{ "date": "2026-10-13", "headline": "Alignment improves the comparison; overall superiority remains unestablished.", "status": "Retrospective development only \u00b7 research checkpoints, not serving defaults", "models": [ { "report": { "model": "m4", "version": "m4-aligned-response-v3", "evaluation": "Retrospective study-held-out development only", "freshHoldout": false, "superiorityEstablished": false, "folds": [ { "heldStudy": "NMDRHT", "trainingStudy": "Tani2012", "trainingGenes": 7687, "heldGenes": 10990, "overlappingGenes": 7400, "selected": { "aligned_v3": { "configurationId": 7, "configuration": { "objective": "regression_l1", "num_leaves": 7, "reg_lambda": 100 }, "innerValidationSpearman": 0.5013894542249832, "refitSeconds": 0.23768475000000056 }, "ExtraTrees": { "configurationId": 6, "configuration": { "min_samples_leaf": 10, "max_features": 0.33 }, "innerValidationSpearman": 0.49031060771063295, "refitSeconds": 0.6308780840000008 }, "ridge": { "configurationId": 4, "configuration": { "alpha": 100.0 }, "innerValidationSpearman": 0.4559025040967242, "refitSeconds": 0.02955270800000065 } }, "metrics": { "aligned_v3": { "spearman": 0.4335844659177479, "mae": 0.5878342773871215, "r2": -0.19522628649454532 }, "ExtraTrees": { "spearman": 0.460358527758317, "mae": 0.5627660246653088, "r2": -0.08756873699188161 }, "ridge": { "spearman": 0.4474211485032731, "mae": 0.5655987582260991, "r2": -0.10327427061122418 }, "training_median": { "spearman": null, "mae": 0.6192952740562929, "r2": -0.32025102005434203 } }, "comparisons": [ { "baseline": "ExtraTrees", "deltaSpearman": -0.026774061840569074, "adjustedPairedCi": [ -0.0355059363859636, -0.0192666652188042 ], "positiveDevelopmentInterval": false }, { "baseline": "ridge", "deltaSpearman": -0.013836682585525173, "adjustedPairedCi": [ -0.023726793758993118, -0.003127364129797424 ], "positiveDevelopmentInterval": false } ], "medianMinusCandidateMaeCi95": [ 0.026590269374316994, 0.03684174842146735 ] }, { "heldStudy": "Tani2012", "trainingStudy": "NMDRHT", "trainingGenes": 10990, "heldGenes": 7687, "overlappingGenes": 7400, "selected": { "aligned_v3": { "configurationId": 5, "configuration": { "objective": "regression", "num_leaves": 31, "reg_lambda": 100 }, "innerValidationSpearman": 0.53242974004939, "refitSeconds": 0.6209211669999988 }, "ExtraTrees": { "configurationId": 5, "configuration": { "min_samples_leaf": 5, "max_features": 1.0 }, "innerValidationSpearman": 0.5227826557211279, "refitSeconds": 2.131525833000005 }, "ridge": { "configurationId": 3, "configuration": { "alpha": 10.0 }, "innerValidationSpearman": 0.487373841739424, "refitSeconds": 0.03840629199999768 } }, "metrics": { "aligned_v3": { "spearman": 0.4421142867048064, "mae": 0.669704994173427, "r2": -0.06463337015493598 }, "ExtraTrees": { "spearman": 0.4803894812436631, "mae": 0.6519250968197047, "r2": -0.026556403605596968 }, "ridge": { "spearman": 0.403809064736498, "mae": 0.6911266931167692, "r2": -0.10344529827393822 }, "training_median": { "spearman": null, "mae": 0.6711025223525823, "r2": -0.11712206651913015 } }, "comparisons": [ { "baseline": "ExtraTrees", "deltaSpearman": -0.038275194538856694, "adjustedPairedCi": [ -0.04988419097258626, -0.026541710283269047 ], "positiveDevelopmentInterval": false }, { "baseline": "ridge", "deltaSpearman": 0.03830522196830838, "adjustedPairedCi": [ 0.02389852852683728, 0.05173833330609522 ], "positiveDevelopmentInterval": true } ], "medianMinusCandidateMaeCi95": [ -0.006664951768823415, 0.009422851449695648 ] } ], "finalResearchFit": { "aligned_v3": { "configuration": { "objective": "regression_l1", "num_leaves": 15, "reg_lambda": 100 }, "meanInnerValidationSpearman": 0.5140200008774387, "artifactSha256": "faf2967f7df5db99594924a8f12a491f63d9762a782b5e01c0c0a3b3efb44614", "artifactBytes": 139812 }, "ExtraTrees": { "configuration": { "min_samples_leaf": 10, "max_features": 0.67 }, "meanInnerValidationSpearman": 0.5058581637780604, "artifactSha256": "3eb97cd1d0994f2f916ebbdc1d0d8d5d40283dcfb23ce1746ff9dbf8e3db035b", "artifactBytes": 16865125 }, "ridge": { "configuration": { "alpha": 100.0 }, "meanInnerValidationSpearman": 0.47150297128543484, "artifactSha256": "216859702a4ebfc0cf0d1afd61c8b2b8eb99505e070574c1849409c9ce4ca5bb", "artifactBytes": 3368 } }, "runtime": [ { "family": "aligned_v3", "batchSize": 1, "bundleLoadSeconds": 0.0019091249999974025, "warmMedianMs": 2.4400209999981826, "warmP95Ms": 2.6791061000039917, "repeats": 30 }, { "family": "aligned_v3", "batchSize": 32, "bundleLoadSeconds": 0.0019091249999974025, "warmMedianMs": 2.4006874999997763, "warmP95Ms": 2.5721643999993655, "repeats": 30 }, { "family": "ExtraTrees", "batchSize": 1, "bundleLoadSeconds": 0.07984408300000467, "warmMedianMs": 27.983562500001113, "warmP95Ms": 28.582096349999375, "repeats": 30 }, { "family": "ExtraTrees", "batchSize": 32, "bundleLoadSeconds": 0.07984408300000467, "warmMedianMs": 27.99731300000019, "warmP95Ms": 28.869254199998906, "repeats": 30 }, { "family": "ridge", "batchSize": 1, "bundleLoadSeconds": 0.0006670829999961825, "warmMedianMs": 2.3696664999981465, "warmP95Ms": 2.629676750004961, "repeats": 30 }, { "family": "ridge", "batchSize": 32, "bundleLoadSeconds": 0.0006670829999961825, "warmMedianMs": 2.3474165000010316, "warmP95Ms": 2.6158706500002173, "repeats": 30 } ], "runtimeScope": "Precomputed feature-table validation, imputation and inference; excludes sequence extraction, network and response-target estimation. One bundle-load measurement is not cold end-to-end latency.", "features": [ "nexon__mean", "nexon__max", "nexon__std", "tx_len__mean", "tx_len__max", "tx_len__std", "cds_len__mean", "cds_len__max", "cds_len__std", "utr5_len__mean", "utr5_len__max", "utr5_len__std", "utr3_len__mean", "utr3_len__max", "utr3_len__std", "stop_to_lastEJ__mean", "stop_to_lastEJ__max", "stop_to_lastEJ__std", "downstream_junctions__mean", "downstream_junctions__max", "downstream_junctions__std", "stop_upstream_50nt__mean", "stop_upstream_50nt__max", "stop_upstream_50nt__std", "max_exon_len__mean", "max_exon_len__max", "max_exon_len__std", "mean_exon_len__mean", "mean_exon_len__max", "mean_exon_len__std", "last_exon_len__mean", "last_exon_len__max", "last_exon_len__std", "cds_fraction__mean", "cds_fraction__max", "cds_fraction__std", "GC_tx__mean", "GC_tx__max", "GC_tx__std", "GC_utr5__mean", "GC_utr5__max", "GC_utr5__std", "GC_cds__mean", "GC_cds__max", "GC_cds__std", "GC_utr3__mean", "GC_utr3__max", "GC_utr3__std", "utr3_mfe_nt__mean", "utr3_mfe_nt__max", "utr3_mfe_nt__std", "isoform_count" ], "trainingRows": 18677, "trainingGenes": 11277, "machine": "macOS-27.0.1-arm64-arm-64bit", "threadLimit": 4, "limitations": [ "Native HCT116 transcript geometry summaries reused for both studies; not assay-specific isoform composition.", "SLAM versus BRIC estimation, cell line and perturbation differences persist.", "Same-time point estimate cancels time baseline, but no covariance-aware uncertainty or raw-read refitting was performed.", "All source outcomes were previously used or belong to an already-used study." ] }, "alignment": { "inputGenes": 14578, "labelRows": 19744, "missingTargets": 0, "rowsByStudy": { "NMDRHT": 10990, "Tani2012": 7687 }, "features": [ "nexon__mean", "nexon__max", "nexon__std", "tx_len__mean", "tx_len__max", "tx_len__std", "cds_len__mean", "cds_len__max", "cds_len__std", "utr5_len__mean", "utr5_len__max", "utr5_len__std", "utr3_len__mean", "utr3_len__max", "utr3_len__std", "stop_to_lastEJ__mean", "stop_to_lastEJ__max", "stop_to_lastEJ__std", "downstream_junctions__mean", "downstream_junctions__max", "downstream_junctions__std", "stop_upstream_50nt__mean", "stop_upstream_50nt__max", "stop_upstream_50nt__std", "max_exon_len__mean", "max_exon_len__max", "max_exon_len__std", "mean_exon_len__mean", "mean_exon_len__max", "mean_exon_len__std", "last_exon_len__mean", "last_exon_len__max", "last_exon_len__std", "cds_fraction__mean", "cds_fraction__max", "cds_fraction__std", "GC_tx__mean", "GC_tx__max", "GC_tx__std", "GC_utr5__mean", "GC_utr5__max", "GC_utr5__std", "GC_cds__mean", "GC_cds__max", "GC_cds__std", "GC_utr3__mean", "GC_utr3__max", "GC_utr3__std", "utr3_mfe_nt__mean", "utr3_mfe_nt__max", "utr3_mfe_nt__std", "isoform_count" ], "target": "log2(kdeg depleted / control)", "NMDRHT": "Gene-level Nter_12h versus Ctrl_0h minus gene-level Ctrl_12h versus Ctrl_0h; shared reference cancels in the point estimate. No subtraction of p-values or standard errors.", "Tani2012": "Existing exact unique gene-symbol mapping and positive finite half-lives; log2(control half-life / depleted half-life).", "limitations": [ "Native HCT116 transcript geometry summaries reused for both studies; not assay-specific isoform composition.", "SLAM versus BRIC estimation, cell line and perturbation differences persist.", "Same-time point estimate cancels time baseline, but no covariance-aware uncertainty or raw-read refitting was performed.", "All source outcomes were previously used or belong to an already-used study." ] } }, { "report": { "model": "m5", "version": "m5-aligned-response-v3-unique-gene", "evaluation": "Retrospective study-held-out development only", "freshHoldout": false, "superiorityEstablished": false, "folds": [ { "heldStudy": "Aliouat2019", "trainingStudy": "NMDRHT", "trainingGenes": 4886, "heldGenes": 5268, "overlappingGenes": 4628, "selected": { "aligned_v3": { "configurationId": 3, "configuration": { "objective": "regression", "num_leaves": 15, "reg_lambda": 100 }, "innerValidationSpearman": 0.39060245436872715, "refitSeconds": 0.4062323750000001 }, "ExtraTrees": { "configurationId": 11, "configuration": { "min_samples_leaf": 20, "max_features": 1.0 }, "innerValidationSpearman": 0.38146417015778716, "refitSeconds": 0.9601016249999983 }, "ridge": { "configurationId": 5, "configuration": { "alpha": 1000.0 }, "innerValidationSpearman": 0.3101909283209963, "refitSeconds": 0.03565383400000144 } }, "metrics": { "aligned_v3": { "spearman": -0.22324970058358445, "mae": 0.4162220341902543, "r2": -0.5125058950446244 }, "ExtraTrees": { "spearman": -0.22457119377350954, "mae": 0.397041721506036, "r2": -0.39332799163656285 }, "ridge": { "spearman": -0.18132987167786624, "mae": 0.3829945280504677, "r2": -0.32786521895898124 }, "training_median": { "spearman": null, "mae": 0.3322701935771819, "r2": -0.06066183658803492 } }, "comparisons": [ { "baseline": "ExtraTrees", "deltaSpearman": 0.0013214931899250826, "adjustedPairedCi": [ -0.013219411545348675, 0.014575565281474728 ], "positiveDevelopmentInterval": false }, { "baseline": "ridge", "deltaSpearman": -0.041919828905718215, "adjustedPairedCi": [ -0.06861976353214111, -0.016660210476622476 ], "positiveDevelopmentInterval": false } ], "medianMinusCandidateMaeCi95": [ -0.08870814578492349, -0.07920677685256998 ] }, { "heldStudy": "NMDRHT", "trainingStudy": "Aliouat2019", "trainingGenes": 5268, "heldGenes": 4886, "overlappingGenes": 4628, "selected": { "aligned_v3": { "configurationId": 7, "configuration": { "objective": "regression_l1", "num_leaves": 7, "reg_lambda": 100 }, "innerValidationSpearman": 0.25526878492156335, "refitSeconds": 0.256466249999999 }, "ExtraTrees": { "configurationId": 11, "configuration": { "min_samples_leaf": 20, "max_features": 1.0 }, "innerValidationSpearman": 0.237193116387699, "refitSeconds": 1.0717916659999958 }, "ridge": { "configurationId": 5, "configuration": { "alpha": 1000.0 }, "innerValidationSpearman": 0.18847582101087143, "refitSeconds": 0.0364466249999964 } }, "metrics": { "aligned_v3": { "spearman": -0.3239511717926555, "mae": 0.4814249878156927, "r2": -0.18106044525186404 }, "ExtraTrees": { "spearman": -0.29411083012195716, "mae": 0.4801350096990822, "r2": -0.1758065116633376 }, "ridge": { "spearman": -0.2687791969669815, "mae": 0.47074092915764515, "r2": -0.146660634266651 }, "training_median": { "spearman": null, "mae": 0.4484788185378922, "r2": -0.06096868568711877 } }, "comparisons": [ { "baseline": "ExtraTrees", "deltaSpearman": -0.029840341670698323, "adjustedPairedCi": [ -0.048919625059811137, -0.009012976492301145 ], "positiveDevelopmentInterval": false }, { "baseline": "ridge", "deltaSpearman": -0.055171974825673986, "adjustedPairedCi": [ -0.07861225272018352, -0.029545447305432203 ], "positiveDevelopmentInterval": false } ], "medianMinusCandidateMaeCi95": [ -0.03535947472680609, -0.0304771572252909 ] } ], "finalResearchFit": { "aligned_v3": { "configuration": { "objective": "regression_l1", "num_leaves": 7, "reg_lambda": 100 }, "meanInnerValidationSpearman": 0.31853720140881625, "artifactSha256": "ab779b15b50375ab8dbf222d6d6e3173de4024f598f671bf29a1b8dcb8033054", "artifactBytes": 69651 }, "ExtraTrees": { "configuration": { "min_samples_leaf": 20, "max_features": 1.0 }, "meanInnerValidationSpearman": 0.30932864327274306, "artifactSha256": "54ae26009acb4cec984eed913c79e09e8d3d8510224177f639c589b7d56a6131", "artifactBytes": 4929834 }, "ridge": { "configuration": { "alpha": 1000.0 }, "meanInnerValidationSpearman": 0.24933337466593386, "artifactSha256": "6c55b1833c67dca587d80f553ee901617498f18a0dd977a68a271f4c26405c3e", "artifactBytes": 5434 } }, "runtime": [ { "family": "aligned_v3", "batchSize": 1, "bundleLoadSeconds": 0.00148200000000287, "warmMedianMs": 3.7453539999994234, "warmP95Ms": 4.004454499997222, "repeats": 30 }, { "family": "aligned_v3", "batchSize": 32, "bundleLoadSeconds": 0.00148200000000287, "warmMedianMs": 3.790500000000918, "warmP95Ms": 4.529866599998298, "repeats": 30 }, { "family": "ExtraTrees", "batchSize": 1, "bundleLoadSeconds": 0.0387482499999976, "warmMedianMs": 29.240417000000463, "warmP95Ms": 30.395949850001713, "repeats": 30 }, { "family": "ExtraTrees", "batchSize": 32, "bundleLoadSeconds": 0.0387482499999976, "warmMedianMs": 29.296228999996288, "warmP95Ms": 30.132825000001162, "repeats": 30 }, { "family": "ridge", "batchSize": 1, "bundleLoadSeconds": 0.0006935830000003307, "warmMedianMs": 3.7776244999996322, "warmP95Ms": 4.779768749999034, "repeats": 30 }, { "family": "ridge", "batchSize": 32, "bundleLoadSeconds": 0.0006935830000003307, "warmMedianMs": 3.726103999998287, "warmP95Ms": 4.123624699997919, "repeats": 30 } ], "runtimeScope": "Precomputed feature-table validation, imputation and inference; excludes sequence extraction, network and response-target estimation. One bundle-load measurement is not cold end-to-end latency.", "features": [ "n_uorfs_aug", "n_uorfs_near", "max_uorf_len", "dist_cap_to_first_uorf", "first_uorf_is_aug", "uorf_overlaps_main_orf", "min_dist_stop_to_cds", "kozak_max", "kozak_first", "utr_len", "utr_gc", "utr_dg", "kozak_x_dist", "n_uorfs_x_utrlen", "gc_x_dg", "main_kozak", "cds_len", "utr3_len", "dg_per_nt", "utr_kmer_A", "utr_kmer_C", "utr_kmer_G", "utr_kmer_T", "utr_kmer_AA", "utr_kmer_AC", "utr_kmer_AG", "utr_kmer_AT", "utr_kmer_CA", "utr_kmer_CC", "utr_kmer_CG", "utr_kmer_CT", "utr_kmer_GA", "utr_kmer_GC", "utr_kmer_GG", "utr_kmer_GT", "utr_kmer_TA", "utr_kmer_TC", "utr_kmer_TG", "utr_kmer_TT", "utr_kmer_AAA", "utr_kmer_AAC", "utr_kmer_AAG", "utr_kmer_AAT", "utr_kmer_ACA", "utr_kmer_ACC", "utr_kmer_ACG", "utr_kmer_ACT", "utr_kmer_AGA", "utr_kmer_AGC", "utr_kmer_AGG", "utr_kmer_AGT", "utr_kmer_ATA", "utr_kmer_ATC", "utr_kmer_ATG", "utr_kmer_ATT", "utr_kmer_CAA", "utr_kmer_CAC", "utr_kmer_CAG", "utr_kmer_CAT", "utr_kmer_CCA", "utr_kmer_CCC", "utr_kmer_CCG", "utr_kmer_CCT", "utr_kmer_CGA", "utr_kmer_CGC", "utr_kmer_CGG", "utr_kmer_CGT", "utr_kmer_CTA", "utr_kmer_CTC", "utr_kmer_CTG", "utr_kmer_CTT", "utr_kmer_GAA", "utr_kmer_GAC", "utr_kmer_GAG", "utr_kmer_GAT", "utr_kmer_GCA", "utr_kmer_GCC", "utr_kmer_GCG", "utr_kmer_GCT", "utr_kmer_GGA", "utr_kmer_GGC", "utr_kmer_GGG", "utr_kmer_GGT", "utr_kmer_GTA", "utr_kmer_GTC", "utr_kmer_GTG", "utr_kmer_GTT", "utr_kmer_TAA", "utr_kmer_TAC", "utr_kmer_TAG", "utr_kmer_TAT", "utr_kmer_TCA", "utr_kmer_TCC", "utr_kmer_TCG", "utr_kmer_TCT", "utr_kmer_TGA", "utr_kmer_TGC", "utr_kmer_TGG", "utr_kmer_TGT", "utr_kmer_TTA", "utr_kmer_TTC", "utr_kmer_TTG", "utr_kmer_TTT" ], "trainingRows": 10154, "trainingGenes": 5526, "machine": "macOS-27.0.1-arm64-arm-64bit", "threadLimit": 4, "limitations": [ "Count quantification and transcript annotation remain source-specific; Aliouat stable transcript IDs map to GENCODE42 then exclude every gene with multiple mapped transcript rows; no effective-length offsets.", "NMDRHT acute 12h versus 0h and Aliouat 72h shRNA/control differ; no time-matched untreated NMDRHT RPF arm is established.", "Representative transcript features are not condition-specific. Existing feature bank is biased toward prior NMDRHT response availability.", "Joint negative-binomial estimator shares dispersion assumptions across assays and treats libraries as unpaired; published Aliouat pairing is not modeled; NMDRHT cross-assay sample pairing has not been verified.", "No source significance or response-magnitude filtering." ] }, "alignment": { "rowsByStudy": { "Aliouat2019": 5268, "NMDRHT": 4886 }, "features": [ "n_uorfs_aug", "n_uorfs_near", "max_uorf_len", "dist_cap_to_first_uorf", "first_uorf_is_aug", "uorf_overlaps_main_orf", "min_dist_stop_to_cds", "kozak_max", "kozak_first", "utr_len", "utr_gc", "utr_dg", "kozak_x_dist", "n_uorfs_x_utrlen", "gc_x_dg", "main_kozak", "cds_len", "utr3_len", "dg_per_nt", "utr_kmer_A", "utr_kmer_C", "utr_kmer_G", "utr_kmer_T", "utr_kmer_AA", "utr_kmer_AC", "utr_kmer_AG", "utr_kmer_AT", "utr_kmer_CA", "utr_kmer_CC", "utr_kmer_CG", "utr_kmer_CT", "utr_kmer_GA", "utr_kmer_GC", "utr_kmer_GG", "utr_kmer_GT", "utr_kmer_TA", "utr_kmer_TC", "utr_kmer_TG", "utr_kmer_TT", "utr_kmer_AAA", "utr_kmer_AAC", "utr_kmer_AAG", "utr_kmer_AAT", "utr_kmer_ACA", "utr_kmer_ACC", "utr_kmer_ACG", "utr_kmer_ACT", "utr_kmer_AGA", "utr_kmer_AGC", "utr_kmer_AGG", "utr_kmer_AGT", "utr_kmer_ATA", "utr_kmer_ATC", "utr_kmer_ATG", "utr_kmer_ATT", "utr_kmer_CAA", "utr_kmer_CAC", "utr_kmer_CAG", "utr_kmer_CAT", "utr_kmer_CCA", "utr_kmer_CCC", "utr_kmer_CCG", "utr_kmer_CCT", "utr_kmer_CGA", "utr_kmer_CGC", "utr_kmer_CGG", "utr_kmer_CGT", "utr_kmer_CTA", "utr_kmer_CTC", "utr_kmer_CTG", "utr_kmer_CTT", "utr_kmer_GAA", "utr_kmer_GAC", "utr_kmer_GAG", "utr_kmer_GAT", "utr_kmer_GCA", "utr_kmer_GCC", "utr_kmer_GCG", "utr_kmer_GCT", "utr_kmer_GGA", "utr_kmer_GGC", "utr_kmer_GGG", "utr_kmer_GGT", "utr_kmer_GTA", "utr_kmer_GTC", "utr_kmer_GTG", "utr_kmer_GTT", "utr_kmer_TAA", "utr_kmer_TAC", "utr_kmer_TAG", "utr_kmer_TAT", "utr_kmer_TCA", "utr_kmer_TCC", "utr_kmer_TCG", "utr_kmer_TCT", "utr_kmer_TGA", "utr_kmer_TGC", "utr_kmer_TGG", "utr_kmer_TGT", "utr_kmer_TTA", "utr_kmer_TTC", "utr_kmer_TTG", "utr_kmer_TTT" ], "mapping": { "NMDRHT": { "estimatedTargets": 12869, "finiteTargets": 12869, "mappedRows": 4886 }, "Aliouat2019": { "estimatedTargets": 12256, "finiteTargets": 12256, "mappedRows": 5268 } }, "target": "Common DESeq2 assay-by-condition interaction log2FC, RPF change minus RNA change.", "limitations": [ "Count quantification and transcript annotation remain source-specific; Aliouat stable transcript IDs map to GENCODE42 then exclude every gene with multiple mapped transcript rows; no effective-length offsets.", "NMDRHT acute 12h versus 0h and Aliouat 72h shRNA/control differ; no time-matched untreated NMDRHT RPF arm is established.", "Representative transcript features are not condition-specific. Existing feature bank is biased toward prior NMDRHT response availability.", "Joint negative-binomial estimator shares dispersion assumptions across assays and treats libraries as unpaired; published Aliouat pairing is not modeled; NMDRHT cross-assay sample pairing has not been verified.", "No source significance or response-magnitude filtering." ] } } ], "targetAudit": { "NMDRHTNewVersusPublishedTE": { "genes": 5300, "spearman": 0.9999591251824214 }, "AliouatNewVersusPriorDerivedTE": { "genes": 3460, "spearman": -0.025048139243098237 }, "crossStudyAlignedTargets": { "genes": 4628, "spearman": -0.19095056980735883 }, "status": "Unresolved source/estimator disagreement; no M5 promotion", "interpretation": "Different estimators and gene/transcript inclusion can change results. Mapping ambiguity was removed before the final diagnostic rerun, but disagreement remains. No sign flip, sample relabeling, response rescaling, or cohort selection was used to improve transfer scores.", "countMappingCorrection": { "RNA": { "sourceTranscripts": 17105, "unmappedTranscripts": 537, "ambiguousGeneRowsExcluded": 85 }, "RPF": { "sourceTranscripts": 17105, "unmappedTranscripts": 537, "ambiguousGeneRowsExcluded": 85 } }, "limitations": [ "NMDRHT target ranks closely reproduce published estimates, providing a direction/estimator consistency check.", "Aliouat re-estimates disagree with prior derived contrasts; cannot attribute negative transfer solely to biology or model architecture.", "A joint unpaired estimator does not exploit the published Aliouat paired-library design.", "Current pooled M5 bundle is research-only, unsuitable for promotion pending reconciliation." ] }, "holdoutSearch": { "date": "2026-10-13", "qualifiedFreshHumanHoldout": null, "searchScope": "Targeted extension of the preceding qualification audit; not exhaustive. New candidate outcome files were not read or evaluated.", "candidates": [ { "name": "GSE303542", "status": "Excluded from primary UPF1 response holdout", "evidence": "Public GEO sample metadata shows human WT/eRF1 perturbation samples; UPF1 knockout samples belong to Neurospora crassa. This is not a matched human UPF1 response study.", "source": "https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE303542", "metadata": "GSE303542-metadata.soft.gz" }, { "name": "GSE304550 \u00b7 human motor neurons", "status": "Secondary proxy candidate; primary endpoint unqualified", "evidence": "Human UPF1/control RNA-seq exists, but the reviewed dataset measures abundance; reported REMBRANDTS analysis infers stability from RNA-seq rather than measuring a paired decay time course. No matched RPF response is established. It must not be substituted for direct decay or TE measurements.", "source": "https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE304550", "metadata": "GSE304550-metadata.soft.gz" }, { "name": "MDC UPF1 SLAM-seq / Ribo-seq records", "status": "Already-used study lineage", "evidence": "B\u00f6hm/Landthaler HCT116 AID-UPF1 records are NMDRHT source experiments, not another independent laboratory study.", "source": "https://edoc.mdc-berlin.de/id/eprint/25702/" } ], "nextRequirement": "An untouched human study with directly measured paired UPF1/control decay rates for M4, or replicated matched RNA/RPF conditions for M5; audit source lineage before freezing the final cohort." }, "protocol": { "createdUtc": "2026-10-13T13:11:19.690938+00:00", "status": "Development only; no untouched final study", "historicalStudyTransition": "NMDRHT, Tani2012/Imamachi2017 DRA000591 and Aliouat2019 GSE126385 are explicitly admitted as development sources for this revision. Historical test reports remain immutable. No results here constitute fresh independent confirmation.", "m4": "Direct gene-level decay targets, input-only gene geometry summaries. Align log2 depleted/control rate direction; use matched-time NMDRHT contrasts if available; record remaining assay and cell differences.", "m5": "Common gene-level replicate-aware RNA/RPF interaction estimator on both available count matrices; normalize within assay, use identical filtering and estimation rules; retain timing/perturbation and quantification differences.", "developmentEvaluation": "Two outer folds per model, hold out one entire study. On the other study use a deterministic gene-hash 80/20 inner split, identical across methods. Fit preprocessing only on inner training. Tune 12 configurations per family on inner validation Spearman; refit selected configuration on the outer training study, then predict the held-out study. These are retrospective development diagnostics, not fresh tests.", "families": [ "aligned_v3", "ExtraTrees", "ridge" ], "fitsPerFamilyPerOuterFold": 12, "finalFit": "Choose configuration by mean inner-validation Spearman across source studies, never by outer transfer score. Fit both development studies with equal total study weight. Preserve all older checkpoints; store new research-only bundles separately.", "metrics": [ "Spearman", "MAE", "R2", "training median MAE" ], "uncertainty": "Paired gene bootstrap within each outer study, 1000 resamples. Bonferroni adjustment across 8 candidate/baseline comparisons (two models, two studies, two strong baselines). Retrospective intervals cannot support confirmatory superiority.", "featureScope": "Reuse validated frozen input features. M4 gene mean/max/std and isoform count. M5 existing representative transcript feature bank is conditional on earlier label availability; do not describe it as unbiased whole-transcriptome coverage.", "runtime": "Measure model bundle loading and warm feature-table validation/imputation/prediction equally on local hardware, thread limit 4; 30 repeats, batches 1/32. This excludes source sequence/geometry extraction, HTTP and target-estimation cost; not full sequence-to-result runtime.", "holdout": "Keep new candidate response tables unopened while inspecting metadata. No new final evaluation without source/sample independence and endpoint qualification." }, "nextDecision": "M4: ExtraTrees is the strongest tested family in both transfer directions; retain it as the accuracy reference. M5: reconcile raw-count estimates with published contrasts and pairing metadata before further model tuning. Neither new candidate is promoted." }