DiffFit: Differentiable Fitting of Molecule Structures to a Cryo-EM mAP

Abstract

We introduce DiffFit, a differentiable algorithm for fitting protein atomistic structures into experimental Cryo-Electron Microscopy (cryo-EM) volume maps. This process is crucial in structural biology for reconstructing large meso-scale models of complex protein assemblies and complete cellular structures. Unlike current methods requiring manual fitting followed by automated fine-tuning, DiffFit enables automatic fitting with visual inspection and interactive revision. Our approach employs differentiable 3D rigid transformations and a novel loss function based on multi-resolution volume arrays and negative space exploitation. Evaluations on three use cases using realistic datasets demonstrate that DiffFit significantly outperforms previous methods. DiffFit is available as an open-source plugin (github.com/nanovis/DiffFitViewer) in ChimeraX, with all supplemental materials accessible at osf.io/5tx4q.

Cite

Text

Luo et al. "DiffFit: Differentiable Fitting of Molecule Structures to a Cryo-EM mAP." ICML 2024 Workshops: Differentiable_Almost_Everything, 2024.

Markdown

[Luo et al. "DiffFit: Differentiable Fitting of Molecule Structures to a Cryo-EM mAP." ICML 2024 Workshops: Differentiable_Almost_Everything, 2024.](https://mlanthology.org/icmlw/2024/luo2024icmlw-difffit/)

BibTeX

@inproceedings{luo2024icmlw-difffit,
  title     = {{DiffFit: Differentiable Fitting of Molecule Structures to a Cryo-EM mAP}},
  author    = {Luo, Deng and Alsuwaykit, Zainab and Khan, Dawar and Strnad, Ondrej and Isenberg, Tobias and Viola, Ivan},
  booktitle = {ICML 2024 Workshops: Differentiable_Almost_Everything},
  year      = {2024},
  url       = {https://mlanthology.org/icmlw/2024/luo2024icmlw-difffit/}
}