Remove remote BLAST support and improved local BLAST

[ShinThomas]

Summary

Removed remote/online BLAST implementation from the refactor-v2 codebase while preserving the local BLAST workflow. Updated BLAST support in the refactor-v2 codebase by removing remote BLAST functionality and improving the local BLAST workflow.

Changes

• Removed RemoteBlastSearcher from src/lverage/blast.py

• Removed remote-only BLAST dependencies and constants:

  • requests
  • NCBI_EFETCH_URL
  • unused SeqIO

• Deleted tests/v2/test_remote_blast.py

• Preserved LocalBlastSearcher

• Preserved shared BLAST parsing and ortholog-record logic

• Confirmed no changes were required in:

  • src/lverage/pipeline.py
  • src/lverage/ortholog_searcher.py

Documentation

docs/api/blast.rst still contains references to remote BLAST and RemoteBlastSearcher.

Remote BLAST changes

• Removed RemoteBlastSearcher
• Removed remote-only BLAST dependencies and constants
• Deleted tests/v2/test_remote_blast.py
• Preserved the local BLAST implementation

Local BLAST changes

• Updated blastp invocation to explicitly read the protein query from stdin using -query -
• Updated query coverage calculation to use BLAST query coordinates
• Added fallback sequence retrieval using the accession when retrieval by BLAST hit ID fails
• Skips hits when the full protein sequence cannot be retrieved by either identifier

Testing

Focused local BLAST tests:

12 passed

The local BLAST tests cover:

• local blastp command construction
• FASTA query input through stdin
• coordinate-based query coverage
• complete sequence retrieval with blastdbcmd
• fallback from hit ID to accession
• failure handling when sequence retrieval fails

[ShinThomas]

Documentation build currently fails because docs/api/blast.rst still references the removed RemoteBlastSearcher. The docs directory was inspected but intentionally not modified.

[AnthonyGarza124]

This changes “query coverage” from alignment-column length to the query-coordinate span of the best HSP. That seems reasonable and avoids gap-driven values above 100%, but it is slightly different from counting the exact query residues aligned.

I’m happy to accept this behavior for now. Could we add or track a follow-up test with gapped HSPs (and possibly multiple HSPs) to document the intended semantics?