Percepta: Why stereochemistry is generally ignored in property predictions

Description
Percepta typically does not treat stereoisomers differently when predicting many physicochemical and ADME/Tox endpoints. For most covered properties, the available data and statistical constraints do not support robust, stereo‑specific models.

This applies both to classic physchem properties (logP, pKa, etc.) and many ADME/Tox endpoints, where the experimental data often do not distinguish reliably between stereoisomers.

Solution

For physicochemical endpoints (logP, pKa, etc.):

• Experimental evidence usually shows little systematic effect of stereochemistry on these properties within experimental error.
• Ignoring stereochemistry is therefore acceptable for most use cases.

For ADME/Tox endpoints:

While biology is often stereo‑selective, practical modeling is limited by:

Data limitations:

• Insufficient stereo‑specific datasets for many properties.
• Often classification data only (e.g., 0/1 for toxicity) where isomers end up in the same class.

Statistical limitations:

• No consistent mapping between a specific configuration and effect across structural classes.
• Stereo‑specific descriptors would become very complex, leading to too many variables and low statistical significance.

Cases where stereochemistry can matter strongly:

• A small number of properties associated with very specific ligand–protein interactions or specialized toxins.
• Percepta currently focuses on general‑purpose models rather than a small subset of highly stereo‑sensitive systems.

Customer guidance:

• Use Percepta predictions as approximate for stereoisomer pairs unless there is clear evidence that your endpoint is strongly stereo‑specific and that experimental data confirm large differences.
• For critical stereo‑selective systems, rely on experimental studies supported by Percepta predictions, rather than the predictions alone.