Grain protein function prediction based on improved FCN and bidirectional LSTM

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput sequencing necessitates advanced methods for predicting protein function from amino acid sequences.
• Existing models often overlook the critical sequential order and long-range dependencies within amino acid chains.
• Accurate protein function prediction is vital for understanding and improving crop traits.

Purpose of the Study:

• To develop and evaluate a novel intelligent model for predicting grain protein function.
• To address the limitations of existing methods in capturing amino acid sequence order and long-term dependencies.
• To improve the accuracy and reliability of protein function prediction in important crop species.

Main Methods:

• Utilized a dataset comprising grain proteins from soybean, maize, indica, and japonica sourced from UniProtKB.
• Proposed the PBiLSTM-FCN model, integrating Fully Convolutional Networks (FCN) for sequence order and bidirectional Long Short-Term Memory (BiLSTM) for long-term dependencies.
• Conducted experimental comparisons against existing models to assess performance.

Main Results:

• The PBiLSTM-FCN model demonstrated superior performance compared to existing methods.
• The model effectively captured long-range dependencies and the order of amino acid sequences, leading to higher prediction accuracy.
• Interpretability analyses confirmed the model's effectiveness by comparing predicted functions with actual protein functions.

Conclusions:

• The PBiLSTM-FCN model represents a significant advancement in predicting grain protein function.
• The model's ability to handle sequence order and long-term dependencies offers a more accurate approach to bioinformatics tasks.
• This work provides a valuable tool for functional genomics and crop improvement research.