Show HN: CUDA, Shmuda: Fold Proteins on a MacBook

(latentspacecraft.com)
11 points geoffitect | 4 comments | 17 Nov 25 00:08 UTC | HN request time: 0.683s | source

Alphafold3 used to be fodder for HPC clusters; now I've got a port running smoothly on Apple Silicon. If you have an M-series Mac (~2023-present), you can generate protein structures from sequences in minutes. Give it a try!

GitHub repo: https://github.com/latent-spacecraft/openfold-3-mlx

1. cyrusradfar ◴[17 Nov 25 01:03 UTC] No.45949987[source]▶
Curious to try, thanks for putting this out.

What did you find to be the most surprising challenge in this project?

replies(1): >>45950022 #
2. geoffitect ◴[17 Nov 25 01:11 UTC] No.45950022[source]▶
^How convoluted MSA and template handling is under the hood in openfold. I’m working on a simple offline replacement next. Compared to that, the attention functions were fairly easy to convert to MLX ops.
3. ray__ ◴[25 Nov 25 16:31 UTC] No.46047485[source]▶
Can you fold very large proteins/complexes with the large amount of VRAM available on Macs? Ram limitations forcing folding runs to proteins ~<1500 is an annoying nit for a lot of protein folding workflows for me—I'd be curious to see if this helps.