KAUST Research Conference

Computational Advances in Structural Biology

May 1 - 3, 2023 Auditorium between building 4 & 5

Machine Learning for Reconstructing Molecular Flexibility in Cryo- Electron Microscopy


Abstract:

Cryogenic electron microscopy (cryo-EM) is a powerful technique to 
obtain the 3D structure of macromolecules from thousands of noisy 
projection images. Since these macromolecules are flexible by nature, 
the areas shows lower resolution and gives a blurry reconstruction. We 
propose a novel method incorporated in a software package named 
dynamight, that represents the molecule with gaussian basis functions 
and estimates deformation fields for every experimental image by a 
variational autoencoder. We further use the estimated deformations to 
better resolve the flexible regions in the reconstruction using a 
filtered backprojection algorithm along curved lines. We present results 
on real data showing that we obtain improved 3D reconstruction.

  • Share this: