{
  "artifact_type": "empirical_real_fragment_coordinate_downsampling",
  "clinical_claim": false,
  "dataset": "CRAG liver matched fragment-coordinate cohort",
  "samples": 16,
  "matched_pairs": 8,
  "method": "Independent Bernoulli thinning of each real author-derived fragment, implemented exactly from disjoint per-bin short, long and other counts",
  "common_observed_autosomal_bins": 567,
  "results": [
    {
      "retained_fraction": 0.05,
      "replicates": 200,
      "median_retained_fragments": 1241863.5,
      "approx_aligned_depth_assuming_2x100bp_and_3.1Gb": 0.08012022580645162,
      "coverage_dispersion_auc_median": 0.9375,
      "coverage_dispersion_auc_95pct_range": [
        0.90625,
        0.96875
      ],
      "short_long_auc_median": 0.578125,
      "short_long_auc_95pct_range": [
        0.578125,
        0.578125
      ],
      "favorable_matched_pair_fraction_median": 1.0,
      "favorable_matched_pair_fraction_95pct_range": [
        0.875,
        1.0
      ],
      "median_within_sample_coverage_score_sd_across_replicates": 0.001506741194020201
    },
    {
      "retained_fraction": 0.1,
      "replicates": 200,
      "median_retained_fragments": 2484520.5,
      "approx_aligned_depth_assuming_2x100bp_and_3.1Gb": 0.16029164516129032,
      "coverage_dispersion_auc_median": 0.96875,
      "coverage_dispersion_auc_95pct_range": [
        0.9375,
        0.96875
      ],
      "short_long_auc_median": 0.578125,
      "short_long_auc_95pct_range": [
        0.578125,
        0.578125
      ],
      "favorable_matched_pair_fraction_median": 1.0,
      "favorable_matched_pair_fraction_95pct_range": [
        0.875,
        1.0
      ],
      "median_within_sample_coverage_score_sd_across_replicates": 0.001244907031004877
    },
    {
      "retained_fraction": 0.25,
      "replicates": 200,
      "median_retained_fragments": 6211870.5,
      "approx_aligned_depth_assuming_2x100bp_and_3.1Gb": 0.4007658387096774,
      "coverage_dispersion_auc_median": 0.96875,
      "coverage_dispersion_auc_95pct_range": [
        0.953125,
        0.96875
      ],
      "short_long_auc_median": 0.578125,
      "short_long_auc_95pct_range": [
        0.578125,
        0.578125
      ],
      "favorable_matched_pair_fraction_median": 1.0,
      "favorable_matched_pair_fraction_95pct_range": [
        1.0,
        1.0
      ],
      "median_within_sample_coverage_score_sd_across_replicates": 0.0009063176618424008
    },
    {
      "retained_fraction": 0.5,
      "replicates": 200,
      "median_retained_fragments": 12423014.0,
      "approx_aligned_depth_assuming_2x100bp_and_3.1Gb": 0.8014847741935484,
      "coverage_dispersion_auc_median": 0.96875,
      "coverage_dispersion_auc_95pct_range": [
        0.96875,
        0.96875
      ],
      "short_long_auc_median": 0.578125,
      "short_long_auc_95pct_range": [
        0.578125,
        0.578125
      ],
      "favorable_matched_pair_fraction_median": 1.0,
      "favorable_matched_pair_fraction_95pct_range": [
        1.0,
        1.0
      ],
      "median_within_sample_coverage_score_sd_across_replicates": 0.0007031857296871259
    },
    {
      "retained_fraction": 1.0,
      "replicates": 200,
      "median_retained_fragments": 24843109.5,
      "approx_aligned_depth_assuming_2x100bp_and_3.1Gb": 1.6027812580645162,
      "coverage_dispersion_auc_median": 0.96875,
      "coverage_dispersion_auc_95pct_range": [
        0.96875,
        0.96875
      ],
      "short_long_auc_median": 0.578125,
      "short_long_auc_95pct_range": [
        0.578125,
        0.578125
      ],
      "favorable_matched_pair_fraction_median": 1.0,
      "favorable_matched_pair_fraction_95pct_range": [
        1.0,
        1.0
      ],
      "median_within_sample_coverage_score_sd_across_replicates": 0.0
    }
  ],
  "interpretation": "More retained fragments reduce Monte Carlo measurement variation in the coverage-dispersion score. Discrimination can still plateau or vary; this does not establish an optimal clinical sequencing depth.",
  "limitations": [
    "Computational thinning of coordinate fragments is not an independently prepared library.",
    "The source library's extraction, PCR duplication and alignment cannot be undone.",
    "Only eight cancer/control pairs are present, so AUROC has very low resolution.",
    "Approximate depth assumes two sequenced 100 bp mates per retained fragment and a 3.1 Gb haploid genome.",
    "The coverage score uses neutral GC/mappability placeholders and is not a production CNA caller."
  ]
}