Update subsection: Structure-based prediction of bioactivity

[0ut0fcontrol]

Hi,
@mapleleaf-soar and I found AI models (e.g. AtomNet #56, ACNN #287) achieve high performance on DUD-E and PDBbind data sets because they learned the bias in the data sets. REF #1008

For example:

We found ACNN model can achieve "state-of-the-art" in PDBbind data set using ligand alone.😂

Others also found the bias in DUD-E. REF #1009 #1010

I believe the subsection Structure-based prediction of bioactivity need to be updated and try to work on it.

• The impact of deep learning in treating disease and developing new treatments
  • Drug development
    • Structure-based prediction of bioactivity

Any comments and suggestions are welcome!

[agitter]

Thanks for the suggestions. Would you like to write a short update to this section? Discussing bias in data sets and evaluations is definitely of interest to us.

I could likely review that pull request, though I can't promise a specific timeline. If you want to make a pull request, I recommend

• Skimming add discussion in interpretability section and updates to molecular design section and discussion sections (issues with previous PR fixed) #988 to make sure your edits won't introduce conflicts. If so, let's try to merge that one first.
• Being explicit about which other models discussed in the review are affected by this dataset bias.

[cgreene]

I agree with @agitter : a short update to that section as a pull request would be great - thanks - with the same caveats :)