google-deepmind / google-deepmind/alphafold
Port to Metal and macOS
- Dominant language
- Python
- Stars
- 14.9k
- Forks
- 2.9k
- PR merge metrics
- No merged PRs in 30d
Description
I've been taking a shot at porting this from NVIDIA/CUDA to M1/Metal, so that I can run it on my personal MacBook Pro (1 TB disk). I anticipated being able to port the entire framework in a single day, but faced significant challenges regarding disk space. It would have been much less challenging if the total database collection were even just halved in size.
I wish to engineer proteins that never existed before in biology, and may have no evolutionary history. AlphaFold would let me rapidly search the solution space of amino acid sequences, until encountering proteins that suit my needs. I fear that AlphaFold, heavily tuned for MSAs, would perform poorly on proteins foreign to biological evolution. Nevertheless, this would be a great tool in my toolbox.
I'm thinking of either (1) purchasing an external HDD or (2) seeing whether you reduce the collective database size from 600 (actually 700) GB to something like 200 GB. When/if I finish the port, would you consider merging these contributions into the main branch? I don't think that's very likely, but "never say never".
Fork and documentation of my porting efforts: https://github.com/philipturner/alphafold-metal
Contributor guide
Assessment
This issue has not been assessed yet.