Correction: DeepRNAac4C: a hybrid deep learning framework for RNA N4-acetylcytidine site prediction

Guohua Huang¹Runjuan Xiao²Chunying Peng¹Jinyun Jiang¹Weihong Chen^1*

• ¹湖南财政经济学院, Changsha, China
• ²Shaoyang University, Shaoyang, China

Correction on: Huang, G.; Xiao, R.; Peng, C; Jiang, J; Chen, W. DeepRNAac4C: a hybrid deep learning framework for RNA N4-acetylcytidine site prediction. Front. Genet., 2025, *16* (), 1-13. https://www.frontiersin.org/articles/10.3389/fgene.2025.1622899 Text correction To better reflect the broader application of the methods, the phrase has been revised from "applied to predict ac4C sites efficiently" to the more comprehensive "widely applied to the field of molecular biology". A correction has been made to the section 1 Introduction, Paragraph 3, Line 3: "widely applied to the field of molecular biology" The original version of this article has been updated. Text correction To correct an erroneous citation. The reference "Zhao et al., 2019" has been replaced with the correct citation "Su et al., 2023". A correction has been made to the section 2 Materials and methods, 2.1 Datasets, Paragraph 1, Line 2: "In this study, we employed the same dataset as iRNA-ac4C (Su et al., 2023)" The original version of this article has been updated. Text correction For terminological precision and clarity, the sentence "To minimize sequence redundancy, we applied CD-HIT (Li and Godzik, 2006) with a similarity threshold of 0.8, filtering out highly similar sequences" is replaced by "The CD-HIT (Li and Godzik, 2006) with a sequence identity threshold of 0.8 was used to filter out highly similar sequences". A correction has been made to the section 2 Materials and methods, 2.1 Datasets, Paragraph 2, Lines 1-3: "The CD-HIT (Li and Godzik, 2006) with a sequence identity threshold of 0.8 was used to filter out highly similar sequences" The original version of this article has been updated. Equations Equation 6 in [2 Materials and methods, 2.2 Methodology, 2.2.4 BiLSTM, Paragraph 9] was erroneously given as ℎ□□= □□□□ tanh (□□□□ ). The correct equation is ℎ□□= □□□□⋅ tanh (□□□□ ). Text correction Terminology has been standardized for consistency. "Onehot" is now correctly spelled as "One-hot". Additionally, the typo "StratTableegies" has been corrected to "strategies". A correction has been made to the section 3 Results, 3.1 Performance comparison with various encoding methods, Paragraph 2, Lines 2-4: "It is evident that the One-hot+SVM method delivers superior overall performance compared to other encoding strategies" Text correction To remove the redundant and illogical word "both". A correction has been made to the section 3 Results, 3.2 Model selection, 3.2.1 Performance comparison with different combinations, Paragraph 3, Line 9: "By stacking BiLSTM and BiGRU, the model benefits from a combination of bidirectional processing," Text correction Incorrect citation in the footnote of Table 5. The citation "Zhao et al., 2019" has been corrected to "Lai and Gao, 2023". A correction has been made to the section 3 Results, 3.4 Performance comparison with state-of-the-art methods, Paragraph 2, Table 5 in Page 9: "Models marked with an asterisk (*) refer to previously published results [see (Lai and Gao, 2023)]." Text correction To adopt the author-date citation style and provide the complete reference. The numerical citation "[54]" has been replaced with "(McInnes et al., 2018)", and the corresponding bibliographic entry has been added to the References section. One correction has been made to the section 3 Results, 3.5 Visualization with UMAP, Paragraph 1, Line 3: "the Uniform Manifold Approximation and Projection (UMAP) (McInnes et al., 2018)" The other correction has been made to the section References: "McInnes L., Healy J., Melville J. (2018). Umap: Uniform manifold approximation and projection for dimension reduction. arXiv:180203426." Text correction Substantive revision to the rationale for robustness validation. The original text "Validating the robustness of DeepRNAac4C is crucial to ensure the model's generalization capability, which refers to its performance across different datasets and real-world applications. In practical research, data diversity is common, and robustness validation confirms that the model maintains its effectiveness despite such diversity. Additionally, verifying the robustness of the model aids in reproducing research results, ensuring the reproducibility and reliability of scientific findings" has been corrected. A correction has been made to the section 3 Results, 3.6 Robustness analysis of DeepRNAac4C, Paragraph 1, Lines 1-9: "The robustness of DeepRNAac4C is crucial to ensure the method's stability. In real-world application, training data may contain noise. The method sensitive to noise could not be applied in practice. Therefore, a test for robustness is essential." Text correction To accurately report the experimental results, we have corrected the text to match the data in Figure 7: 1) The range of accuracy fluctuations has been changed to 0.79–0.84; 2) The lower bound for the overall accuracy has been corrected to above 0.79. A correction has been made to the section 3 Results, 3.6 Robustness analysis of DeepRNAac4C, Paragraph 3, Lines 7 and 17: "the model's prediction accuracy fluctuates between 0.79 and 0.84" "the overall performance slightly decreases, but the overall accuracy remains above 0.79" Text correction To adhere to the standard spelling of the term, the title of Subsection 3.7 is corrected to "Web server". A correction has been made to the section 3 Results, 3.7 Web server: "3.7 Web server"