Figure 6
The comparison of persistent Dirac and persistent weighted Dirac models with conventional methods from MoleculeNet74 on the FreeSolv database. Both types of persistent Dirac models performed better than existing conventional methods such as MoleculeNet’s XGBoost (1.74 ± 0.15 kcal/mol), Multitask (1.87 ± 0.07 kcal/mol), Random Forest (RF) (2.03 ± 0.22 kcal/mol) and Kernel Ridge Regression (KRR) (2.11 ± 0.07 kcal/mol) models. Note that a lower RMSE value indicates better result.
