Given the merged per-contig hits from the WF1 screen (all contigs of one sample BLASTed against the bundled metazoan mitogenome database), pick the reference the sample as a whole matches best and keep the contigs that match it convincingly.
Usage
select_mito_contigs(
hits,
min_identity = 70,
min_aligned_length = 300,
min_aligned_fraction = 0.5,
max_candidates = 20,
max_references = 5
)Arguments
- hits
data frame of BLAST hits with columns `qseqid`, `saccver`, `pident`, `length` (aligned bases), `bitscore` and `qlen`.
- min_identity
Percent identity required against the winning reference (default = 70)
- min_aligned_length
Aligned bases required (default = 300)
- min_aligned_fraction
Fraction of the contig the alignment must cover (default = 0.5)
- max_candidates
Most contigs carried into confirmation, counted across all references (default = 20)
- max_references
Most references the vote may award, a safety net on very messy assemblies (default = 5)
Value
a list with `accession` (the winning references in the order they won, NA when there are no hits), `candidates` (character vector of contig names, best first) and `evidence` (one row per contig scored, with the numbers behind the call and a `reason` for anything dropped)
Details
The vote runs in rounds. Each round scores only the contigs that hit that round's winning reference, then drops all of them from the pool, pass or fail, so every contig is judged against exactly one reference. Rounds keep the anti-splitting behavior within a mitogenome while letting a second, unrelated mitogenome (a contaminant, or a mixed sample) win its own round. The search stops at the first round that selects nothing: references are visited strongest first, so a reference carrying only NUMT-grade hits means the weaker ones behind it carry less.
The fraction rule is the NUMT filter: a real mitochondrial contig is almost entirely mitochondrial, while a nuclear scaffold carrying a NUMT aligns over a tiny slice of itself.