Methods and reproducibility
Methods and Reproducibility
Last scientific review: 2026-10-12.
This page summarizes archived computational work. The handoff involved curation, source checking and figure preparation only; it did not rerun prediction models or begin new experiments. Supporting claims: C044scope
Reference identity and transcript reconstruction
NCBI RefSeq, genomic reference and transcript/exon mappings were reconciled before interpretation. The primary reference is NM_206933.4 / NP_996816.3 on GRCh38; NM_007123.6 / NP_009054.6 is retained as the distinct classical short reference. Genomic introns use 0-based half-open intervals; retained exonic-base coordinates and transcript/CDS positions use 1-based numbering. 1 2 Supporting claims: C001sequence factC004sequence factC011predictionC016conditional
The proposed transcript carrying a 50-nt deletion was reconstructed by replacing the exon-21 donor while keeping other long-reference exons and junctions unchanged. cDNA deletion and independent genomic-exon concatenation agreed. Translation from the unchanged start verified the conditional frame and stop; it did not establish RNA or protein production. Supporting claims: C016conditionalC017conditionalC019unresolved
RNA models
SpliceAI 1.3.1 used the mean of five original models. Pangolin used three v2 models per tissue across heart, liver, brain and testis; the executed classification heads did not separately label donor and acceptor. Model outputs retain their own scales and are not averaged across tools. 6 7 Supporting claims: C010prediction
AlphaGenome used client 0.10.0, API v1 and the saved ALL_FOLDS selection for 131,072-bp and 1,048,576-bp inputs. No immutable server checkpoint identifier was recorded. Raw REF/ALT junction, site and usage outputs were archived separately from official variant-scorer aggregates and quantiles. The eye and brain track metadata, exact intervals, values and coordinate maps are retained in the independent junction-review archive. 8 Supporting claims: C011predictionC012predictionC014predictionC044scope
Local protein models and controls
ColabFold 1.6.3 / AlphaFold2 pTM produced twelve unrelaxed 346-residue predictions: parameter sets 1–3, seeds 17 and 18, three recycles, templates and dropout off. Homolog MSA rows were matched, and comparisons used model/seed identity rather than output rank. RMSD and pLDDT remain structural descriptors, not functional assays. 9 10 Supporting claims: C021predictionC022predictionC023prediction
Rosetta Cartesian relaxation/ddG calculations used matched WT/mutant states, geometry checks, identity/reverse controls and a spatial-neighbor-flexibility sensitivity. The primary strained results were excluded from native-effect magnitude claims. ThermoMPNN predictions were retained separately on their learned scale, including inconsistent reverse results. 11 12 Supporting claims: C024predictionC025predictionC026prediction
OpenMM 8.6.1 simulated capped LG1, residues 1517–1709, using Amber ff14SB/TIP3P, 300 K, 1 bar, 0.15 M salt and a 2-fs step. C1672–C1709 disulfide connectivity was an annotation-based modeling assumption, not experimental confirmation. The final primary simulations used matched His1535 HID assignments and zero production backbone restraints; an earlier 5-ns tautomer sensitivity was retained separately. Glycans and full-length domain context were absent. 13 Supporting claims: C027prediction
Three primary WT/G1526R pairs reached 20 ns each. Comparisons retained replica variability and used a common 10–20 ns window. Equilibrium convergence, native folding free energy and cellular function were not established. Supporting claims: C027predictionC028predictionC029unresolved
Reviewable artifacts
The evidence registry records source-repository commit, relative path and SHA-256 for each supporting artifact. The claims registry links public statements to evidence category, source, method, limitation and review status. The methods registry records versions and parameters.
The junction-review archive contains original raw arrays, metadata, exact reconstruction, checksums and an offline verifier. Smaller evidence tables accompany the figures. Full structure/trajectory archives remain preserved in the research repository; no public download URL has been assigned to those larger archives. Source-file hashes support later reconciliation without asserting that an unpublished archive is already publicly hosted.
Reproducibility checks concern computation and provenance. They are not experimental validation, independent peer review or evidence of therapeutic efficacy. Supporting claims: C033unresolvedC044scope
Methods registry
Versions, parameters and the limitations stated at the time of the archived computation. Values are recorded as they were used; nothing here was re-run during content preparation.
| Method | Version / build | Recorded parameters | Stated limitation |
|---|---|---|---|
| M01 Reference and sequence reconciliation | NCBI RefSeq NM_206933.4/NP_996816.3 and NM_007123.6/NP_009054.6; GRCh38; archived Ensembl/GENCODE mappings | genomic_variant: chr1:216175303 C>T · gene_strand: - · cds_variant: c.4576G>A · cds_start_cdna_1based: 440 | Sequence identity does not establish biological fate. |
| M02 SpliceAI | 1.3.1 | models: 5 · aggregation: mean · input_bp: 30001 · output_positions: 20001 · output_orientation: genomic_forward | Site-level inference; not a complete junction or RNA quantity. |
| M03 Pangolin | source 5cf94b8db938c658391b4305cd7ce33297d44ff7; v2 weights | models_per_tissue: 3 · tissues: ["heart","liver","brain","testis"] · head: P(splice) classification; no donor/acceptor distinction · input_bp: 30001 | Tissue predictions are not independent biological experiments. |
| M04 AlphaGenome | client 0.10.0; API v1; ALL_FOLDS selection; immutable server checkpoint unrecorded | contexts: [{"length":131072,"start_0based":216109766,"end_0based_exclusive":216240838},{"length":1048576,"start_0based":215498223,"end_0based_exclusive":216546799}] · chromosome: chr1 · ontology_terms: ["UBERON:0000019","UBERON:0000955"] · tracks: ["embryonic camera-type eye / total RNA-seq","adult brain / polyA plus RNA-seq"] · modalities: ["splice_sites","splice_site_usage","splice_junctions","rna_seq"] · official_scorers: Separate archived recommended variant scorers; no ontology filter supplied to score_variant · raw_transformations: none | Predicted junction outputs are not measured RNA abundance or complete transcript probabilities. |
| M05 Conditional transcript reconstruction | independent exon concatenation and cDNA deletion; archived sequence records | reference: NM_206933.4 · deletion_cds_1based_inclusive: [4578,4627] · deletion_cdna_1based_inclusive: [5017,5066] · deleted_nt: 50 · all_other_junctions: assumed annotated · product_aa_if_translated: 1527 · stop_codon_1based: 1528 | Mature RNA and protein production are not observed. |
| M06 Matched ColabFold / AlphaFold2 pTM ensemble | ColabFold 1.6.3 | construct_residues: [1364,1709] · model_parameter_sets: [1,2,3] · seeds: [17,18] · recycles: 3 · early_stop_tolerance: 0 · msa: 256:512 · templates: false · dropout: false · relaxation: false · pairing: model and seed, not rank | Local predicted fold; no native function or full-length domain placement established. |
| M07 Rosetta Cartesian stability and controls | 2025.51; commit 612b6ef9e9d4638e6f1e47fabee02587d9453733 | construct_residues: [1517,1709] · relax_decoys_per_input: 3 · ddg_rounds: 5 · score_function: ref2015_cart · controls: ["identity G10G","reverse R10G from independently prepared Arg models","flex_bb sensitivity in two inputs"] · primary_geometry_failures: 26/30 mutant structures · units: REU; no kcal/mol conversion | Primary effect magnitudes invalid; limited sampling and protocol dependence persist. |
| M08 ThermoMPNN | commit 370f76ec62bd929f7425e311d8df04a0d094990f; archived default weights | construct_lengths: [193,346] · forward_and_reverse_inputs: six WT and six independently predicted Arg models per context · weights: strict loading; no retraining · units: predicted kcal/mol on training scale | Forward/reverse disagreement and native-context transfer uncertainty. |
| M09 OpenMM exploratory MD | OpenMM 8.6.1; MDTraj 1.11.1.post2; PDBFixer 1.12.0 | construct_residues: [1517,1709] · terminal_caps: ACE/NME · force_field: Amber ff14SB · water: TIP3P · temperature_K: 300 · pressure_bar: 1 · salt_M: 0.15 · step_fs: 2 · primary_replicas: 6 · production_ns_each: 20 · preparation_sensitivity_ns: 5 · total_ns: 125 · analysis_window_ns: [10,20] · primary_H1535: HID · production_backbone_restraints: 0 · glycans_modeled: false · disulfide_assumption: C1672–C1709; by-similarity annotation · C1528: neutral free thiol | Folded isolated basin; no equilibrium/free-energy, localization or degradation conclusion. |
| M10 Evidence interpretation | reference annotation and construct-specific publication review | scope: Separate sequence, annotation, prediction, conditional consequence, experiment and unknowns · new_cohorts: none | Negative literature statements are limited to the reviewed evidence, not proof of universal absence. |
| M11 Scientific content curation | this handoff; 2026-10-12 | new_predictions: false · new_wet_lab_experiments: false · website_implementation: false · public_case: USH2A G1526R only | Curation does not constitute independent expert peer review or clinical validation. |
Evidence registry
Each evidence entry groups the artifacts supporting a set of claims. Internal source paths stay in the research repository; published files are listed by name with their integrity hashes.
| Evidence | Type | Published files |
|---|---|---|
| USH2A-E001 Variant and reference-sequence identity | sequence validation | reference_sequences.fasta, transcript_mapping.csv |
| USH2A-E002 Classical isoform and exon architecture | reference annotation | domain_feature_evidence.json, exon_map.csv |
| USH2A-E003 USH2A gene/background annotation | reference annotation | Supporting artifacts remain in the research repository. |
| USH2A-E004 General cytoplasmic interaction experiments | published experiment not G1526R | Supporting artifacts remain in the research repository. |
| USH2A-E005 Three-model splice-site comparison | computational prediction | cross_model_site_comparison.csv |
| USH2A-E006 AlphaGenome raw junction evidence package | archived prediction and reconstruction | alphagenome_junction_comparison.csv, alphagenome_splice_site_comparison.csv |
| USH2A-E007 Exact hypothetical RNA reconstruction | conditional sequence consequence | codon_comparison.csv, cryptic_partner_ranking.csv, deletion_sequence_evidence.json, reconstructed_consequences.csv |
| USH2A-E008 Matched local protein-structure ensemble | computational prediction | original_structure_model_metrics.csv, structure_baseline_variability.csv, wt_mutant_comparison.csv |
| USH2A-E009 Rosetta geometry qualification and controls | computational quality control | rosetta_geometry_qc.csv, stability_results.csv |
| USH2A-E010 ThermoMPNN forward/reverse predictions | computational prediction | thermompnn_results.csv |
| USH2A-E011 Exploratory MD and preparation audit | computational prediction | md_physical_qc.csv, md_preparation_audit.csv, md_replica_summary.csv, md_wt_mutant_comparison.csv |
| USH2A-E012 Integrated hypotheses and evidence gaps | interpretation with explicit uncertainty | Supporting artifacts remain in the research repository. |
| USH2A-E013 Discriminating experimental proposals | research design not completed experiment | Supporting artifacts remain in the research repository. |
| USH2A-E014 Public research scope specified by the project initiator | project brief not biological evidence | Supporting artifacts remain in the research repository. |
| USH2A-E015 Recorded computation and provenance | execution and integrity records | Supporting artifacts remain in the research repository. |
Evidence snapshots for download
Files below are byte-identical selections from the archived study. SHA-256 values are computed from the published bytes and compared with the evidence registry during the build.
| File | Bytes | SHA-256 | Registry match |
|---|---|---|---|
| alphagenome_junction_comparison.csv | 214,274 | 0ca7c308d4cee0c41acfbb3d3b98635e961c755660810f3742a4a082a076d67c | exact |
| alphagenome_splice_site_comparison.csv | 28,270 | c29c2c092859ecdbf0b1051aee6cc4328a99ef4d230716552f464674564dd01d | exact |
| codon_comparison.csv | 13,008 | 639acd9cd81078e425f08f865a0be7cc21dfb72222807a1e69b4fd836f1fdb05 | exact |
| cross_model_site_comparison.csv | 2,745 | abc02040bf83a158f5f9c419f4278a98d2d7e341891d823bb0a2a22ce1a258be | exact |
| cryptic_partner_ranking.csv | 51,835 | 4373685c4a2d0e48ee71be41511c6826b5926cf92ec13ebc2a243fa8a4f9de8b | exact |
| deletion_sequence_evidence.json | 875 | 6d698a265c9ff811c81dc42d5d1f39f2b56ebabb57afd99031d82c6598e9587c | exact |
| domain_feature_evidence.json | 2,574 | 9af0acb0c5e3d0fa68164ca035538acc3eb33cfb8d07c219de24dd420639aa75 | exact |
| exon_map.csv | 14,064 | 665882681c86ee3d2a9507431cd365710400d7438983a5beec9a3370aa64e79d | exact |
| md_physical_qc.csv | 2,135 | 76f444785d057248f244ea22a41f19f2eb3fcd700df9f1305af83be45214a912 | exact |
| md_preparation_audit.csv | 2,712 | 634d3aeb9783cb7b3fd407a5eab01bca6a513af69b0dea9468a63eea1f6e20a0 | exact |
| md_replica_summary.csv | 16,197 | 7f2975aea0dfcf641600c6984f69bef0b39cf89cef440df2a458338f945440b2 | exact |
| md_wt_mutant_comparison.csv | 3,789 | 9529a95fc97d3d0661db4f9440fab239bb940f2fa0cbccf48fa1e164342832ea | exact |
| original_structure_model_metrics.csv | 3,785 | 22747e176442f6a98e052673f50d5cac5f29ca190776fbb07b55f1db3d6fc41f | exact |
| reconstructed_consequences.csv | 5,064 | a977de4a24893bac444e331221edcd717e0a7228c965a5b18f332aa4c0e9b337 | exact |
| reference_sequences.fasta | 126,478 | 11148d40a32d7971403988a317d25dd892fb4124473688ee824b11cca3c8e2b8 | exact |
| rosetta_geometry_qc.csv | 51,876 | ab7eac95a44fbb40264227e696ce3ddb4426943668a67625548b571acfec0ac1 | exact |
| stability_results.csv | 11,935 | 1ef1b7998550431132e2808bd8e899c3703273937cb0ceec13c296411c57c8f7 | exact |
| structure_baseline_variability.csv | 2,635 | d39559d38ea5f53117fb17ff510c47719a4ca86833c433d20192ab83c6bf6681 | exact |
| thermompnn_results.csv | 10,359 | 3fd81fed55b1a1d43f7ef60c57fbee44ab8c93ef2a74d30d0021ae5ab97237c2 | exact |
| transcript_mapping.csv | 760 | c75528f9c54ce6787d783d0d6648dfceaf3cf9630b9b1779a6a02264dcb23e73 | exact |
| wt_mutant_comparison.csv | 916 | 202e3087f537996720f4191b268c1ddf117de4608de0524ae5daa7e94b92e0a1 | exact |
Interpretation boundaries for these files
- Model predictions and official scorer aggregates are different outputs.
- A missing value is not a zero, and no score is a measured mature-RNA fraction.
- Reconstructed RNA/protein products are hypothetical unless separately demonstrated.
- Failed Rosetta geometry is retained as method evidence, not a native-energy result.
- MD replica statistics describe limited sampling, not equilibrium or biological replicates.