NBCR-ac4C: A Deep Learning Framework Based on Multivariate BERT for Human mRNA N4-Acetylcytidine Sites Prediction

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• N4-acetylcytidine (ac4C) is vital for gene expression and disease.
• Wet lab identification of ac4C is laborious and expensive.
• Existing computational methods lack semantic understanding of sequence data.

Purpose of the Study:

• To propose a novel deep learning model, NBCR-ac4C, for accurate ac4C site prediction.
• To leverage pretrained models for enhanced feature extraction from nucleotide sequences.
• To improve upon current state-of-the-art methods in ac4C identification.

Main Methods:

• Utilized Nucleotide Transformer and DNABERT2 for contextual nucleotide sequence embedding.
• Employed Convolutional Neural Network (CNN) and ResNet18 for feature extraction.
• Trained and validated the NBCR-ac4C model on extensive datasets.

Main Results:

• NBCR-ac4C achieved 83.51% accuracy (ACC) and 89.58% AUROC on an independent test set.
• The model outperformed general learning-based methods for ac4C prediction.
• NBCR-ac4C showed superior performance compared to the SOTA model LSA-ac4C.

Conclusions:

• NBCR-ac4C offers a more efficient and accurate approach to identifying ac4C sites.
• The deep learning framework effectively captures sequence-based relationships for biological insights.
• The model facilitates further research in ac4C-related gene regulation and disease mechanisms.