read_feature¶

Purpose¶

Computes read-level quality and bias features for each candidate mutation.

Upstream¶

• genotyping

Required inputs¶

• bam_file
• ind_geno_filter_mutation_list

Input interpretation¶

Input key	Source	Required	Interpretation
bam_file	input_details.bam_file	Yes	Read-level evidence source for mismatch/mapQ/baseQ/query-position features.
ind_geno_filter_mutation_list	genotyping output	Yes	Candidate locus list used to define per-region read feature extraction targets.

Parameters¶

From steps.read_feature:

• cell_info (optional barcode-to-cell mapping)
• downsample
• downsample_target_depth
• max_region_size
• max_variants_per_region
• seed

Parameter interpretation highlights¶

Parameter	Interpretation
cell_info	Optional barcode-to-cell map for cell-aware feature derivation.
downsample, downsample_target_depth	Controls depth normalization for robust comparisons across loci.
max_region_size, max_variants_per_region	Region partition controls for balancing runtime and memory.
seed	Reproducibility for stochastic operations (for example downsampling).

Outputs¶

• read_feature: output_dir/read_feature/read_feature.txt
• parquet mirror: read_feature.parquet

Tuning notes¶

• Regions are batched and processed in multiprocessing workers.
• Feature extraction uses mutation-local read information and produces one row per candidate.