duan_jandClaude Fable 5
4ffc1b152d
Separate module from PanDDA (different library): self-contained python 3.9 venv (torch==1.13.1 caps at py3.10), torch from the pytorch CPU wheel index, ~1 GB of CNN scoring models pre-fetched at build time because the package dir is read-only for users under /opt/psi. ccp4/9.0 runtime dep only, for pandda.inspect / pandda.export. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0152MuuYUaapP7CtxrDAfvBh
1.2 KiB
PanDDA2
PanDDA 2 (pandda_2_gemmi) — multi-dataset density analysis for fragment
screening. https://github.com/xchem/PanDDA2
Not on PyPI: installed with uv into a self-contained python 3.9 venv from a
sha-pinned commit of xchem/PanDDA2 master (no usable release tags upstream;
module version 1.0.0 follows the README's announced version — bump the sha in
build to update). Torch comes from the pytorch CPU wheel index to avoid the
~2 GB CUDA build.
The ~1 GB of CNN scoring models that PanDDA2 normally gdown-fetches on first
run are downloaded at build time into the installed package dir, because that
dir is read-only for users under /opt/psi. If Ra blocks Google Drive, download
the 5 files by hand (ids in build) and copy them into the dir named in the
die message.
ccp4/9.0 is a runtime dependency (auto-loaded) for pandda.inspect /
pandda.export. The build needs no ccp4 — the venv is self-contained.
Upstream README claims inspect/export "do not function correctly" outside
ccp4/7.0.067; verify with 9.0 before trusting exports.
Verify after load:
pandda2.analyse --help
$PANDDA2_ENV/bin/python -c 'from pandda_gemmi.cnn import LitEventScoring'