(474a) Predicting Activation Pathway of Integrin Using Generative Adversarial Networks and Targeted Molecular Dynamics Simulations

Integrins are transmembrane proteins that play a critical role in cell adhesion and cell signaling.[1,2] Upon activation, integrins undergo a large-scale conformational transition from a bent to an extended state. Impaired conformational transitions of integrin are linked to various diseases such as bleeding disorders, immunodeficiency, and cancer.[3] To this end, understanding integrin activation is of fundamental interest and is critical to the development of therapeutics targeting integrin.[1-5]

While crystal structures of bent conformations of different integrins are well-characterized, only fractions of extended conformations are available. Additionally, there is a limited understanding of intermediate conformations of integrin using experimental methods because of their transient nature. Alternatively, studying integrin activation using conventional molecular dynamics simulations require inordinate computational resources because integrins are large heterodimer proteins comprising ~1770 residues. Our goal is to address this problem in characterizing unknown intermediate conformations of integrin by combining generative adversarial networks (GAN)[6] and targeted molecular dynamics simulations.

We demonstrate our approach by training GAN using coarse-grained representations of É‘_IIbβ₃ integrin in an active, two partially inactive and an inactivate state, which are sampled via unbiased all-atom molecular dynamics simulations.[4] We utilize adaptive diffusion maps[7] to learn the latent space of the integrin. Targeted molecular dynamics simulations are employed to predict the all-atom conformations from the coarse-grained transition state structures generated by GAN. Thereby, allowing access to previously unknown all-atom conformations along the activation pathway of integrin. In our talk, we will present our approach and the predicted all-atom conformations of É‘_IIbβ₃ integrin in the transition region and discuss their biophysical validity and significance.

References

[1] Bidone TC, Polley A, Jin J, Driscoll T, Iwamoto DV, Calderwood DA, Schwartz MA, Voth GA. Coarse-grained simulation of full-length integrin activation. Biophysical journal. 2019 Mar 19;116(6):1000-10.

[2] Banno A, Ginsberg MH. Integrin activation. Biochemical Society Transactions. 2008 Apr 1;36(2):229-34.

[3] Zhang Y, Wang H. Integrin signalling and function in immune cells. Immunology. 2012 Apr;135(4):268-75.

[4] Goodman SL, Picard M. Integrins as therapeutic targets. Trends in pharmacological sciences. 2012 Jul 1;33(7):405-12.

[5] Tong D, Soley N, Kolasangiani R, Schwartz MA, Bidone TC. αIIbβ3 integrin intermediates: from molecular dynamics to adhesion assembly. Biophysical Journal. 2022 Dec 23.

[6] Sidky H, Chen W, Ferguson AL. Molecular latent space simulators. Chemical Science. 2020;11(35):9459-67.

[7] Wang J, Gayatri MA, Ferguson AL. Mesoscale simulation and machine learning of asphaltene aggregation phase behavior and molecular assembly landscapes. The Journal of Physical Chemistry B. 2017 May 11;121(18):4923-44.