iR1mA-LSTM: Identifying N1-Methyladenosine Sites in Human Transcriptomes using Attention-based Bi-LSTM
T. T. T. Do, T.-H. Nguyen-Vo, Q. H. Trinh, P.-U. Nguyen-Hoang, L. Nguyen, and B. P. Nguyen∗
Methylation is the most frequently occurring epigenetic modification that accounts for over 50% of the total modification forms. Among the methylation sites in the adenosine nucleobase, N1-methyladenosine (1mA) is a significant post-transcriptional alteration found in myriad types of RNA molecules. The alteration of gene sequences caused by this methylation affects numerous biological processes, such as genome editing, cellular differentiation, and gene expression, causing dangerous diseases. Many experimental techniques were developed to determine 1mA in RNA sequences. These approaches, however, are not cost- and time-effective solutions for wide screening in laboratories with limited budgets. To partially address this problem, several computational methods have been introduced to assist experimental scientists in identifying 1mA sites. In this paper, we present a more effective computational model called iR1mA-LSTM to predict 1mA sites in human transcriptomes using the Long Short-term Memory networks enhanced by an attention mechanism to improve the predictive power.
We conducted repeated experiments to fairly estimate the robustness and stability of the model performance and benchmarked our model with other methods on the same independent test set. The results show that iR1mA-LSTM outperformed other methods with both the area under the receiver operating characteristic curve and the area under the precision-recall curve values of over 0.99.
Source code and data are available upon request.
- Source 1: Click here
- Source 2: Click here
Trang T. T. Do, Thanh-Hoang Nguyen-Vo, Quang H. Trinh, Phuong-Uyen Nguyen-Hoang, Loc Nguyen, Binh P. Nguyen. iR1mA-LSTM: Identifying N1-Methyladenosine Sites in Human Transcriptomes Using Attention-Based Bidirectional Long Short-Term Memory. In: N.H. Phuong, V. Kreinovich (eds) Deep Learning and Other Soft Computing Techniques. Studies in Computational Intelligence, vol 1097. Springer, Cham.. DOI: 10.1007/978-3-031-29447-1_5