new conformer benchmark data | @David Hahn | LPW- origin of non-symmetrical distribution of ddE? DH - rematched MM minima to ‘best match' QM conformer CIB - y axis to log may give more information about the distribution (tail of the distribution)? HJ - couldnt see the clear information form log distribution
CIB - how can we do fair comparison OPLS3e with other general force fields? Different approaches (OPLS3e: bespoke forcefield. Other ffs: general ff, aiming to fast efficient ff including very wide chemistry) DH -off ff an be improved by adding bespoke parameters
LPW - in 1d ddE, heavier tails on the left for all ffs. if we draw marginal distribution of the 2d, are we getting the same plot with 1d?(DH- should be the same) in openff-1.2.0, concentrated area is more toward positive side. (DH - but can see more outliers on the negative side. ) CIB- scatter plot. is the functional form of OPLS3e similar with openff functional form? DM - offsite charge might be the major difference HJ - outliers around the positive ddE (DH - only shown in unconstrained 1.2.0). HJ will check what are the molecules assigned to the outliers DM - big band around -10 in both OPLS3e and openff-1.2.0 → need to study origin of the trend
|
short discssion about “goodness-of-fit metrics” | | DM - goodness-of-fit metrics (how to assess/benchmark): for geometry based metric, energetic based metric, single number that can show the goodness-of-fit (for example, average of rmsd or else) LPW - performance could be dependent on the chemical space that the validation set covers. |