Rdkit: Mol rendering within DataFrames in a Jupyter Notebook is broken with Pandas 0.25.1

[mrcblt]

Description:

• Python Version: 3.7.4
• RDKit Version: 2019.03.4
• Pandas Version: 0.25.1
• Platform: macOS/Linux

Starting with Pandas version 0.25.1, Mol objects within DataFrames are no longer drawn in a Jupyter notebook. The HTML string is displayed instead of the drawn molecule (<img src="data:image/png;base64,iVBORw0KGgoAAA...). With version 0.25.0 and below everything works fine.

Code to reproduce in a Jupyter Notebook:

from rdkit.Chem import PandasTools
df = PandasTools.LoadSDF('example.sdf')
df.head()

Current workaround aside from the switch to Pandas 0.25.0:

from rdkit.Chem import PandasTools
from IPython.core.display import HTML

def show(df):
    return HTML(df.to_html(notebook=True))

df = PandasTools.LoadSDF('example.sdf')
show(df.head())

[greglandrum]

Confirmed. This is due to some change in pandas v0.25.x
Now we just need to figure out how to work around it.

[hellerm2]

Hi Greg,

the current workaround from above needs, at least in my hands, the following line:

PandasTools.RenderImagesInAllDataFrames(images=True)

Without that line, only a string gets displayed.

[fredrikw]

I looked at the Pandas code and they no longer call to_html from _repr_html_ but instead go directly to a "DataFrameFormatter". So, the patched to_html will not be called for repr. The best solution might be to patch the DataFrameFormatter.to_html() instead.

[fredrikw]

Or on second thought, that would probably be too harsh since it will patch every dataframe, not only the ones with structures.
So maybe patching _repr_html_ in addition to to_html?

[greglandrum]

Some additional info here, as I collect it:

• Things work fine with pandas 0.25.0, the breakage first appears in 0.25.1
• This snippet works with pandas 0.25.3 from IPython.display import HTML;HTML(df.head().to_html())
• This was the PR that did it: https://github.com/pandas-dev/pandas/pull/27991/files