LGC-DBP: the method of DNA-binding protein identification based on PSSM and deep learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• DNA Binding Proteins (DBPs) are vital for key biological processes like replication, transcription, and repair.
• Current sequence-based DBP prediction methods often underutilize deep learning's potential for complex pattern recognition.

Purpose of the Study:

• To develop an advanced deep learning model, LGC-DBP, for enhanced DNA Binding Protein prediction.
• To improve the accuracy and interpretability of DBP recognition by capturing intricate sequence-based features.

Main Methods:

• Developed LGC-DBP, a model combining Long Short-Term Memory (LSTM), Gated Inception Convolution, and Improved Channel Attention.
• Protein sequences were converted to Position Specific Scoring Matrices (PSSM) for deep learning framework input.
• Gated Inception Convolution integrated gating units with Graph Convolutional Network (GCN) and Dilated Convolution; Improved Channel Attention used multi-input sigmoid functions.

Main Results:

• LGC-DBP achieved a prediction accuracy of 88.26%.
• The model obtained a Matthews correlation coefficient of 0.701.
• Performance metrics surpassed existing DBP prediction methods, demonstrating superior multi-dimensional data integration and nonlinear interaction capture.

Conclusions:

• LGC-DBP represents a significant advancement in DNA Binding Protein recognition.
• The model effectively captures complex, nonlinear relationships within protein sequence data.
• The findings highlight the potential of integrating advanced deep learning architectures for biological sequence analysis.