map_utils.py

​

High-level description

This code defines a function map_intbcs that processes a molecule table to resolve allele ambiguity for each cellBC/intBC pair. It identifies the most frequent allele based on read count and UMI count, then filters out alignments with other alleles for each pair.

​

Symbols

​

map_intbcs

​

Description

This function takes a molecule table as input and returns a modified allele table where each cellBC/intBC pair is associated with a single allele. It achieves this by grouping the input table by cellBC, intBC, and allele, then selecting the allele with the highest read count (and UMI count as a tie-breaker) for each group. Alignments with other alleles for the same cellBC/intBC pair are removed.

​

Inputs

​

Outputs

​

Internal Logic

1. Drops rows with missing intBC values from the input molecule_table.
2. Groups the remaining rows by cellBC, intBC, and allele.
3. Aggregates the groups by summing readCount and counting UMI.
4. Sorts the aggregated table first by UMI in descending order, then by readCount in descending order.
5. Identifies duplicate cellBC/intBC pairs, keeping only the first occurrence (which corresponds to the highest read and UMI count).
6. Creates a set of tuples representing the selected cellBC, intBC, and allele combinations.
7. Filters the original molecule_table to keep only rows where the cellBC, intBC, and allele combination exists in the set created in step 6.
8. Logs the number of removed alleles and UMIs during the filtering process.
9. Returns the filtered molecule_table.

​

References

• logger: Used for logging debug messages. Imported from cassiopeia.mixins.
• utilities.log_molecule_table: A decorator function used to log statistics of the molecule table before and after the function execution. Imported from cassiopeia.preprocess.

​

Side Effects

• Logs debug messages indicating the number of alleles and UMIs removed during the filtering process.

​

Performance Considerations

The function uses pandas groupby and apply operations, which can be computationally expensive for large datasets. Consider optimizing these operations or using alternative data structures if performance becomes a bottleneck.