LogP vs logD

Here are a few references on problems calculating logD

• 1. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5053177/ - paper from David Mobley about just calculated logP (RMSE of 1.5 log units)
• 2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3496770/ - paper demonstrating that logP calculations are reasonable, but pKa and logD are not good (figure 6 tells a story...)
• 3. https://blogs.sciencemag.org/pipeline/archives/2012/08/22/watch_that_little_letter_c - blog from Derek Lowe discussing these matters.

Josh writes:

I don't have a reference for the dataset being inappropriate for the methods out there, that requires some more technical reading. Generally, the software is trained on drug-like molecules, usually referencing fragment tables that simply won't exist for these excipient molecules (some way too greasy, and some way too polar, like the dyes). Further reinforcing this point is reference 2, where they show that commercial tools are better (relatively) for neutral molecules than charged ones (dyes and everal others are charged ions in our case).