ADOPT: intrinsic protein disorder prediction through deep bidirectional transformers.

Area of Science:

• Proteomics
• Computational Biology
• Biochemistry

Background:

• Intrinsically disordered proteins (IDPs) play crucial roles in biological processes and disease.
• Experimental analysis of dynamic IDPs is challenging.
• Computational prediction of protein disorder from amino acid sequences is an active research area.

Purpose of the Study:

• To introduce ADOPT (Attention DisOrder PredicTor), a new computational method for predicting protein disorder.
• To evaluate ADOPT's performance against existing disorder prediction tools.
• To identify key features contributing to accurate disorder prediction.

Main Methods:

• ADOPT utilizes a self-supervised encoder based on a deep bidirectional transformer (Evolutionary Scale Modeling library).
• A supervised disorder predictor is trained on a curated dataset of nuclear magnetic resonance chemical shifts.
• The training dataset is balanced for disordered and ordered residues.

Main Results:

• ADOPT demonstrates superior performance in predicting protein disorder compared to existing state-of-the-art predictors.
• ADOPT offers rapid prediction, typically within seconds per sequence.
• Effective prediction can be achieved using fewer than 100 features.

Conclusions:

• ADOPT provides a highly accurate and efficient computational approach for identifying intrinsically disordered proteins and regions.
• The tool facilitates research into IDPs, crucial for understanding biological functions and developing targeted therapeutics.
• ADOPT is accessible as a standalone package and a web server for broader scientific use.