Autocycler trim

Basics

Long-read contigs sometimes contain excess sequence which needs to be trimmed off. This often occurs with circular sequence (most bacterial replicons) where overlap between the start and end of a contig. For small plasmids, this can sometimes involve the entire sequence being duplicated (or more) in a single contig. Linear sequences with hairpin ends can also create excess sequence which needs to be trimmed.

This step is performed on a per-cluster basis. For each contig, Autocycler looks for two types of excess sequence: start-end circular overlap and hairpin overlap. Contigs are trimmed as necessary. After trimming, Autocycler discards any contigs with outlier lengths, so the remaining contigs in the cluster will all be in close agreement with each other.

Usage

Usage: autocycler trim [OPTIONS] --cluster_dir <CLUSTER_DIR>

Options:
  -c, --cluster_dir <CLUSTER_DIR>    Autocycler cluster directory containing 1_untrimmed.gfa file
                                     (required)
      --min_identity <MIN_IDENTITY>  Minimum alignment identity for trimming alignments [default: 0.75]
      --max_unitigs <MAX_UNITIGS>    Maximum unitigs used for overlap alignment, set to 0 to disable
                                     trimming [default: 5000]
      --mad <MAD>                    Allowed variability in cluster length, measured in median absolute
                                     deviations, set to 0 to disable exclusion of length outliers
                                     [default: 5.0]
  -t, --threads <THREADS>            Number of CPU threads [default: 8]
  -h, --help                         Print help
  -V, --version                      Print version

Notes

• Uses a dynamic-programming alignment algorithm, but based on unitigs, not bases. This saves a lot of time.
• --max_unitigs 0 will turn off trimming, e.g. if you've manually trimmed the sequences yourself.
• Cannot distinguish between artefactual and genuine duplications. E.g. if a plasmid really is doubled, Autocycler trim will still cut it down to a single copy.

Toy example

Of all the input contigs in the toy example, only b1 and b2 contain overlap. These are their paths through the unitig graph, with the overlap highlighted and the overlap-free trimmed paths below:

Note that the alignment does not need to be exact. Contig b2 has a variant in its overlap (unitig 38 vs 34), but the alignment is still sufficiently high identity for trimming to occur.

After trimming is complete, the cluster graph is simplified and saved as 2_trimmed.gfa. The result is similar to the untrimmed graph but slightly simpler due to the removed pieces: