RT @ZhongingAlong: 🚨New paper at #NeurIPS2024! We present Hydra, a mixture of neural fields 🐉🐉🐉 for ab initio cryo-EM reconstruction of com…

@bj_charles Yes, something like this. Maybe with a slope after 30A that progressively increases throughout optimization to make things more stable?

@mipt98 @pedrobeltrao Hi! For crystallography, I would say that once peak finding, indexing and integration have been done, our technique could potentially be used to refine an atomic structure given scattering amplitudes. This is an exciting direction!

@bj_charles This would definitely be a useful application! However in this case the "likelihood" (which quantifies how much the model fits the measurements) would be a non-continuous function and additional tricks would be necessary to make gradient-based optimization effective.

This work is the fruit of a fantastic collaboration with @ericryanchan, Sara Fridovich-Keil, @fredericpoitev1, @ZhongingAlong and @GordonWetzstein! We will make the code available as soon as possible. 5/

Shout out to @MGrzadkows26472 for the software development, #EZLab for beta-testing and all the team @fredericpoitev1, @Frances84469032, @OliBClarke, @GordonWetzstein, @ZhongingAlong! We're excited to get feedback from the community! 7/

DRGN-AI handles low-SNR particles (e.g. here, spike protein, 438 kDa) as well as datasets with many "junk" images (e.g. DSL1/SNARE complex, see paper). 4/

We validate our algorithm on several benchmark EMPIAR datasets - except we analyze the full particle stack instead of a filtered version like previous tools. 3/

DRGN-AI relies on a two-step hybrid approach combining exhaustive search and gradient-based optimization, making it robust to noise and computationally efficient. 2/

RT @GoogleAI: Presenting MELON, a technique that can determine object-centric camera poses entirely from scratch while reconstructing the o…

RT @ZhongingAlong: We present a new algorithm, cryoFIRE ❄️🔥, for fast ab initio heterogeneous reconstruction, appearing at @NeurIPSConf thi…

preprint: