LTPConstraint: a transfer learning based end-to-end method for RNA secondary structure prediction

BMC Bioinformatics

Table 9 Table of comparative results to examine transfer learning

Length	Family	Transfer_learning				Non_transfer_learning
Length	Family	Precision	Recall	F1-score	ave-F1	Precision	Recall	F1-score	ave-F1
Encoding_128	5SrRNA	0.9857	0.9804	0.9831	0.8489	0.9777	0.8128	0.8876	0.6862
	tRNA	0.9985	0.9992	0.9988		0.8968	0.7873	0.8385
	PDB	0.6695	0.3050	0.4190		0.2985	0.1600	0.2083
	SPR	0.9929	0.9971	0.9950		0.8833	0.7489	0.8106
Encoding_512	grpl	0.8304	0.8894	0.8589	0.7051	0.3002	0.0843	0.1317	0.1235
	RNP	0.5334	0.7000	0.6054		0.2432	0.1647	0.1964
	SRP	0.7130	0.7378	0.7252		0.2236	0.3636	0.2769
	telomerase	0.3752	0.8728	0.5248		0.0033	0.0072	0.0045
	tmRNA	0.7550	0.8767	0.8113		0.0041	0.9996	0.0081

In the group of transfer learning, two kinds of pre-trained model are still trained by using the data of Rfam_128 and Rfam_512. After the results of each group are available, two average F1 value is calculated for the families with two kinds of coding length

ISSN: 1471-2105