Prediction of polyreactive and nonspecific single-chain fragment variables through structural biochemical features and protein language-based descriptors

Table 1 Performance of models with different ML algorithms and single descriptors

Descriptor	Method	Train AUC	Valid AUC	Test AUC	Accuracy	Precision	Recall	F1-score
F46	GBM	0.917 ± 0.006	0.705 ± 0.129	0.805	0.730	0.753	0.607	0.672
	LGBM	0.858 ± 0.009	0.704 ± 0.124	0.811	0.731	0.700	0.720	0.710
	RF	1.000 ± 0.000	0.697 ± 0.123	0.795	0.728	0.722	0.658	0.689
	XGB	0.951 ± 0.003	0.701 ± 0.125	0.810	0.732	0.705	0.710	0.708
UniRep	GBM	1.000 ± 0.000	0.591 ± 0.177	0.821	0.747	0.741	0.686	0.712
	LGBM	0.926 ± 0.006	0.606 ± 0.178	0.816	0.734	0.704	0.722	0.713
	RF	1.000 ± 0.000	0.575 ± 0.178	0.815	0.736	0.728	0.671	0.699
	XGB	0.999 ± 0.000	0.596 ± 0.181	0.824	0.740	0.718	0.709	0.713
TAPE	GBM	1.000 ± 0.000	0.647 ± 0.160	0.824	0.745	0.741	0.679	0.709
	LGBM	0.919 ± 0.005	0.657 ± 0.155	0.810	0.731	0.703	0.713	0.708
	RF	1.000 ± 0.000	0.638 ± 0.160	0.815	0.745	0.747	0.666	0.704
	XGB	0.998 ± 0.000	0.651 ± 0.156	0.822	0.746	0.729	0.706	0.717
ESM-1b	GBM	0.979 ± 0.003	0.603 ± 0.182	0.807	0.727	0.742	0.616	0.673
	LGBM	0.922 ± 0.006	0.608 ± 0.176	0.807	0.730	0.697	0.722	0.710
	RF	1.000 ± 0.000	0.593 ± 0.176	0.814	0.736	0.730	0.669	0.698
	XGB	0.998 ± 0.000	0.604 ± 0.177	0.821	0.741	0.718	0.713	0.716
ESM-1v	GBM	1.000 ± 0.000	0.594 ± 0.150	0.819	0.743	0.761	0.639	0.695
	LGBM	0.919 ± 0.005	0.602 ± 0.172	0.813	0.730	0.694	0.729	0.711
	RF	1.000 ± 0.000	0.582 ± 0.171	0.816	0.740	0.735	0.674	0.703
	XGB	0.949 ± 0.002	0.597 ± 0.175	0.808	0.728	0.702	0.701	0.701

ISSN: 1471-2105