umi_combine¶

Purpose¶

Aggregates read-level evidence into UMI-level counts across candidate loci.

Upstream¶

• mpileup

Required inputs¶

• in_filter_bam
• filter_mpileup_file
• db_path (chunk database from mpileup)
• sequence_type

Input interpretation¶

Input key	Source	Required	Interpretation
in_filter_bam	bam_processing output	Yes	Filtered BAM used to aggregate read evidence into UMI-level counts.
filter_mpileup_file	mpileup output	Yes	Candidate loci table that defines UMI aggregation targets.
db_path / chunk manifest	mpileup output	Yes	Chunk metadata controlling parallel chunk processing.
sequence_type	top-level config	Yes	Controls mode-specific handling during evidence aggregation.

Parameters¶

Most controls are currently internal defaults in the step implementation (for example chunk-level parallel workers and buffer sizes), while threads are inherited from run.threads.

Parameter interpretation¶

Parameter area	Interpretation
run.threads / runtime parallel settings	Controls chunk-level parallel processing throughput.
internal filtering flags	Duplicate/secondary/QC-fail/supplementary filtering behavior is defined in implementation defaults unless exposed in your version config.

Outputs¶

• spot_count_file: output_dir/umi_combine/spot.count.parquet
• error_count_file: output_dir/umi_combine/error.count.parquet

Tuning notes¶

• This step is parallelized by chunk files loaded from split_chunk.db.
• Outputs are parquet and used by both genotyping and RNA_feature.