Track Provenance
5 properties
The processing/derivation operations applied to the data to produce the
file — what was done to it, as opposed to
Track Interpretation,
which records what the data is. Where
output_type names a result
kind (a noun — "peaks"), these slots name operations
applied: which read subset the data was computed from
(read_multiplicity), what
QC filtering was applied (filter_status),
what scaling normalization was applied
(normalization), what retention
cut / thresholding was applied
(thresholding), and the
epistemic origin of the values — observed vs. predicted
(derivation).
read_multiplicity
The read mapping-multiplicity subset the data was computed from (all mapped reads vs uniquely-mapping reads only) — the read-selection operation applied. Factored out of compound DataType terms (e.g. "signal of all reads"). Absent means unspecified.
filter_status
The quality-control filtering operation applied (filtered vs unfiltered). Factored out of compound DataType terms (e.g. "filtered peaks", "unfiltered alignments"). Absent means unspecified.
normalization
The scaling normalization applied to a signal (raw vs depth-normalized vs percentage-normalized) — the scaling operation applied. Factored out of compound DataType terms (e.g. "raw signal", "normalized signal", "percentage normalized signal"). Does not cover statistic transforms (p-value, fold-change, z-score, enrichment) or smoothing, which are distinct output_types. Absent means unspecified (not raw).
thresholding
The retention cut / thresholding operation applied to produce the file, and its basis (idr vs mapping_quality vs significance). Factored out of compound DataType terms (e.g. "IDR thresholded peaks", "mapping quality thresholded contact matrix"). Cross-cuts peaks, contact matrices, and regulatory links. Absent means no threshold / unspecified.
derivation
The epistemic origin of the file's values (observed vs predicted) — whether they were empirically measured from experimental data or generated by a computational/statistical model. Factored out of compound DataType / FeatureType terms (e.g. "observed signal profile", "predicted signal profile", "predicted transcription start sites"). `control` is excluded: it denotes an experimental role or normalization reference, not the measured-vs-model axis. Absent means unspecified (not a default of observed).
read_multiplicity and
filter_status were relocated
here from Track Interpretation, while
normalization,
thresholding, and
derivation (observed vs. predicted)
were faceted out of the compound DataType / FeatureType terms: all describe a
processing operation, not content. Track Provenance remains the planned home for
further processing facets — bias-correction — when those are
faceted out of the compound DataType terms. That axis remains baked into the
output_type base for now.