Predicting RNA sequence-structure likelihood via structure-aware deep learning

BMC Bioinformatics

Table 2 Models performance on TeDS

Metric	NU-ResNet		NUMO-ResNet		ENTRNA
	\(\varvec{\vartheta }^{*}_{a}\)	\(\varvec{\vartheta }^{*}_{\ell }\)	\(\varvec{\vartheta }^{*}_{a}\)	\(\varvec{\vartheta }^{*}_{\ell }\)
Accuracy	93.75%	92.19%	90.63%	96.88%	73.44%
AUCROC	0.9736	0.9824	0.9824	0.9912	0.7275
MCC	0.875	0.8442	0.8141	0.9375	0.5130
Precision	93.75%	93.55%	93.33%	96.88%	66.67%
Recall	93.75%	90.63%	87.5%	96.88%	93.75%
Specificity	93.75%	93.75%	93.75%	96.88%	53.13%

Models with parameters that optimize the validation accuracy are referred to as \(\varvec{\vartheta }^{*}_{a}\), while models with parameters that optimize the validation loss are referred to as \(\varvec{\vartheta }^{*}_{\ell }\)

ISSN: 1471-2105