Identification and prediction of the most likely electroporation sites in cell membranes assisted by molecular dynamics and machine learning
Authors: Lea Rems, Xinru Tang, Fangwei Zhao, Sergio Pérez-Conesa, Ilaria Testa, Lucie Delemotte
Publication: Rems, L., Tang, X., Zhao, F., Pérez-Conesa, S., Testa, I., & Delemotte, L. (2022). Identification of electroporation sites in the complex lipid organization of the plasma membrane. Elife, 11, e74773.
In this repository you can find links to our open source code and data:
- Type in your title Lea developed by [Lea Rems](https://www.biophysics.se/index.php/alumni/lea-rems/
- Bayesian survival analysis code developed by Sergio Pérez-Conesa. )
- OSF repository containing the data and machine learning models of this work, developed by Lea Rems et al.
The plasma membrane of a biological cell is a complex organization of numerous types of lipids and membrane proteins. In physiological conditions the plasma membrane tightly regulates transmembrane transport. However, different factors can transiently disrupt the membrane integrity with formation of transmembrane pores. One of such factors is an electric field, which can cause membrane electroporation. While electroporation is already used in many applications in medicine and biotechnology, there are still many open questions about the nature of pores formed in plasma membranes and their location. In this study we propose a method combining coarse grained molecular dynamics simulations and machine learning methods, that can allow us to identify and predict electroporation sites in membranes mimicking realistic composition of plasma membranes. We ran ~1000 electroporation simulations on membranes with >60 different lipids types and we identified features that govern the membrane’s propensity for poration. We used Bayesian survival analysis to model the poration kinetics and its variation with membrane composition. We then trained different machine learning models to investigate how well they can predict poration sites in different membranes, and we were typically able to get more than 80% accurate prediction of porated and nonporated locations. We anticipate that these approaches will allow us to efficiently predict membrane sites, which act as weak points for electromechanical perturbation, and model the corresponding electroporation kinetics, in systems with greater complexity and size, including model plasma membranes with proteins and actin cytoskeleton.
Please make sure to contact Lea Rems (lea.rems at fe.uni-lj.si) or Lucie Delemotte (lucie.delemotte at scilifelab.se) if you have any questions or suggestions!