Skip to main content

Advertisement

Table 2 Mean absolute errors for total energies (\(U_0\) and E in kcal/mol), HOMO and LUMO energies (in eV) using different ML methods on different training and prediction datasets

From: Dataset’s chemical diversity limits the generalizability of machine learning predictions

MethodTrainTest\(U_0\)EHOMOLUMO
EN (CM)QM9QM921.122.00.340.64
KRR (CM)QM9QM94.95.20.180.25
SchNetQM9QM90.31.00.040.03
EN (CM)PC9PC938.20.470.66
KRR (CM)PC9PC922.80.310.36
SchNetPC9PC91.60.060.05
SchNetQM9PC9(A)3.00.070.06
SchNetQM9PC9(B)8.90.330.27
SchNetPC9QM9(A)3.40.050.05
SchNetPC9QM9(B)4.20.120.11
  1. (A) corresponds to the subset of molecules that belongs to QM9 and PC9, whereas (B) indicates molecules exclusive to the dataset