diagFDR 0.1.2

Diagnostics

• dfdr_ipe(): decoy-reference PEP concordance (internal PEP calibration error, IPE) with the target-decoy comparator D_b/T_b (mean target PEP per bin, weights n_b = T_b + D_b, bins with mean target PEP <= 0.5 and n_b >= 200). The headline IPE uses it; IPE_pooled gives the same summary with the pooled comparator D_b/(T_b + D_b).
• dfdr_headline() returns the decoy-support index support_index (D^-1/2) and the FDP-CV benchmark CV_benchmark (sqrt(2) * D^-1/2).
• flag_headline(): review alerts, not a verdict. Flag values OK, CAUTION*, ALERT_*, NE_* (not evaluable) and NA; status NO_ALERT, CAUTION or ALERT. IPE alerts above 0.05. The equal-chance flag is NE_two_pass on two-pass exports and CAUTION_imbalance_pass_unknown on a DIA-NN list whose pass status was not given.
• dfdr_alert_basis(): evidential basis of each alert (derived check, simulation-informed alert, heuristic alert, descriptive).
• dfdr_p_summary() / dfdr_p_interpretation(): pi0 by the eight cp4p estimators on targets + decoys, targets and decoys, grouped into a converging set (consensus value) and the others; p-value threshold, and score or PEP threshold, giving an estimated FDR of alpha on targets with the consensus pi0 and with pi0 = 1 (Benjamini-Hochberg); comparison with target-decoy counting; pooled targets + decoys check; caveats.
• dfdr_pep_summary() / dfdr_pep_interpretation(): whether PEPs agree with, under- or overestimate the decoy-derived reference, and the sum of the PEPs of the accepted targets against the number of accepted decoys.

Inputs

• as_dfdr_tbl() and the DIA-NN builders take two_pass (set TRUE for --reanalyse / match-between-runs reports); the DIA-NN builders say once per session when it is not given.
• dfdr_run_all(compute_pseudo_pvalues = TRUE) warns, and records the reason in the list metadata (p_source), when a list has no score column (DIA-NN 2.x reports carry no CScore). pseudo_pvalue_score = "pep" uses -PEP as an approximate surrogate score.
• mokapot_competed_universe(q_mode = c("as_exported", "recount")): mokapot computes q-values over all PSMs it receives; "recount" recomputes (D+1)/T q-values on the competed winners.
• DIA-NN builders keep the largest score of a collapsed unit (higher is better), and the smallest q-value and PEP.

HTML report

• Executive summary with alerts to review, each tagged with its evidential basis; lists without alert shown as “no internal alert”; checks that are not evaluable shown as “n/e”.
• Every table is followed by a legend of its columns; small values are shown in scientific notation; formulas are written in HTML (no MathJax needed).
• “What the p-value diagnostics say” and a PEP interpretation, written from the summaries above.
• The recomputed decoy ratio is labelled D/T; the Jaccard guide treats overlap as descriptive.
• dfdr_render_report() accepts a relative out_dir and renders from a temporary copy of the template.
• A plot that cannot be built for a list (e.g. too few p-values) is skipped with a warning.