Table 3.

Regression and classification performance of simple MLP with varying similarity cutoffs

Similarity cutoff	Number of train sets	PCC	MSE	CI	Prec	Recall	BACC
1	313,414	0.867	0.691	0.854	0.858	0.908	0.862
0.6	289,107	0.848	0.780	0.843	0.886	0.843	0.855
0.59	259,290	0.824	0.872	0.829	0.872	0.841	0.845
0.58	226,521	0.777	1.097	0.806	0.847	0.832	0.824
0.57	183,555	0.719	1.380	0.780	0.839	0.778	0.797
0.56	148,252	0.653	1.722	0.750	0.828	0.704	0.762
0.55	113,519	0.578	2.027	0.715	0.786	0.655	0.718
0.54	81,362	0.534	2.163	0.704	0.789	0.622	0.709
0.53	61,755	0.495	2.572	0.687	0.795	0.497	0.670
0.52	46,652	0.425	2.971	0.660	0.788	0.372	0.624
0.51	36,447	0.376	3.096	0.639	0.760	0.366	0.612
0.5	29,831	0.328	3.462	0.624	0.750	0.291	0.586

The number of test data points is fixed at 80,578. As the similarity to the test set diminishes, indicated by lower cutoff values, the performance metrics (PCC: Pearson correlation coefficient, MSE: mean squared error, CI: concordance Index, Prec: precision, Recall, and BACC: balanced Accuracy) generally decline.