I have been reading some literature on molecular energy prediction using machine learning techniques and I have noticed that one of the input parameters in many of the available models is "chirality". In these works "chirality" is defined as an atom feature. Meaning that for every atom in the molecule, a one-hot or null vector is used to assign R/S to the atom. As far as I understand, chirality is a molecular property and only the atom which is the chiral center can be assigned R/S.

I was wondering if my understanding is correct and if not, what am I missing. Thanks!

PS: One such a paper is this.

  • 3
    $\begingroup$ I don't think you are missing anything. An atom is assigned as R/S or nonchiral (null vector). While atoms can be chiral centers, chirality is inherently a molecular property, with the whole molecule not being superimposable on its mirror image. $\endgroup$
    – Tyberius
    Sep 12, 2019 at 0:38
  • $\begingroup$ Thanks @Tyberius. I just found an exact description in an implementation of such algorithm, stating "chirality`: (categorical) R, S, or not a Chiral center (one-hot encoded)." which makes it clear that it only assigns R/S to chiral center. $\endgroup$
    – Blade
    Sep 12, 2019 at 1:22
  • 3
    $\begingroup$ Just to clarify: a molecule may be achiral and have an even number of stereogenic ("chiral") centers greater than 0 (meso compound). A chiral molecule may also have more than one stereogenic center. $\endgroup$
    – TAR86
    Sep 12, 2019 at 5:49


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.