Key Parameters
The most important parameters and switches. For the full list, see the CLI reference.
Choice of the Aligner
--fpa switches from the default A* aligner to the faster but less capable four-point aligner; --costs sets the cost model. See Aligners and Cost model.
Padding (--padding)
The amount of reference sequence to extract around the cluster for realignment context. More padding gives better alignment results but costs more computation.
- Default: 200
- More padding = better results, slower
- Common values: 100–500 bp
Memory Limit (--memory-limit) and Number of Threads (--threads)
Resource allocations for the aligner. This does not apply to the IO and buffering overhead. The memory will be divided among all threads. Each thread will perform one alignment at a time.
- Default: scales to available memory on the system
- Less memory = more alignments will fail due to OOM, faster
- Examples:
40gb,2gb,500mb
Timeout (--aligner-timeout)
Both aligners respect an optional per-alignment timeout (e.g. 30s, 5m). Alignments that exceed this limit are reported as soft failures and the original variant calls are kept.
Clustering
--cluster-strategy, --cluster-max-gap, --cluster-min-records and --cluster-min-mass decide which mutations are realigned together. See Clustering parameters.
Phasing (for VCF data)
--phasing-bam enables read-based phasing of ambiguous clusters. See Phasing parameters.
Output Filtering (for VCF data)
By default, all records are output. You can filter to focus on interesting results:
--only-realigned: only records that were successfully realigned--realigned-and-context N: realigned records plus N bases of surrounding context