Experimental fluorescent-protein structures, downloaded from RCSB and kept here so no experiment folder re-downloads them. Data only — no code.
| path | what it is |
|---|---|
experimental/ |
174 .pdbx entries, one per PDB ID that some part of the repo resolves a design window against |
1GFL.pdbx, 1G7K.pdbx, 3M22.pdbx |
the three whole-barrel scaffold references hard-coded in fpdesign/pockets.py's STRUCT map (avGFP, DsRed, eqFP578) |
| consumer | how |
|---|---|
fpdesign/build_design_windows.py |
STRUCTDIR = REPO / "structures" / "experimental" |
msa_conservation/conservation.py |
1GFL.pdbx, for the barrel-core reference |
in-silico-test/ |
in-silico-test/structures is a symlink to this folder; its design_common.STRUCT_DIR resolves here |
fpdesign/pockets.py calls biotite.database.rcsb.fetch(pdb, "pdbx", structdir), which downloads
only on a cache miss. So a PDB ID that isn't here yet is fetched on first use and written into
experimental/ — nothing needs pre-seeding, and deleting a file only costs one re-download.
The cache is a strict superset of every PDB ID referenced by live code in the repo: the
scaffold_pdb fields of the campaign and in-silico-test design windows, every row of
in-silico-test/structure_hits.csv (the "this protein has a ≥97%-identity PDB entry" table, so a
future cohort draw finds its structures already cached), and the three hard-coded scaffold
references above.
23 entries that only the archived peak-conditioned experiment ever touched were moved to
archive/structures/experimental/ (untracked). They are not referenced by any live code, and
pockets.py would re-fetch any of them on demand if that changed.