PSSP-MVIRT: peptide secondary structure prediction based on a multi-view deep learning architecture.

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• Peptide secondary structure prediction is crucial for understanding peptide function and therapeutic potential.
• Accurate prediction aids in drug discovery and development.

Purpose of the Study:

• To develop an advanced computational method for accurate peptide secondary structure prediction.
• To improve the understanding of peptide functional mechanisms through structural insights.

Main Methods:

• Proposed Peptide Secondary Structure Prediction based on Multi-View Information, Restriction and Transfer learning (PSSP-MVIRT), a multi-view deep learning approach.
• Integrated sequential, evolutionary, and hidden state information using a multi-view fusion strategy.
• Employed a hybrid Convolutional Neural Network (CNN) and Bi-directional Gated Recurrent Unit (Bi-GRU) architecture.
• Utilized transfer learning to address limited training data.

Main Results:

• PSSP-MVIRT significantly outperformed existing state-of-the-art methods on independent tests.
• The model demonstrated superior performance in segment-level prediction, indicating strong local feature extraction capabilities.
• Case studies confirmed the method's robust performance in predicting secondary structures for novel peptides.

Conclusions:

• The PSSP-MVIRT method offers a significant advancement in peptide secondary structure prediction.
• The developed webserver (http://server.malab.cn/PSSP-MVIRT) provides a valuable tool for researchers.
• This work facilitates the application of accurate peptide structure prediction in therapeutic development.