Table 2. Performance of Predicting Thermostability with Regression Tasks Using the HotProtein Dataset^a.

	HP-S2C2		HP-S2C5		HP-S
method	Spearman	Pearson	Spearman	Pearson	Spearman	Pearson
3D GCN50	0.490 ± 0.019	0.469 ± 0.019	0.291 ± 0.053	0.301 ± 0.074	N/A	N/A
TAPE51	0.432 ± 0.061	0.386 ± 0.065	0.367 ± 0.063	0.364 ± 0.047	0.504 ± 0.013	0.453 ± 0.031
ESM-IF135	0.589 ± 0.040	0.547 ± 0.036	0.373 ± 0.036	0.377 ± 0.035	N/A	N/A
ESM-1B29	0.890 ± 0.018	0.893 ± 0.024	0.712 ± 0.043	0.804 ± 0.023	0.807 ± 0.001	0.809 ± 0.001
HotProtein15	0.906 ± 0.010	0.923 ± 0.012	0.754 ± 0.035	0.837 ± 0.019	0.823 ± 0.001	0.827 ± 0.003
ProCeSa-ESM-1B	0.930 ± 0.012	0.958 ± 0.013	0.856 ± 0.053	0.900 ± 0.039	0.852 ± 0	0.858 ± 0
ProCeSa-ESM-1B (+Cont)	0.933 ± 0.014	0.961 ± 0.015	0.861 ± 0.050	0.904 ± 0.045	0.855 ± 0.001	0.861 ± 0.001
ProCeSa-ESM-C	0.942 ± 0.001	0.969 ± 0.000	0.919 ± 0.000	0.949 ± 0.002	0.892 ± 0.000	0.897 ± 0.001
ProCeSa-ESM-C (+Cont)	0.943 ± 0.001	0.969 ± 0.000	0.921 ± 0.002	0.954 ± 0.005	0.894 ± 0.001	0.900 ± 0.001

^aCorrelation coefficients of Spearman and Pearson are reported for all three datasets; 95(%) confidence intervals are computed via the 10-fold evaluation on HP-S²C2/C5 and 3 replicates on HP-S.