The methods and their shared pipeline

Every analytical method in BOLD Forensics runs as a job: you choose records in a recordset, fill in a form in the Analysis panel, and the job runs in its own container away from the workbench. The methods differ in what they compute, but they share one pipeline, one set of record filters and one report layout. This page describes those shared parts so that each method page can concentrate on what is specific to it.

The five stages

Each job runs the same five stages in order. A stage that fails stops the job, and the job's error names the stage that failed.

The validate stage checks only parameters, so a job whose records are unsuitable (for example, unaligned sequences sent to a column-by-column distance method) passes validation and fails or misleads later. See Job lifecycle for how a job moves from submission to result.

Record filters common to all methods

The filter stage uses one shared filter program. Its rules apply before any method sees a record.

• Marker. Only records whose marker_code equals the chosen marker are kept. Identification and the sequence download apply no marker filter.
• Sequence length. Records outside the form's minimum and maximum are dropped. Most methods count only A, C, G and T; identification and diagnostic characters count every character except gaps. Barcode gap ignores the form and uses a fixed window (see its page).
• Ambiguity threshold. A record is dropped when more than 10% of its characters, after gaps are removed, are not A, C, G or T. This threshold is fixed and appears on no form.
• Records without a sequence. A record with no sequence counts as 100% ambiguous and is always dropped, including from the specimen download.
• Stop-codon and misidentification checkboxes. Every analysis form offers both. No method applies the misidentification filter. Only identification reads the stop-codon box, and its page explains why to leave it unticked.

The report shows the number of input records and the number that survived the filter, but not which records were dropped or why. If the filtered count is lower than you expect, the ambiguity threshold and the records-without-sequence rule are the usual causes.

Alignment

Distance summary, barcode gap, diagnostic characters and tree construction offer an "Align Sequences" choice. MAFFT runs MAFFT in automatic mode, which lets MAFFT pick its own strategy for the data. None passes the sequences on exactly as stored. The distance and diagnostic methods compare sequences column by column, so choose None only when the stored sequences are already aligned to a common length. Programs and versions are listed on Software versions.

What reports have in common

Every analysis job produces analysis_report.html and a PDF copy of it inside the result package. The report contains:

• the job ID and the recordset's title and description;
• the parameters as you submitted them, not as the job applied them;
• the input and filtered record counts;
• the time spent in each stage and the container's CPU and memory use;
• the method's own charts and tables;
• a list of output files.

Two catches matter for anyone citing a report. The parameter list is a copy of the form, so an option the job ignores still appears there as if applied; each method page says which options those are. The output file list is declared by the job type, not checked against the package, so a file can be listed and absent. The report does not record program versions, citations, the reference database build or random seeds. See Report anatomy for a section-by-section walk-through.

The methods

Which method should I use? helps you choose between them, and Running an analysis covers the form itself.

Good to know

• Distance summary and barcode gap compute every pair in Python, so run time grows with the square of the record count.
• No method records the program versions it ran with. Known limitations collects these points across all methods.

Related

• Running an analysis
• Analysis queue and results
• Job lifecycle
• Known limitations