Thread by @MoreheadAlex

Thrilled to share new state-of-the-art 3D molecule generation results with our new Geometry-Complete Diffusion Model (GCDM). I'll be presenting GCDM tomorrow at #ICLR2023 MLDD. Look forward to seeing you there!
Paper: arxiv.org/abs/2302.04313
Code: github.com/BioinfoMachine…

We find that our model's (1) sensitivity to molecular chirality and (2) application of attention for each molecular graph's edges both uniquely contribute to our model's success.

This model configuration lets us achieve significant improvements in generating 3D molecules with specific molecular properties.

We can even scale up nicely to generate 3D molecules using the GEOM-Drugs dataset. In fact, GCDM is the first method (to the best of our knowledge) that can generate any stable large molecules for GEOM-Drugs in this setting.

Lastly, as a first-of-its-kind result, we show that diffusion models for 3D molecules can successfully and effectively optimize existing molecules for specificity to particular molecular properties, offering improvements to their molecular stability along the way.