[AtharvaTilewale]

Problem

colabfold_search ... --use-templates 1 without --db2 fails with

FileNotFoundError: Database . does not exist

even though the template database is present in the database directory (#748).

Root cause

--db1 and --db3 default to the database names that setup_databases.sh creates, but --db2 defaulted to Path(""). With --use-templates 1 that empty path is appended to the list of used databases and formats as ., which produces the misleading error message. As noted in the issue thread, the template database simply had no default name.

What the fix does

Give --db2 (and the template_db argument of mmseqs_search_monomer) the default pdb100_230517, the name setup_databases.sh creates, the same way --db1 (uniref30_2302_db) and --db3 (colabfold_envdb_202108_db) already have defaults. --use-templates 1 now works without --db2, and if that database really is missing the error names it instead of ..

Validation

Added tests/test_search.py, which runs colabfold_search through main() (with run_mmseqs stubbed out) against a temp database directory holding only the UniRef stub files, and asserts the resulting FileNotFoundError names the template database.

Pre-patch (git stash push colabfold/mmseqs/search.py, test kept):

$ pytest tests/test_search.py -q
E   AssertionError: Regex pattern did not match.
E     Expected regex: 'pdb100_230517'
E     Actual message: 'Database . does not exist'
1 failed

Post-patch:

$ pytest tests/test_search.py -q
1 passed

$ colabfold_search --help
  --db2 DB2             Templates database (default: pdb100_230517)

$ pytest -q
29 passed, 20 skipped, 7 errors

The 7 errors are tests/test_colabfold.py prediction tests, which need network access to download parameters/MSAs; they error identically on unmodified main in this environment (baseline there is 28 passed, 20 skipped, 7 errors).

Fixes #748

Disclosure: this change was prepared with AI assistance and validated locally as described above.