Evaluation of model performance on genes with intermediate abundance ratios. The gene represented here had been excluded from the training and evaluation subsets of their respective studies. (A) The RNAlight repository includes a file of predictions for all canonical lncRNA transcripts. The accuracy of binary labels was tabulated across those transcripts having a mean CNRCI value in lncATLAS. (B) Code and data were obtained from the lncLocator 2.0 repository. The code was modified to apply middle exclusion to the training subset but the inverse filter to the testing subset for the default 100 epochs. The experiment was run on the H1.hESC cell line which that study used as a baseline. (C) The TACOS web server was given 10 sequence files, one for each of its 10 supported cell lines. Each file represented canonical transcripts of randomly selected genes having an intermediate CNRCI value for that cell line. (D) It was not feasible to perform a similar test on DeepLncRNA. The data developed for the study, which did not use lncATLAS, were provided in the online supplement without the genes filtered by middle exclusion
| . | Model, authors . | Model performance on excluded genes . |
|---|---|---|
| A | RNAlight Yuan et al. (16) | 55% accuracy was computed from a file of predictions on all genes, available in the source code repository. |
| B | lncLocator 2.0 Lin et al. (14) | 59% accuracy was obtained on the ‘baseline’ cell line by training the model with middle exclusion and evaluating on middle genes from the test set. |
| C | TACOS Jeon et al. (15) | 56% accuracy was obtained after submitting 100 randomly selected sequences for each of the 10 supported cell lines, 1000 total. |
| D | DeepLncRNA Gudenas & Wang (13) | N/A. The filtered portion of the data was not included in the supplemental data files. |
| . | Model, authors . | Model performance on excluded genes . |
|---|---|---|
| A | RNAlight Yuan et al. (16) | 55% accuracy was computed from a file of predictions on all genes, available in the source code repository. |
| B | lncLocator 2.0 Lin et al. (14) | 59% accuracy was obtained on the ‘baseline’ cell line by training the model with middle exclusion and evaluating on middle genes from the test set. |
| C | TACOS Jeon et al. (15) | 56% accuracy was obtained after submitting 100 randomly selected sequences for each of the 10 supported cell lines, 1000 total. |
| D | DeepLncRNA Gudenas & Wang (13) | N/A. The filtered portion of the data was not included in the supplemental data files. |
Evaluation of model performance on genes with intermediate abundance ratios. The gene represented here had been excluded from the training and evaluation subsets of their respective studies. (A) The RNAlight repository includes a file of predictions for all canonical lncRNA transcripts. The accuracy of binary labels was tabulated across those transcripts having a mean CNRCI value in lncATLAS. (B) Code and data were obtained from the lncLocator 2.0 repository. The code was modified to apply middle exclusion to the training subset but the inverse filter to the testing subset for the default 100 epochs. The experiment was run on the H1.hESC cell line which that study used as a baseline. (C) The TACOS web server was given 10 sequence files, one for each of its 10 supported cell lines. Each file represented canonical transcripts of randomly selected genes having an intermediate CNRCI value for that cell line. (D) It was not feasible to perform a similar test on DeepLncRNA. The data developed for the study, which did not use lncATLAS, were provided in the online supplement without the genes filtered by middle exclusion
| . | Model, authors . | Model performance on excluded genes . |
|---|---|---|
| A | RNAlight Yuan et al. (16) | 55% accuracy was computed from a file of predictions on all genes, available in the source code repository. |
| B | lncLocator 2.0 Lin et al. (14) | 59% accuracy was obtained on the ‘baseline’ cell line by training the model with middle exclusion and evaluating on middle genes from the test set. |
| C | TACOS Jeon et al. (15) | 56% accuracy was obtained after submitting 100 randomly selected sequences for each of the 10 supported cell lines, 1000 total. |
| D | DeepLncRNA Gudenas & Wang (13) | N/A. The filtered portion of the data was not included in the supplemental data files. |
| . | Model, authors . | Model performance on excluded genes . |
|---|---|---|
| A | RNAlight Yuan et al. (16) | 55% accuracy was computed from a file of predictions on all genes, available in the source code repository. |
| B | lncLocator 2.0 Lin et al. (14) | 59% accuracy was obtained on the ‘baseline’ cell line by training the model with middle exclusion and evaluating on middle genes from the test set. |
| C | TACOS Jeon et al. (15) | 56% accuracy was obtained after submitting 100 randomly selected sequences for each of the 10 supported cell lines, 1000 total. |
| D | DeepLncRNA Gudenas & Wang (13) | N/A. The filtered portion of the data was not included in the supplemental data files. |
This PDF is available to Subscribers Only
View Article Abstract & Purchase OptionsFor full access to this pdf, sign in to an existing account, or purchase an annual subscription.
