Prediction of DNA binding proteins using local features and long-term dependencies with primary sequences based on...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• DNA-binding proteins (DBPs) are crucial for various biological processes.
• Existing machine learning and deep learning methods for DBP prediction often require manual feature engineering or struggle with long DNA sequence dependencies.

Purpose of the Study:

• To propose PDBP-Fusion, a novel deep learning method for accurate DNA-binding protein identification.
• To overcome limitations of traditional methods by integrating local features and long-term dependencies from primary sequences.

Main Methods:

• Utilized Convolutional Neural Network (CNN) for local feature extraction.
• Employed Bi-directional Long Short-Term Memory (Bi-LSTM) network to capture long-term sequence dependencies.
• Integrated feature extraction, model training, and prediction within a single framework.

Main Results:

• Achieved 86.45% sensitivity, 79.13% specificity, 82.81% accuracy, and 0.661 Matthews Correlation Coefficient (MCC) on the PDB14189 dataset.
• Demonstrated at least a 9.1% increase in MCC compared to other advanced prediction models.
• Showcased superior performance and robustness on the independent PDB2272 dataset.

Conclusions:

• PDBP-Fusion accurately and effectively predicts DNA-binding proteins directly from primary sequences.
• The method successfully integrates local and long-range sequence information.
• An online server for PDBP-Fusion is available for public use.