Sampling Conformational Ensembles of Highly Dynamic Proteins via Generative Deep Learning.

Area of Science:

• Biophysics
• Computational Biology
• Structural Biology

Background:

• Protein conformational ensembles are critical for biological function, but their study is computationally challenging.
• Intrinsically disordered proteins (IDPs) exhibit dynamic conformational landscapes crucial for function and disease-related aggregation.
• Understanding protein dynamics requires advanced methods for comprehensive sampling of conformational spaces.

Purpose of the Study:

• To develop a novel deep learning model, Internal Coordinate Net (ICoN), for efficient protein conformational sampling.
• To investigate the conformational landscape of amyloid-beta 42 (Aβ42) monomer using the ICoN model.
• To identify novel protein conformations and rationalize experimental findings through computational analysis.

Main Methods:

• Introduction of Internal Coordinate Net (ICoN), a deep learning model trained on molecular dynamics simulation data.
• Interpolation in the learned latent space to generate novel synthetic protein conformations.
• Application of ICoN to comprehensively sample the conformational landscape of Aβ42 monomer.

Main Results:

• ICoN successfully learned physical principles of protein conformational changes.
• Generated synthetic conformations revealed distinct clusters explaining experimental observations.
• Identified novel conformations with detailed atomistic interactions not present in training data.
• New conformations showed side chain rearrangements consistent with experimental data (EPR, amino acid substitution).

Conclusions:

• Deep learning, via ICoN, offers a powerful and transferable approach for protein conformation sampling.
• The method effectively utilizes natural atomistic motions for enhanced sampling of protein dynamics.
• ICoN provides a comprehensive view of protein conformational landscapes, aiding in understanding function and disease.