{ "schema": "https://aptatlas.org/data/v4.6/combine_engine_spec.schema.json", "name": "combine_engine_spec", "title": "Help Me Combine engine specification", "version": "4.6", "last_updated": "2026-11-04", "licence": "CC-BY-4.0", "licence_url": "https://creativecommons.org/licenses/by/4.0/", "citation": "Whelan CD, Smith-Byrne K. Atlas of Proteomic Technologies: an evidence-based framework for selecting and combining commercial proteomics platforms. bioRxiv 2026. https://doi.org/10.64898/2026.09.10.750723", "archive_doi": "10.5281/zenodo.22697191", "third_party": [ { "name": "ChEMBL", "licence": "CC-BY-SA-3.0", "url": "https://www.ebi.ac.uk/chembl/", "relationship": "derived", "note": "The listed fields are derived from ChEMBL (CC BY-SA 3.0): they are counts and percentages computed against a ChEMBL drug-target list. The per-protein list itself is not published here.", "fields": [ "$.denominators.drug_targets", "$.platforms.*.coverage.drug_targets", "$.pairs[*].coverage_pct", "$.pairs[*].coverage_ci_lo", "$.pairs[*].coverage_ci_hi", "$.pairs[*].net_new_targets", "$.pairs[*].net_new_pct", "$.pairs[*].class_balance" ] } ], "data": { "engine_version": "v3.7", "generated_from": "src/engine/combine.js + src/data/helpMeCombine.json (Atlas of Proteomic Technologies)", "source_locations": { "priorities_derive_axes_quality_dims": "src/engine/combine.js: PRIORITIES, derive(), AXES, QUALITY_DIMS, DEFAULT_QWEIGHTS", "quality_function": "src/engine/combine.js: customQuality()", "scoring_function": "src/engine/combine.js: scorePairs()", "pair_coverage": "src/engine/combine.js: pairCoverage(), pairNetNew(), cov()", "data": "src/data/helpMeCombine.json" }, "axes": [ { "key": "coverage", "default_weight": 1, "invert": false }, { "key": "complementarity", "default_weight": 1, "invert": false }, { "key": "quality", "default_weight": 1, "invert": false }, { "key": "cost", "default_weight": 1, "invert": true }, { "key": "class_balance", "default_weight": 0.5, "invert": false }, { "key": "sample_volume", "default_weight": 0, "invert": true }, { "key": "ms_fit", "default_weight": 0, "invert": false } ], "priority_to_lever": { "coverage": { "axis": "coverage", "set_weight_to": 5 }, "complementarity": { "axis": "complementarity", "set_weight_to": 5 }, "precision": { "quality_dim": "precision" }, "specificity": { "quality_dim": "specificity" }, "sensitivity": { "quality_dim": "sensitivity" }, "quantification": { "quality_dim": "quantification_type" }, "pqtl": { "quality_dim": "pqtl_accuracy" }, "throughput": { "quality_dim": "throughput" }, "cost": { "axis": "cost", "set_weight_to": 5 }, "volume": { "axis": "sample_volume", "set_weight_to": 5 }, "tissue": { "axis": "ms_fit", "set_weight_to": 5 } }, "priority_combination_rule": "Start from the default axis weights above. Each selected priority overrides its lever: an axis priority sets that axis weight to 5; a quality-dimension priority is collected. If one or more quality-dimension priorities are selected, the quality axis weight is set to 5 and the quality sub-dimension weights are set to 1 for exactly the selected dimensions and 0 for all others. If no quality-dimension priority is selected, the quality sub-dimension weights take their defaults (intrinsic 1, maturity 0). Weights are never summed across priorities or renormalised outside the composite weighted mean.", "quality_dims": [ { "key": "precision", "group": "intrinsic", "default_weight": 1 }, { "key": "specificity", "group": "intrinsic", "default_weight": 1 }, { "key": "sensitivity", "group": "intrinsic", "default_weight": 1 }, { "key": "quantification_type", "group": "intrinsic", "default_weight": 1 }, { "key": "pqtl_accuracy", "group": "intrinsic", "default_weight": 1 }, { "key": "throughput", "group": "maturity", "default_weight": 0 }, { "key": "evidence_depth", "group": "maturity", "default_weight": 0 }, { "key": "sample_flexibility", "group": "maturity", "default_weight": 0 } ], "quality_formula": "quality(pair) = ( sum over sub-dims d with weight w_d > 0 of w_d * (app_scores[a][d] + app_scores[b][d]) / 2 ) / ( sum of w_d ). This is the AVERAGE of the two platforms on each property (not best-of-pair). If no sub-dimension has weight > 0, quality = 0.", "raw_axis_values": "coverage = pair coverage in the chosen denominator; complementarity = net-new percent in the chosen denominator; quality = quality(pair) above; cost = total_cost; class_balance = class_balance; sample_volume = joint_volume_uL; ms_fit = 1 if the pair contains a mass-spectrometry platform (Seer Proteograph XT or Biognosys TrueDiscovery) else 0.", "pair_coverage_rule": "EXACT for all four numeric denominators as of v2.9 (P13: the independence-model estimate has been removed). drug_targets and fda_biomarkers are exact set intersections against their external reference lists (1,449 ChEMBL drug targets; 216 FDA-approved biomarkers, Bhowmick et al. 2021). proteome_canonical and proteome_atlas_union are exact unions of each platform's own accession-level list, computed directly from public/proteins.json.", "class_balance_rule": "Normalised Shannon entropy of the target classes among the drug targets the pair covers: nine ChEMBL/UniProt classes (target_class_9; GPCRs within membrane receptors, nuclear receptors within transcription factors), Unclassified excluded, divided by ln(9). Derived from the ChEMBL drug-target list. Rounded to 3 decimals.", "normalisation": "Per axis, min-max across the fifteen pairs to [0,1]: n = (x - min)/(max - min); for inverted axes (cost, sample_volume) n = (max - x)/(max - min). If max == min the range is treated as 1 so the axis contributes 0.", "composite_formula": "composite(pair) = ( sum over axes of weight[axis] * n[axis] ) / ( sum over axes of weight[axis] ). Winner of a scenario = the pair with the highest composite (discovery mode, all fifteen pairs eligible).", "denominators": { "drug_targets": 1449, "proteome_atlas_union": 14340, "proteome_canonical": 20190, "fda_biomarkers": 216 }, "pathway_coverage_disclosure": "The UI's optional Reactome pathway selector shows each pair's precomputed pathway coverage (source: reactome_pathways.json, pathway_coverage.json) as a descriptive figure alongside composite, with a view-only toggle to re-order the ranked list by it. It is not an eighth axis (not in the seven-entry axes array above), not a twelfth priority (not in priority_to_lever above), and not a fifth coverage denominator (not in denominators above): it enters no formula in this spec and does not affect composite, so it is outside the priority_sweep's input space and requires no change to any scored quantity here.", "priority_sweep": "All combinations of 1, 2, or 3 of the eleven priorities (11 + 55 + 165 = 231), evaluated over the four numeric denominators = 924 scenarios. Fully deterministic; no RNG.", "bootstrap_ci": "Per pair, 2,000-resample bootstrap over the drug-target universe (n=1,449): draws ~ Binomial(1449, coverage_pct/100) / 1449 via numpy.random.default_rng(seed=42), 2.5th/97.5th percentiles. Computed once per pair, in pair order, immediately after the platform/pair data is loaded and before the sweeps run (order affects the RNG stream and must match for exact reproduction). FDA-biomarker intervals (coverage_fda_ci_lo, coverage_fda_ci_hi) use the same method over n=216 with coverage_pct_fda, on a separate generator with the same seed, so they do not alter the drug-target stream.", "platforms": { "nomic-omni": { "display_name": "Nomic Omni", "ptm_proteoform_capable": false, "coverage": { "drug_targets": 24, "proteome_canonical": 5.2, "proteome_atlas_union": 7.4, "fda_biomarkers": 41.7 }, "app_scores": { "proteome_coverage": 2, "precision": 4, "specificity": 4, "sensitivity": 3, "cost_efficiency": 5, "throughput": 5, "quantification_type": 4, "sample_flexibility": 2, "evidence_depth": 1, "pqtl_accuracy": 3 } }, "olink-explore-ht": { "display_name": "Olink Explore HT", "ptm_proteoform_capable": false, "coverage": { "drug_targets": 40.4, "proteome_canonical": 26.9, "proteome_atlas_union": 37.8, "fda_biomarkers": 58.8 }, "app_scores": { "proteome_coverage": 4, "precision": 2, "specificity": 4, "sensitivity": 4, "cost_efficiency": 3, "throughput": 5, "quantification_type": 2, "sample_flexibility": 3, "evidence_depth": 5, "pqtl_accuracy": 3 } }, "illumina-protein-prep": { "display_name": "SomaSeq", "ptm_proteoform_capable": false, "coverage": { "drug_targets": 65.3, "proteome_canonical": 46.2, "proteome_atlas_union": 65, "fda_biomarkers": 90.3 }, "app_scores": { "proteome_coverage": 5, "precision": 5, "specificity": 2, "sensitivity": 4, "cost_efficiency": 3, "throughput": 4, "quantification_type": 2, "sample_flexibility": 3, "evidence_depth": 5, "pqtl_accuracy": 2 } }, "nulisa": { "display_name": "Alamar NULISA", "ptm_proteoform_capable": false, "coverage": { "drug_targets": 10.4, "proteome_canonical": 1.9, "proteome_atlas_union": 2.7, "fda_biomarkers": 14.4 }, "app_scores": { "proteome_coverage": 1, "precision": 5, "specificity": 4, "sensitivity": 5, "cost_efficiency": 4, "throughput": 3, "quantification_type": 2, "sample_flexibility": 3, "evidence_depth": 3, "pqtl_accuracy": 3 } }, "seer-proteograph": { "display_name": "Seer Proteograph XT", "ptm_proteoform_capable": true, "coverage": { "drug_targets": 57.8, "proteome_canonical": 49.9, "proteome_atlas_union": 70.2, "fda_biomarkers": 88.4 }, "app_scores": { "proteome_coverage": 5, "precision": 3, "specificity": 4, "sensitivity": 3, "cost_efficiency": 3, "throughput": 3, "quantification_type": 3, "sample_flexibility": 3, "evidence_depth": 4, "pqtl_accuracy": 5 } }, "biognosys-truediscovery": { "display_name": "Biognosys TrueDiscovery", "ptm_proteoform_capable": true, "coverage": { "drug_targets": 33.7, "proteome_canonical": 17.7, "proteome_atlas_union": 24.9, "fda_biomarkers": 73.1 }, "app_scores": { "proteome_coverage": 4, "precision": 3, "specificity": 4, "sensitivity": 3, "cost_efficiency": 3, "throughput": 3, "quantification_type": 3, "sample_flexibility": 5, "evidence_depth": 3, "pqtl_accuracy": 3 } } }, "display_name_to_slug": { "Nomic Omni": "nomic-omni", "Olink Explore HT": "olink-explore-ht", "SomaSeq": "illumina-protein-prep", "Alamar NULISA": "nulisa", "Seer Proteograph XT": "seer-proteograph", "Biognosys TrueDiscovery": "biognosys-truediscovery" }, "pairs": [ { "a": "Nomic Omni", "b": "Olink Explore HT", "coverage_pct": 44.3, "coverage_ci_lo": 41.7, "coverage_ci_hi": 47, "coverage_pct_canonical": 27.9, "coverage_pct_atlas_union": 39.2, "coverage_pct_fda": 68.1, "coverage_fda_ci_lo": 62, "coverage_fda_ci_hi": 74.1, "net_new_pct": 3.86, "net_new_targets": 56, "total_cost": 3.5, "class_balance": 0.875, "joint_volume_uL": 12 }, { "a": "Nomic Omni", "b": "SomaSeq", "coverage_pct": 66.5, "coverage_ci_lo": 64.2, "coverage_ci_hi": 68.9, "coverage_pct_canonical": 46.4, "coverage_pct_atlas_union": 65.4, "coverage_pct_fda": 92.1, "coverage_fda_ci_lo": 88.4, "coverage_fda_ci_hi": 95.4, "net_new_pct": 1.17, "net_new_targets": 17, "total_cost": 4, "class_balance": 0.911, "joint_volume_uL": 65 }, { "a": "Nomic Omni", "b": "Alamar NULISA", "coverage_pct": 25.3, "coverage_ci_lo": 23, "coverage_ci_hi": 27.4, "coverage_pct_canonical": 5.7, "coverage_pct_atlas_union": 8, "coverage_pct_fda": 45.4, "coverage_fda_ci_lo": 38.9, "coverage_fda_ci_hi": 52.3, "net_new_pct": 1.24, "net_new_targets": 18, "total_cost": 3, "class_balance": 0.765, "joint_volume_uL": 35 }, { "a": "Nomic Omni", "b": "Seer Proteograph XT", "coverage_pct": 64.5, "coverage_ci_lo": 62.1, "coverage_ci_hi": 67, "coverage_pct_canonical": 51.2, "coverage_pct_atlas_union": 72, "coverage_pct_fda": 91.7, "coverage_fda_ci_lo": 88, "coverage_fda_ci_hi": 95.4, "net_new_pct": 6.76, "net_new_targets": 98, "total_cost": 4, "class_balance": 0.918, "joint_volume_uL": 255 }, { "a": "Nomic Omni", "b": "Biognosys TrueDiscovery", "coverage_pct": 43.8, "coverage_ci_lo": 41.3, "coverage_ci_hi": 46.3, "coverage_pct_canonical": 19.9, "coverage_pct_atlas_union": 28.1, "coverage_pct_fda": 79.2, "coverage_fda_ci_lo": 73.6, "coverage_fda_ci_hi": 84.3, "net_new_pct": 10.08, "net_new_targets": 146, "total_cost": 4, "class_balance": 0.85, "joint_volume_uL": 50 }, { "a": "Olink Explore HT", "b": "SomaSeq", "coverage_pct": 70.5, "coverage_ci_lo": 68, "coverage_ci_hi": 72.8, "coverage_pct_canonical": 54.9, "coverage_pct_atlas_union": 77.4, "coverage_pct_fda": 94.9, "coverage_fda_ci_lo": 91.7, "coverage_fda_ci_hi": 97.7, "net_new_pct": 5.24, "net_new_targets": 76, "total_cost": 5.5, "class_balance": 0.918, "joint_volume_uL": 57 }, { "a": "Olink Explore HT", "b": "Alamar NULISA", "coverage_pct": 41.3, "coverage_ci_lo": 38.6, "coverage_ci_hi": 43.8, "coverage_pct_canonical": 27.1, "coverage_pct_atlas_union": 38.1, "coverage_pct_fda": 62, "coverage_fda_ci_lo": 55.6, "coverage_fda_ci_hi": 68.1, "net_new_pct": 0.83, "net_new_targets": 12, "total_cost": 4.5, "class_balance": 0.867, "joint_volume_uL": 27 }, { "a": "Olink Explore HT", "b": "Seer Proteograph XT", "coverage_pct": 70, "coverage_ci_lo": 67.8, "coverage_ci_hi": 72.3, "coverage_pct_canonical": 59.6, "coverage_pct_atlas_union": 83.8, "coverage_pct_fda": 94.4, "coverage_fda_ci_lo": 91.2, "coverage_fda_ci_hi": 97.2, "net_new_pct": 12.22, "net_new_targets": 177, "total_cost": 5.5, "class_balance": 0.923, "joint_volume_uL": 247 }, { "a": "Olink Explore HT", "b": "Biognosys TrueDiscovery", "coverage_pct": 53.7, "coverage_ci_lo": 51.2, "coverage_ci_hi": 56.2, "coverage_pct_canonical": 36.2, "coverage_pct_atlas_union": 51, "coverage_pct_fda": 84.7, "coverage_fda_ci_lo": 79.6, "coverage_fda_ci_hi": 89.4, "net_new_pct": 13.25, "net_new_targets": 192, "total_cost": 5.5, "class_balance": 0.888, "joint_volume_uL": 42 }, { "a": "SomaSeq", "b": "Alamar NULISA", "coverage_pct": 65.8, "coverage_ci_lo": 63.4, "coverage_ci_hi": 68.3, "coverage_pct_canonical": 46.3, "coverage_pct_atlas_union": 65.2, "coverage_pct_fda": 91.7, "coverage_fda_ci_lo": 88, "coverage_fda_ci_hi": 94.9, "net_new_pct": 0.48, "net_new_targets": 7, "total_cost": 5, "class_balance": 0.91, "joint_volume_uL": 80 }, { "a": "SomaSeq", "b": "Seer Proteograph XT", "coverage_pct": 78.1, "coverage_ci_lo": 75.9, "coverage_ci_hi": 80.2, "coverage_pct_canonical": 66.3, "coverage_pct_atlas_union": 93.3, "coverage_pct_fda": 98.1, "coverage_fda_ci_lo": 95.8, "coverage_fda_ci_hi": 99.5, "net_new_pct": 12.77, "net_new_targets": 185, "total_cost": 6, "class_balance": 0.927, "joint_volume_uL": 300 }, { "a": "SomaSeq", "b": "Biognosys TrueDiscovery", "coverage_pct": 69.4, "coverage_ci_lo": 67, "coverage_ci_hi": 71.8, "coverage_pct_canonical": 50.1, "coverage_pct_atlas_union": 70.5, "coverage_pct_fda": 96.3, "coverage_fda_ci_lo": 93.5, "coverage_fda_ci_hi": 98.6, "net_new_pct": 4.07, "net_new_targets": 59, "total_cost": 6, "class_balance": 0.913, "joint_volume_uL": 95 }, { "a": "Alamar NULISA", "b": "Seer Proteograph XT", "coverage_pct": 61.4, "coverage_ci_lo": 58.9, "coverage_ci_hi": 63.9, "coverage_pct_canonical": 50.4, "coverage_pct_atlas_union": 71, "coverage_pct_fda": 90.3, "coverage_fda_ci_lo": 86.1, "coverage_fda_ci_hi": 94, "net_new_pct": 3.66, "net_new_targets": 53, "total_cost": 5, "class_balance": 0.919, "joint_volume_uL": 270 }, { "a": "Alamar NULISA", "b": "Biognosys TrueDiscovery", "coverage_pct": 39.3, "coverage_ci_lo": 37, "coverage_ci_hi": 41.8, "coverage_pct_canonical": 18.7, "coverage_pct_atlas_union": 26.3, "coverage_pct_fda": 77.8, "coverage_fda_ci_lo": 72.2, "coverage_fda_ci_hi": 83.3, "net_new_pct": 5.52, "net_new_targets": 80, "total_cost": 5, "class_balance": 0.844, "joint_volume_uL": 65 }, { "a": "Seer Proteograph XT", "b": "Biognosys TrueDiscovery", "coverage_pct": 60.6, "coverage_ci_lo": 58, "coverage_ci_hi": 63.1, "coverage_pct_canonical": 51.1, "coverage_pct_atlas_union": 72, "coverage_pct_fda": 90.7, "coverage_fda_ci_lo": 86.6, "coverage_fda_ci_hi": 94.4, "net_new_pct": 2.83, "net_new_targets": 41, "total_cost": 6, "class_balance": 0.919, "joint_volume_uL": 285 } ], "coverage_basis": "distinct reviewed human proteins (canonical accessions), v4.2" } }