Table 3.

Comparisons of different methods on the eSOL test dataset

Models	RMSE	${\text{R}}^2$	Accuracy	Precision	Recall	F1	AUC
K-nearest neighbor	0.284	0.214	0.691	0.737	0.486	0.586	0.776
Linear regression	0.280	0.240	0.707	0.685	0.642	0.663	0.777
Random forest	0.255	0.370	0.760	0.750	0.690	0.729	0.825
Protein-Sol	0.253	0.376	0.714	0.689	0.688	0.693	0.808
XGboost	0.252	0.385	0.756	0.748	0.690	0.718	0.829
Support vector machine	0.246	0.411	0.761	0.763	0.684	0.721	0.842
DeepSol	0.241	0.434	0.763	0.771	0.738	0.695	0.845
ProGAN	0.237	0.442	0.763	0.770	0.676	0.720	0.853
SeqVec	0.236	0.458	0.767	0.754	0.715	0.734	0.858
TAPE	0.235	0.461	0.764	0.774	0.710	0.730	0.856
LSTM (All node features)	0.236	0.458	0.765	0.748	0.677	0.730	0.855
GraphSol (No contact)	0.235	0.462	0.763	0.710	0.676	0.729	0.853
GraphSol	0.231	0.483	0.779	0.775	0.693	0.732	0.866
GraphSol (Ensemble)	0.227	0.501	0.782	0.790	0.702	0.743	0.873

Italic values indicate the performance of our purposed model

Bold italic values indicate the performance of our ensemble model by using all folds of models to make a final prediction