatrix("~CMV+AGE+Batch+sex", data=anno,)
                dge_list = DGEList(counts=counts, samples=anno, group_col="sex", genes=mono.var)
                dge_list.estimateGLMCommonDisp(design=design)

                print(design)
                fit = dge_list.glmFit(design=design)
                lrt = glmLRT(fit)
               glmLRT(fit,coef="AGE")

 specifying a coef. it  give an error saying  
ValueError                                Traceback (most recent call last)
Cell In[286], line 1
----> 1 glmLRT(fit,coef="AGE")

File [~/venvs/CODA/lib/python3.9/site-packages/inmoose/edgepy/glmFit.py:454](https://rstudio.pasteur.fr/s/01e0021f2ed724eb5240a/lab/workspaces/auto-n/tree/appa/homes/~/venvs/CODA/lib/python3.9/site-packages/inmoose/edgepy/glmFit.py#line=453), in glmLRT(glmfit, coef, contrast)
    452 tab["pvalue"] = LRT_pvalue
    453 tab.index = glmfit.counts.index
--> 454 res = DGELRT(tab, glmfit)
    455 res.comparison = coef_name
    456 res.df_test = df_test

File [~/venvs/CODA/lib/python3.9/site-packages/inmoose/edgepy/DGELRT.py:41](https://rstudio.pasteur.fr/s/01e0021f2ed724eb5240a/lab/workspaces/auto-n/tree/appa/homes/~/venvs/CODA/lib/python3.9/site-packages/inmoose/edgepy/DGELRT.py#line=40), in DGELRT.__init__(self, df, glmfit, *args, **kwargs)
     40 def __init__(self, df, glmfit, *args, **kwargs):
---> 41     super().__init__(df, *args, **kwargs)
     42     self.fit = glmfit

File [~/venvs/CODA/lib/python3.9/site-packages/inmoose/diffexp/DEResults.py:45](https://rstudio.pasteur.fr/s/01e0021f2ed724eb5240a/lab/workspaces/auto-n/tree/appa/homes/~/venvs/CODA/lib/python3.9/site-packages/inmoose/diffexp/DEResults.py#line=44), in DEResults.__init__(self, df, *args, **kwargs)
     43 for col in ["log2FoldChange", "lfcSE", "pvalue"]:
     44     if col not in self.columns:
---> 45         raise ValueError(f"{col} missing from results table")

ValueError: log2FoldChange missing from results table

4 ) the tutorial only mentions estimating common dispersion. However, edgeR documentation recommends estimating tagwise dispersion as well. so what i did is estimate GLMCommonDisp followed by estimateGLMTagwiseDisp?
and then feed the tagwise dispersion into the fit function as below . is this correct?

fit = dge_list.glmFit(design=design, dispersion=dge_list.tagwise_dispersion)

fit.design
DesignMatrix with shape (287, 5)
  Intercept  CMV[T.Positive]  Batch[T.batch2]  sex[T.male]  AGE
          1                0                1            1   31
          1                0                0            1   37
          1                0                0            1   36
          1                1                1            1   35
          1                0                1            1   36
          1                1                0            1   37
          1                1                1            1   37
          1                0                1            0   31
          1                0                0            0   41
          1                0                0            0   48
          1                0                1            0   47
          1                1                0            0   47
          1                0                1            0   47
          1                1                1            0   43
          1                0                0            0   42
          1                0                1            0   41
          1                0                1            0   46
          1                0                1            0   40
          1                0                0            0   39
          1                0                0            0   44
          1                0                0            0   45
          1                1                0            0   41
          1                0                1            0   41
          1                0                0            0   39
          1                0                1            0   42
          1                0                1            0   41
          1                0                1            0   47
          1                1                0            0   49
          1                0                1            0   40
          1                0                1            0   40
  [257 rows omitted]
  Terms:
    'Intercept' (column 0)
    'CMV' (column 1)
    'Batch' (column 2)
    'sex' (column 3)
    'AGE' (column 4)
  (to view full data, use np.asarray(this_obj)
    'sex' (column 3)
    'AGE' (column 4)
  (to view full data, use np.asarray(this_obj))

thanks alot for your help

Marwan

[EpigeneMax]

Hi Marwan, thanks for using inmoose.

I'll try to address your questions:

1. With what version of inmoose do you see the error? I am quite confident it is fixed in the latest version, but I'd rather double check that.
2. group and group_col are a convenient way to define the design in simple cases. From your code examples, I see you use your own design matrix, that you pass to every method, so you do not need to worry about group or group_col.
3. in edgepy the normalization factors should be automatically computed, unless you pass your own.
4. correct 👍

Let me know if you have further questions.

[dawe]

hello @EpigeneMax

3. in edgepy the normalization factors should be automatically computed, unless you pass your own.

It would be useful to have TMM size factors explicitly computed (or at least shown in the .samples['norm_factors'] after computation).

@Marwansha in case you want to use RLE instead of TMM, I think it's safe to compute them with the deseq2 module:

dge_list = inmoose.edgepy.DGEList(counts=counts)
dge_list.samples['norm_factors'] = inmoose.deseq2.estimateSizeFactors.estimateSizeFactorsForMatrix(dge_list.counts.T)