Identifying Intrinsically Disordered Protein Regions through a Deep Neural Network with Three Novel Sequence Features

Area of Science:

• Computational Biology
• Bioinformatics
• Protein Science

Background:

• Intrinsically disordered protein regions (IDPRs) play vital roles in numerous cellular functions.
• Accurate IDPR identification is essential for understanding molecular recognition, gene regulation, and signal transduction.
• Developing cost-effective computational methods for IDPR identification is imperative.

Purpose of the Study:

• To develop a novel deep neural network for identifying IDPRs.
• To introduce and evaluate new sequence features for improved IDPR prediction.
• To offer a computationally efficient approach for IDPR identification.

Main Methods:

• A deep neural network architecture combining a VGG19 variant with two MLP networks was designed.
• Three novel sequence features were introduced: persistent entropy and probabilities of di- and tri-amino acid sequences.
• The model was trained and evaluated using simulation data.

Main Results:

• The proposed deep neural structure demonstrated superior performance compared to existing methods.
• The novel sequence features proved effective, even with smaller training datasets.
• The VGG19-based deep neural structure achieved high accuracy in IDPR identification.

Conclusions:

• The developed deep neural structure is effective for identifying IDPRs.
• Novel sequence features, including persistent entropy and amino acid pair probabilities, are valuable for IDPR prediction.
• This approach offers a promising direction for future IDPR identification research.