The angle terms applied to atoms in or around a ring have not received a lot of attention. Concerns have been raised that they may be overly stiff or broad. For example, as briefly mentioned by Bill Swope and Alberto Gobbi: 2022-10-13 Force Field Release Meeting notes
Lorenzo D’Amore compared OpenFF 2.0 performance to GAFF 2.11 with a specific focus on ring moieties: https://openforcefieldgroup.slack.com/archives/CKSHCE7SB/p1670282381642029 (powerpoint also attached here for posterity)
As summarised by David Mobley :
One key thing from Lorenzo’s analysis seems to be that performance of the same angle parameters is different across the several categories of angles he notes:
1. All atoms are in a ring
2. Two atoms are in a ring
3. One atom is in a ring
4. No atoms are in a ring
This probably suggests we should try some fitting experiments — at least for angle parameters with larger errors — where we separate out these categories (either individually or in groups) and fit separately.
In general, Lorenzo concludes that Sage performs better than GAFF when all atoms are in a ring (case 1), but underperforms in all other cases (2, 3, 4).