Prediction of the water solubility by a graph convolutional-based neural network on a highly curated dataset

Table 4 Performance of the consensus GNN for different solubility datasets commonly used in literature

Dataset from literature	Number of data points	Overlap of chemicals with our training data	Consensus GNN		Consensus GNN based on a reduced training set
Dataset from literature	Number of data points	Overlap of chemicals with our training data	r²	rmse	r²	rmse
Delaney	1128	966	0.950	0.488	0.889	0.711
Huuskonen*	1291	1125	0.960	0.427	0.905	0.637

ISSN: 1758-2946