Summary of the lattice energy prediction scores for pentacene, 5A and 5B (respectively 564, 594 and 936 structures). Our best accuracies on these datasets are estimated from average scores from a 4-fold cross validation (75% of the dataset is used for training). Δ-learning refers to the learning of the difference between W99 and DFT energies.

Dataset	MAE [kJ mol−1]	RMSE [kJ mol−1]	R 2
Pentacene (W99)	0.29 ± 0.03	0.49 ± 0.08	0.979
Pentacene (DFT)	0.48 ± 0.04	0.68 ± 0.04	0.984
Pentacene (Δ)	0.51 ± 0.04	0.70 ± 0.06	0.96
5A (W99)	0.41 ± 0.02	0.59 ± 0.04	0.967
5A (DFT)	0.64 ± 0.03	0.91 ± 0.07	0.930
5A (Δ)	0.59 ± 0.03	0.85 ± 0.06	0.85
5B (W99)	0.98 ± 0.03	1.31 ± 0.03	0.877
5B (DFT)	1.09 ± 0.03	1.44 ± 0.04	0.870
5B (Δ)	0.74 ± 0.04	1.00 ± 0.05	0.83