sort dataframe index by the second position from nested dictionaries

I have this code where lig_dec_residue is this nested dictionary which comes from big .dat files:

 lig_dec_residue = {'f1': {}, 'f2': {}, 'f3': {}}


def plot_lig(res):
    df = pd.DataFrame.from_dict(lig_dec_residue)
    df.index = df.index.str.split(' ')
    df.index = df.index.str[0] + ' ' + (df.index.str[1].astype(int) + int(res) - 1).astype(str)
    df = df[df <= -0.25]
    df.dropna(how='all', inplace=True)
    df.plot(kind='bar', edgecolor='black')
    plt.legend(['X var', 'Y var', 'Z var'])
    plt.show()
    plt.close()

and this is the result:

                   f1        f2        f3
ARG 403 -0.265999       NaN -0.390653
LEU 455 -1.948253 -2.125521 -1.988445
PHE 456 -1.974429 -1.835651 -2.177540
ALA 475 -0.796856 -1.032929 -0.968554
GLY 476 -0.262736 -0.744952 -0.257448
ASN 477       NaN       NaN -0.868419
PHE 486 -3.674621 -2.882512 -3.179725
ASN 487 -1.172256 -0.805725 -1.050299
LYS 493 -2.283489       NaN -5.231593
SER 496       NaN       NaN -0.366986
PHE 497       NaN -0.340862       NaN
ARG 498 -1.485091       NaN -1.140743
THR 500 -1.497597 -0.778616 -1.961580
TYR 501 -4.286950       NaN -4.851700
GLY 502 -0.447453 -0.808606 -0.702321
VAL 503 -0.256496 -0.371461 -0.977062
HIS 505 -1.420959       NaN -1.321259
LYS 417       NaN -1.115154       NaN
GLN 493       NaN -2.625195       NaN
GLY 496       NaN -1.232041       NaN
GLN 498       NaN -2.271338       NaN
ASN 501       NaN -4.152646       NaN
TYR 505       NaN -2.469813       NaN

Pandas plots the last six entries apart from the rest (look at TYR 501, ASN 501: they should be close but they are not!).

The idea is to make a comparison between f1 f2 and f3 with a bar plot.
This is my output:

Is there a way to sort the index properly? I think this output might be due to the lexicographic sorting method.
I know that there’s natsort library, but I can’t make use of if since the dataframes comes from nested dictionaries.
I would like to group the bars based on the number of the index (eg, HIS 505 next to TYR 505) for a direct comparison where applicable.

Thank you!

Ludovico

>Solution :

Use sort_index with a custom key:

df = df.sort_index(key=lambda x: x.str.split().str[1].str.zfill(5))
print(df)

# Output
               f1        f2        f3
ARG 403 -0.265999       NaN -0.390653
LYS 417       NaN -1.115154       NaN
LEU 455 -1.948253 -2.125521 -1.988445
PHE 456 -1.974429 -1.835651 -2.177540
ALA 475 -0.796856 -1.032929 -0.968554
GLY 476 -0.262736 -0.744952 -0.257448
ASN 477       NaN       NaN -0.868419
PHE 486 -3.674621 -2.882512 -3.179725
ASN 487 -1.172256 -0.805725 -1.050299
LYS 493 -2.283489       NaN -5.231593
GLN 493       NaN -2.625195       NaN
SER 496       NaN       NaN -0.366986
GLY 496       NaN -1.232041       NaN
PHE 497       NaN -0.340862       NaN
GLN 498       NaN -2.271338       NaN
ARG 498 -1.485091       NaN -1.140743
THR 500 -1.497597 -0.778616 -1.961580
TYR 501 -4.286950       NaN -4.851700
ASN 501       NaN -4.152646       NaN
GLY 502 -0.447453 -0.808606 -0.702321
VAL 503 -0.256496 -0.371461 -0.977062
HIS 505 -1.420959       NaN -1.321259
TYR 505       NaN -2.469813       NaN

Detail about the key:

>>> df.index.str.split().str[1].str.zfill(5)
Index(['00403', '00455', '00456', '00475', '00476', '00477', '00486', '00487',
       '00493', '00496', '00497', '00498', '00500', '00501', '00502', '00503',
       '00505', '00417', '00493', '00496', '00498', '00501', '00505'],
      dtype='object')

Note: padding with 0 allow you to have a natural sorting when two numbers have not the same length:

>>> '23' > '5'
False

>>> '23' > '05'
True