[Bug] KeyError in score_interactions when some interactions missing from scoring matrix

[cafferychen777]

Bug Description

When running cpdb_statistical_analysis_method.call() with score_interactions=True, a KeyError is raised because some interaction pairs in interaction_scores['interacting_pair'] are not found in the interacting_pair2score dictionary.

Error Location

File: cellphonedb/utils/scoring_utils.py, line 293:

interaction_scores[ct_pair] = [interacting_pair2score[id] for id in interaction_scores['interacting_pair']]

Error Message

KeyError: 'BMP8A_ACVR_1A2B_receptor'

Steps to Reproduce

1. Load a human spatial transcriptomics dataset (e.g., 10x Visium human lymph node)
2. Run CellPhoneDB v5 statistical analysis with score_interactions=True

from cellphonedb.src.core.methods import cpdb_statistical_analysis_method

result = cpdb_statistical_analysis_method.call(
    cpdb_file_path=db_path,
    meta_file_path=meta_file,
    counts_file_path=counts_file,
    counts_data='hgnc_symbol',
    threshold=0.1,
    iterations=100,
    output_path=output_dir,
    score_interactions=True,  # This causes the error
)

Expected Behavior

The scoring should handle missing interactions gracefully, either by:

1. Skipping interactions not found in the scoring matrix
2. Assigning a default score (e.g., 0 or NaN)

Workaround

Setting score_interactions=False avoids the error, but disables the interaction scoring feature.

Environment

• CellPhoneDB version: 5.0.0
• Python version: 3.10
• OS: macOS

Additional Context

This bug was discovered while integrating CellPhoneDB into ChatSpatial, an MCP server for spatial transcriptomics analysis.