Combining knowledge distillation and neural networks to predict protein secondary structure.

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in proteomics

Background:

• Protein secondary structure is crucial for 3D conformation, function, and biological roles.
• Accurate secondary structure prediction enhances understanding of protein interactions and mechanisms.
• Deep learning models excel at processing complex sequence data for improved prediction accuracy.

Purpose of the Study:

• To develop and evaluate the ITBM-KD model for enhanced protein secondary structure prediction.
• To improve prediction accuracy for octapeptides and tripeptides using a combined deep learning approach.
• To provide a robust and generalizable model for protein structure-function analysis.

Main Methods:

• Integration of an improved temporal convolutional network (TCN), bidirectional recurrent neural network (BiRNN), and multilayer perceptron (MLP).
• Utilized one-hot encoding, word vector representation of physicochemical properties, and knowledge distillation with the ProtT5 model.
• Validated the model on classic datasets (TS115, CB513) and a large PDB dataset (15,078 entries).

Main Results:

• The ITBM-KD model demonstrated excellent performance across multiple datasets.
• Achieved high accuracy in predicting secondary structures for octapeptides and tripeptides.
• Verified model robustness and generalizability on extensive protein data.

Conclusions:

• The ITBM-KD model significantly improves protein secondary structure prediction accuracy.
• Provides a valuable tool for understanding protein structure and function, particularly in resource-limited environments.
• Facilitates biomedical research by offering insights into protein interactions and mechanisms.