pirl-unc / tcrsift / 25705970188

Build # 25705970188

Build Type

push

github

Committed by web-flow

Commit Message

Method × method overlap matrix + method-recovery panel (closes #27, v0.8.0) (#28)

* Method × method overlap matrix + method-recovery panel (closes #27)

Implements both chunks of #27 (chunks 3 and 4 of #20). Surfaces in the
CSV outputs and the report PDF the method-stratified views that users
were previously reconstructing post-hoc from clonotypes.csv.

clonotype.build_method_overlap_matrices(filtered, long_df, similarity)
  -> dict[donor, DataFrame]

  Per-donor square method × method overlap matrices over filter-passing
  clones. Off-diagonal cells are pairwise similarity of the
  selected-clone sets in each method; diagonal is the per-method clone
  count (when similarity='count') or 1.0 (when 'jaccard'/'dice').

  Three similarity options:
    - jaccard (default): |A ∩ B| / |A ∪ B|
    - dice:               2|A ∩ B| / (|A| + |B|)
    - count:              raw |A ∩ B| (diagonal = |A|)

  Symmetric. Returns {} when method axis isn't on long_df or filtered
  is empty. When donor isn't on long_df but method is, folds under a
  synthetic donor='all' so output naming stays uniform.

clonotype.build_method_recovery_table(filtered, long_df, tier)
  -> long DataFrame [donor, method, recovered, total, fraction]

  Per-(donor, method) recovery of clones at the target tier.
  - When `tier` matches a tier label in filtered: count clones with
    that tier that appear in each (donor, method) bucket.
  - When `tier='*'` or filtered has no tier column: fall back to all
    filtered clones (lets non-FDR filter modes use the panel too).

plots.plot_method_overlap(matrix, output_path, similarity, donor)
  Annotated seaborn heatmap, fmt depending on metric.

plots.plot_method_recovery(recovery, output_path, tier_label)
  Bar plot of recovery fraction per method, paired by donor; annotated
  with "recovered/total" labels above each bar; methods sorted by
  cross-donor mean recovery.

cmd_run wires both into Step 4 alongside the per-method rankings:

  data/method_ove... (continued)

Coverage Stats

4407 of 6215 relevant lines covered (70.91%)

2.84 hits per line

Source File

Press 'n' to go to next uncovered line, 'b' for previous

Source Not Available