Comparison of prediction accuracies of 363 partition coefficients (log P) for COSMO-RS and GNN for two datasets with their description.

	Description of the log P dataset	Prediction methods	Kendall τ	RMSE
Set A	300 data points (30 depolymerized lignin derivatives, 10 organic solvents)	GNN (Student 35)	0.87	0.51
		COSMO-RS	0.77	0.50
Set B	63 data points (17 drug-like compounds, 4 organic solvents)	GNN (Student 35)	0.70	1.15
		COSMO-RS	0.77	1.00