Protein remote homology detection based on bidirectional long short-term memory.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Protein remote homology detection is crucial for understanding protein structure and function.
• Traditional machine learning methods struggle with extracting discriminative features from variable-length protein sequences.
• Deep learning offers advantages in automatic feature representation learning.

Purpose of the Study:

• To explore the application of deep learning, specifically Bidirectional Long Short-Term Memory (BLSTM), for protein remote homology detection.
• To develop a novel predictor, ProDec-BLSTM, capable of automatically learning effective protein features.

Main Methods:

• Utilized Bidirectional Long Short-Term Memory (BLSTM) networks.
• Developed the ProDec-BLSTM predictor architecture, comprising input, BLSTM, time-distributed dense, and output layers.
• Employed pseudo proteins as input for feature learning.

Main Results:

• ProDec-BLSTM demonstrated superior performance compared to existing methods on a benchmark dataset.
• Achieved higher mean ROC and mean ROC50 scores, indicating improved accuracy in remote homology detection.
• The model successfully learned and extracted discriminative features automatically.

Conclusions:

• ProDec-BLSTM is an effective tool for protein remote homology detection.
• The model's ability to learn and visualize hidden patterns provides additional insights into protein relationships.
• Deep learning approaches, like BLSTM, show significant promise in advancing bioinformatics tasks.