Generative models are necessary to fully capture uncertainty and conformational flexibility of protein structures, but how can we build such models? At the ICLR MLDD workshop, we'll present EigenFold, work led by Bowen Jing with undergrad students Ezra Erives and
@peterpaohuang
!
5
52
266
Replies
@GabriCorso
@br_jimenez
@peterpaohuang
Very cool! The Eigenmode projections reminded me of some old work where we showed that the overall structure of the protein universe—as encoded by SCOP—remains, even if one calculates the "topological" similarity between smoothened chain representations.
0
0
13
@GabriCorso
@peterpaohuang
Super cool work, as usual! Reminds me a lot of Fourier Shape Descriptors from my time in computer vision, visualized nicely by
@3blue1brown
0
0
3
@GabriCorso
@peterpaohuang
What's interesting is that diffusion models usually work the other way around. With more "forward noise" steps, the signal becomes mainly high-frequency and you lose all the low frequency signal.
Here, it's the opposite. You destroy the high-frequency while preserving lower freq
0
0
6