Direct prediction of intrinsically disordered protein conformational properties from sequence

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Intrinsically disordered regions (IDRs) are crucial functional elements in proteins, lacking stable 3D structures and existing as dynamic ensembles.
• The structural heterogeneity of IDRs limits their representation in structural databases and hinders computational prediction of their conformational properties.
• Predicting IDR ensemble behavior from sequence is essential for understanding their diverse biological roles.

Purpose of the Study:

• To develop a deep-learning model, ALBATROSS, for predicting ensemble dimensions of IDRs directly from amino acid sequences.
• To enable proteome-wide characterization of IDR biophysical behavior using sequence-based predictions.
• To provide accessible computational tools for studying IDR conformational properties.

Main Methods:

• Combined rational sequence design, large-scale molecular simulations, and deep learning techniques.
• Developed ALBATROSS, a deep-learning model to predict ensemble dimensions like radius of gyration and end-to-end distance.
• Validated the model's generalizability and applied it for proteome-wide analysis.

Main Results:

• ALBATROSS accurately predicts key ensemble dimensions of IDRs directly from their sequences.
• Demonstrated the generalizability of sequence-ensemble relationships across diverse IDRs.
• Enabled high-throughput characterization of sequence-specific biophysical behavior of IDRs within and between proteomes.

Conclusions:

• ALBATROSS provides a powerful, lightweight, and accessible tool for predicting IDR ensemble properties.
• The model facilitates large-scale investigations into the biophysical underpinnings of IDR function.
• This work opens new avenues for understanding the role of intrinsically disordered proteins in biology.