Research on DNA-Binding Protein Identification Method Based on LSTM-CNN Feature Fusion

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• DNA-binding proteins are crucial for life activities, necessitating efficient identification methods.
• Traditional biotechnology methods for protein identification face limitations in cost and efficiency.
• Machine learning and deep learning offer advanced solutions for processing complex biological data.

Purpose of the Study:

• To develop a novel deep learning framework for accurate identification of DNA-binding proteins.
• To enhance the extraction of sequence and evolutionary features from protein data.
• To improve upon existing computational methods for DNA-binding protein prediction.

Main Methods:

• A deep learning framework integrating parallel Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) was developed.
• The model processes both protein sequence information and evolutionary features.
• Combined features were utilized for training and testing on the PDB2272 dataset.

Main Results:

• The proposed deep learning model demonstrated superior performance in identifying DNA-binding proteins.
• Compared to the PDBP_Fusion model, Accuracy (ACC) increased by 3.82%.
• Matthew's Correlation Coefficient (MCC) saw an improvement of 7.98%.

Conclusions:

• The parallel LSTM-CNN deep learning framework is effective for DNA-binding protein identification.
• The model's ability to extract diverse features contributes to its enhanced performance.
• This approach offers a significant advancement over existing computational methods in bioinformatics.