NIAID progress report for group meeting May 19 (last report) - July 14, 2022.

Tom Goddard
July 14, 2022

NIH 3D Pipeline

• Eric can describe progress.
• 3DPresets bundle installation issues.

DICOM

• Zach can describe progress.

ColabFold: a 5-times faster AlphaFold

• Switching ChimeraX AlphaFold prediction to use ColabFold.
• Uses a separate server to make sequence alignment (48 cores, 768 Gbytes of memory) very fast, hosted at Korean Bioinformation Center (KOBIC, similar to NCBI).
• ColabFold also runs the GPU structure calculation on Google Colab.
• Also optimizes expensive GPU compilation of neural nets.
• Also uses full size databases, and also can use PDB structure templates.
• Last 3 months, 12000 ChimeraX AlphaFold runs (140 per day), by 2700 unique IP addresses.
• Asked ColabFold developers Sergey Ovchinnikov at Harvard, and Martin Steinegger at Seoul National University if ChimeraX can use their server.
• Sergey and Martin are enthusiastic users of ChimeraX, responded immediately with a dozen emails.
• They are fine with ChimeraX using their sequence server and ColabFold.
• The ColabFold sequence server currently handles 12000 alignments per day from the ColabFold web site.
• ColabFold uses a PyPi module on the Google Colab machine which also makes nice matplotlib plots that ChimeraX can show during the computation.

Native Mac M1 ChimeraX - Funded by CZI grant

• Setup nightly technology preview build on Mac Mini (euclid.cgl.ucsf.edu) including notarization.
• Made Mac universal distribution by combining ARM and Intel ChimeraX builds and tested.
• Still waiting for Phil Thomposon of Riverbank to fix PyQt6 commercial wheels. Said 2 days last week, now he says end of this week.
• Mac universal build (ticket #7222) dmg is 569 Mbytes (43% bigger) versus 398 Mbytes for Intel Mac dmg.
• Zach found LZMA compression can make ChimeraX dmg 30-40% smaller (ticket #7274). Needs macOS 10.15.