[I have no experience in the kind of systematic site-of-metabolism annotation you are doing, but] there is another well-known example that puts your question into focus. Bioactivation of vinyl chloride in the liver gives chlorooxirane, a highly potent carcinogen, so clearly the C=C bond is the site of metabolism rather than any one atom.

On the other hand, perhaps the convention is to indicate the initial point of metabolic reaction and refer only to those atoms in the parent structure, not groups that react later? N-demethylation like you see with amitriptyline and erythromycin, begins with N-oxidation, followed by breaking of C-N bonds as a multi-step follow-on process (oxygen and hydrogen transfers, hydroxide elimination and iminium hydrolysis). If the point of SoMs numbering is to help group metabolites by common reactions, then focusing on the N is enough to associate tertiary N-oxide, >NCH2OH, [>N=CH2]+ and >NH metabolites. Also, the process for oxidizing and removing either methyl group is identical, so there is no need to indicate both of these (i.e. the one that gets removed from amitriptyline is automatically atom 20, while C21 is not a SoM).

Dear Dr Kilgore,

Thank you for your response. Indeed, the definition of site of metabolism can be a bit cumbersome. I am looking at sites of metabolism in regard to in silico prediction. The idea that one should look at the atom where the reaction starts how I have generally looked at SoMs. However, it seems scientists have not always looked at it that way. In the case of vinyl chloride, I would label the two carbons as SoMs.

By defining bonds as SoMs, one would have to also consider implicit hydrogen atoms, as in the case of A-hydroxylation (A = C, or N), where the A-H bond will be modified to A-O-H.

To your opinion on:

"Also, the process for oxidizing and removing either methyl group is identical, so there is no need to indicate both of these (i.e. the one that gets removed from amitriptyline is automatically atom 20, while C21 is not a SoM)"

>> I think that for the sake of completion, and machine learning, it would be important to mention both C20 and C21 though, if one decides to add the carbon atoms near to the N atom as SoMs. Both C20 and C21 are technically SoMs. Because of the symmetry, we, humans could see that the removal of either C leads to the same product, just by visualizing. The computer, however, does not see it, and, since the atom neighbourhoods of both C20 and C21 are the same, it will have to predict both as SoMs.

Looking forward to hearing from you, and others.

Thanks,

Yannick

Yannick, I did quite a bit of work on predictive metabolism using Meteor, Symyx and others. Looking at bonds rather than atoms does make sense but there are other constraints and limitations to it. A couple of examples are:

1. A dimethylamine derivative like amitriptyline and a monomethyl derivative (like nortriptyline) can have different rates of demethylation. The C-N bonds do not remain equivalent in mono- and di-methyl forms. Not only implicit hydrogens, but electron cloud density also plays a role.

2. Stereochemical constraints can affect enzyme accessibility of metabolism sites.

In my experience, in several cases, the software used to predict scores of metabolites but the actual number of metabolites detected in plasma, hepatocytes or microsomes used to ......ZERO. Unmetabolized drug, excreted unchanged in urine!

As regards your approach of representing with bonds, you can find papers where people have tried to correlate with atom/group electronegativity and reactivity. These approaches can be adopted in your model.