Dataframe sum columns according to dictionnary of list of columns

I am trying to find the most pytonic way to do the following :

I have a dataframe and a dictionary

import pandas as pd
df = pd.DataFrame([[4,8,52,7,54],[0,20,2,21,35],[2,33,12,1,87]], columns = ['A', 'B', 'C', 'D', 'E'])
dic = {'x':['A','D'], 'y' : ['E'], 'z':['B','C']}

I am trying to have a dataframe with the keys of the dictionary as columns and each columns would be the sum of the columns from the original dataframe in the list of the corresponding key. For example, column ‘x’ would be the row-wise sum of the column ‘A’ and ‘D’.

This is easily doable with some loops but I am wandering if there is any elegant solution similar to list comprehension for example.

>Solution :

Invert your dictionary, rename the columns and groupby.sum:

d2 = {v:k for k,l in dic.items() for v in l}
out = df.rename(columns=d2).groupby(level=0, axis=1).sum()

Or, for recent pandas versions (in which groupby with axis=1 will be deprecated):

d2 = {v:k for k,l in dic.items() for v in l}
out = df.T.rename(d2).groupby(level=0).sum().T

Output:

    x   y   z
0  11  54  60
1  21  35  22
2   3  87  45