Identifying short disorder-to-order binding regions in disordered proteins with a deep convolutional neural network...

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Molecular recognition features (MoRFs) are crucial functional regions in intrinsically disordered proteins (IDPs).
• IDPs are involved in cellular molecular interactions and are linked to various human diseases.
• Accurate identification of MoRFs is vital for understanding protein function and enabling drug design.

Purpose of the Study:

• To develop a powerful and accurate computational model for predicting MoRFs in IDPs.
• To leverage artificial intelligence and deep learning for enhanced MoRF prediction.
• To provide a freely accessible web server for the proposed MoRF prediction tool.

Main Methods:

• Proposed en_DCNNMoRF, an ensemble deep convolutional neural network (DCNN) based predictor.
• Developed two independent DCNN classifiers (DCNNMoRF1 and DCNNMoRF2) utilizing distinct sequence features (PSSM, amino acid factors, and indexes).
• Incorporated a novel two-dimensional attention mechanism and an averaging strategy for improved prediction accuracy.

Main Results:

• The en_DCNNMoRF model demonstrated accuracy comparable to state-of-the-art methods on benchmark datasets.
• The ensemble approach effectively combined predictions from two distinct DCNN architectures.
• A web server for the en_DCNNMoRF predictor is available for public use.

Conclusions:

• The developed en_DCNNMoRF method offers a significant advancement in MoRF prediction accuracy.
• This AI-driven approach facilitates research into IDP functions and disease mechanisms.
• The accessible web server promotes wider application in biological research and drug discovery.