Data-driven prediction of α_IIbβ₃ integrin activation paths using manifold learning and deep generative modeling

Area of Science:

• Biophysics
• Computational Biology
• Structural Biology

Background:

• Integrin heterodimers are key transmembrane proteins regulating cellular functions.
• Integrin malfunction is implicated in various diseases, making them therapeutic targets.
• Understanding integrin activation pathways is crucial but challenging due to transient intermediate states.

Purpose of the Study:

• To computationally predict the full activation pathways of integrins.
• To develop a transferable multiscale technique for biomolecular transition pathway prediction.
• To uncover new therapeutic targets by elucidating integrin biophysical processes.

Main Methods:

• Nonlinear manifold learning applied to coarse-grained molecular dynamics simulations.
• Deep generative models trained for inverse mapping between low-dimensional and high-dimensional molecular spaces.
• Interpolation of molecular configurations to reconstruct activation pathways between known states.

Main Results:

• Plausible predictions of integrin activation pathways for α_IIbβ₃ integrin.
• Successful learning of a low-dimensional embedding of the configurational phase space.
• Generation of intermediate molecular configurations constituting the activation pathways.

Conclusions:

• The developed multiscale technique provides a generic and transferable method for predicting biomolecular transition pathways.
• This approach offers fundamental insights into the biophysical processes of integrin activation.
• The predicted pathways can guide the identification of novel therapeutic strategies for integrin-related diseases.