TPSO-DBP

TPSO-DBP predicts DNA-binding proteins from protein sequences by extracting three-part sequence-order pseudo features and applying a deep neural network to improve identification accuracy.

Key Features:

• Three-Part Sequence-Order Feature Extraction (TPSO): Divides the primary sequence into N-terminal and C-terminal fragments in addition to the full sequence and extracts numerical pseudo features from each part to increase discriminative information.
• Sequence-based single-view features: Incorporates sequence-based single-view features as input to the predictive model.
• Deep neural network architecture: Employs bidirectional long short-term memory (BiLSTM) networks and fully connected (FC) layers to model complex nonlinear relationships between input features and DNA-binding protein labels.
• Performance: Reported accuracy of 87.01%, coverage of 85.30% for DBPs, and Matthew's correlation coefficient of 0.741, outperforming many existing methods.

Scientific Applications:

• Improved DBP identification: Addresses limitations of existing computational methods by leveraging sequence-order information to improve DNA-binding protein prediction.
• Gene regulation and molecular studies: Enables discovery of concealed sequence patterns relevant to gene regulation, molecular biology, and genetic engineering.
• High-coverage protein annotation: Provides a computational approach for researchers to predict DNA-binding proteins with higher accuracy and coverage from protein sequences.

Methodology:

TPSO divides sequences into full, N-terminal, and C-terminal parts to extract numerical pseudo features, integrates sequence-based single-view features, and trains a deep neural network composed of BiLSTM layers followed by fully connected layers.