Unified dataset across VarChAMP's 1%, Sahni et al., Rolland et al., Luck et al. and Lacoste et al.

Variant-level table that unifies the VarChAMP 1% assays (DUAL-IPA abundance, sqY2H protein-protein interactions, imaging) with four earlier studies: Sahni et al. 2015 (Y2H edgotypes and chaperone binding), Rolland et al. 2014 (Y2H), Luck et al. 2020 (Y2H) and Lacoste et al. 2024 (mislocalisation imaging). One row per variant, keyed by the genomic SPDI: 3,739 variants in 1,162 genes. Built by the varchamp_onepercent pipeline (analyses 2.05-mt-int-onepercent-variants and 2.06-mt-int-unified-table) from the VarChAMP allele-metadata table (doi:10.5281/zenodo.23780442), the sqY2H scores (version 2.0, doi:10.5281/zenodo.23924614) and the DUAL-IPA measurements (doi:10.5281/zenodo.15000710). Files File Content unified_slim.parquet, unified_slim.csv The unified table: 3,739 rows × 74 columns (identity, ClinVar, studies and sources, combined calls, 1% assays, study columns). Columns below unified_middle.parquet, unified_middle.csv Slim and 9 annotation columns: plddt, rsa, pLDDT_Category, RSA_Category, am_pathogenicity, inheritance_pattern, disease_modules, disease_module_count, g2p_Secondary structure (DSSP 3-state)* (83 columns) unified_full.parquet Slim and every other column of the public allele-metadata table (1,329 columns), described in that record 2.06-mt-int-unified-table.summary.json Provenance: pipeline commit, SHA-256 of every input, code file and output, parameters, counts variant_changes.csv Every variant only in version 0.1 or only in this version, and why (columns table, spdi, gene_variant, old_studies / new_studies, reason) data_licences.md, THIRD_PARTY_NOTICES.md Terms of the sources of the allele-metadata columns (from the allele-metadata record) Parquet keeps the column types; the CSV files are the same tables, comma-separated, with missing values empty. All three versions describe the variant on its representative clone. Construction Each 1% assay is attached to its clone in the allele-metadata table: DUAL-IPA (reporting cohort, 772 variants), sqY2H PPI (500) and imaging (607). Imaging clones listed without a mutation id are matched on the ORF and CCSB mutation id; 158 imaging variants measured only on clones without sequence confirmation are not in the allele table and are left out. The 1% assays are collapsed to one row per SPDI (1,208 variants) and outer-joined on SPDI with the study alleles matched to a clone (the allele-metadata record's published_study_scores_by_spdi). Study alleles without a clone have no SPDI and are not included (5 Sahni, 50 Lacoste). The representative clone is the 1% one when the variant was assayed in the 1%, else the study's matched clone (always the same ORF when both exist). Identity, ClinVar and annotation come from it. Met1 and predicted-splice variants (SpliceAI any delta score > 0.2) are left out of every source; the cDNA assays do not reproduce pre-mRNA processing. Synonymous variants are removed. Variants per study: 1% 1,206; Sahni 2,275; Lacoste 1,989; Rolland 106; Luck 21 (one variant can be in several). Identifiers and annotation Columns Content spdi Key: genomic SPDI (GRCh38); unique orf_id, mut_id, ccsb_mutation_id Representative clone: the key of the allele-metadata table symbol, gene_variant, aa_change, nt_change HGNC symbol, symbol_aa_change, protein change (3-letter code) and ORF nucleotide change of the representative clone allele_origin collection (VarChAMP allele collection) or lacoste (Lacoste et al. allele not in the collection) clinvar_clnsig_clean, clinvar_clnsig_clean_pp_strict ClinVar class (2026-10 release) from the allele-metadata table: Pathogenic, Benign, VUS, Conflicting, Others; the second keeps likely classes apart in_analysis_set The allele-metadata table's analysis-set flag Provenance Columns Content studies Studies with the variant, ;-separated: 1pct, Sahni, Rolland, Luck, Lacoste sources Datasets with a measurement, ;-separated: 1pct_DualIPA, 1pct_PPI, 1pct_Imaging, Sahni_PPI, Sahni_Chaperone, Rolland_PPI, Luck_PPI, Lacoste_Imaging n_sources, n_assays Number of sources; number of 1% assays Combined calls One call per variant across studies, taken from the first source in the priority order that has one. *_source names the source used; *_discordant is True when both sources have a call and they differ, and empty when only one has a call. Columns Priority Content edgotype_class, edgotype_source 1% > Sahni > Rolland > Luck Edgetic, Quasi-null, Quasi-wild-type, Perturbed-/Unperturbed-/Uncertain-degree-1 (one partner), Uncertain, Gain-of-interaction, Gain-and-loss-of-interaction. 993 variants (1% 500, Sahni 479, Luck 14). Lacoste's own PPI phenotype is not used abundance_hit_unified, abundance_hit_source, abundance_hit_discordant 1% > Sahni 1%: dualipa_hit. Sahni: chaperone binding (Y → True), a proxy for destabilisation, not an abundance measurement. 2,834 variants (731 True); 43 of the 145 measured in both differ imaging_hit_unified, imaging_hit_source, imaging_hit_discordant 1% > Lacoste 1%: img_pheno_hit. Lacoste: mislocalised (1 → True). 2,540 variants (318 True); 16 of the 56 measured in both differ VarChAMP 1% assays Assay Columns Clones dualipa_mut_id, ppi_mut_id, imaging_mut_id: the clone each assay measured (same ORF as the representative clone) DUAL-IPA abundance (772 variants) dualipa_log2fc (log2 fold change of abundance to the matched WT wells); dualipa_call (decrease / increase / no_detected_change, from empirical WT tails); dualipa_n_measurements; dualipa_hit (decrease or increase) sqY2H PPI (500 variants) ppi_norm_dist_rms (interaction-profile distance to the WT); ppi_num_edges; ppi_edgotype and the partner counts ppi_n_ppi, ppi_n_perturbed, ppi_n_unperturbed, ppi_n_uncertain; ppi_hit (empty: no variant-level call yet) Imaging (607 variants) img_pheno_score, img_pheno_hit, img_pheno_hit_strict (phenotype); img_misloc_score, img_misloc_hit (mislocalisation); img_U2OS_cc_t_stat, img_U2OS_cc_t_pval, img_U2OS_cc_t_adj_pval (cell count) and img_U2OS_prot_abun_t_stat, img_U2OS_prot_abun_t_pval, img_U2OS_prot_abun_t_adj_pval (protein abundance): t-tests in U2OS cells, Benjamini-Hochberg adjusted; img_Metadata_Bio_Batch Earlier studies (as mapped in the allele-metadata record) Study Columns Sahni et al. 2015 sahni_allele_id, sahni_edgotype_class, sahni_edgotype_class_source, sahni_n_edges, sahni_chaperone_responsive (Y / N), sahni_n_chaperones_expressed Rolland et al. 2014 rolland_allele_id, rolland_edgotype_class, rolland_edgotype_class_source, rolland_n_edges Luck et al. 2020 luck_allele_id, luck_edgotype_class, luck_edgotype_class_source, luck_n_edges Lacoste et al. 2024 lacoste_allele_id, lacoste_mislocalized (0 / 1), lacoste_mislocalization_phenotype, lacoste_impact_score, lacoste_ppi_phenotype, lacoste_variant_alternative, lacoste_collection, lacoste_source, lacoste_mutation_id For Rolland and Luck the edgotype is derived from the Y2H scores of the variant's partners (*_edgotype_class_source = derived): perturbed with one partner and retained with another is Edgetic, perturbed with two or more is Quasi-null, retained with two or more is Quasi-wild-type, and a single partner gives a degree-1 class. Changes from version 0.2 sqY2H scores from version 2.0 of the sqY2H record (doi:10.5281/zenodo.23924614; version 0.2 used 1.1, doi:10.5281/zenodo.23815551): the spots are scored from the day-4 plate photographs instead of CellProfiler, with thresholds set from the image scores. Same columns. 500 variants with sqY2H data (was 502): 492 in both, 10 no longer measured, 8 newly measured. Three variants measured only by sqY2H leave the table (CLN6 Ala12Thr, HMBS Arg173Gln, KRT6A Glu461Lys) and three enter (HMBS Leu81Pro, TTR Ala56Pro, TTR Tyr89His); the table keeps 3,739 variants. ppi_edgotype: 111 of the 492 variants in both versions change class, mostly Edgetic → Quasi-wild-type (25), Uncertain-degree-1 → Unperturbed-degree-1 (13) and Perturbed-degree-1 → Unperturbed-degree-1 (11). edgotype_class: 1% 500, Sahni 479, Luck 14 (was 502 / 479 / 12). variant_changes.csv: two new reasons, 1pct_no_longer_measured and 1pct_newly_measured, for variants measured only in the 1% (previously counted as study alleles without a clone or newly matched). DUAL-IPA, imaging, ClinVar and the allele-metadata columns are unchanged. Changes from version 0.1 Keyed by SPDI on a representative clone of the allele-metadata table, instead of gene_variant collapsed on spdi_joined. Column names are snake_case: Source / Source_Simple, Edgotype_class / Edgotype_source and clinvar_clinical_significance_unified are now sources / studies, edgotype_class / edgotype_source and clinvar_clnsig_clean. dualipa_score (a Z score) is replaced by dualipa_log2fc, on another scale; ppi_score is ppi_norm_dist_rms; imaging_score / imaging_hit are img_pheno_score / img_pheno_hit. 4,122 variants in version 0.1, 3,739 now, 3,707 in both. Only in version 0.1: 184 predicted-splice variants (now left out everywhere), 158 1% variants measured only on clones without sequence confirmation, 45 study alleles dropped by an allele-table rule (38 SPDI reference mismatches on GRCh38, 4 GNAS, 3 on a partial EGLN1 ORF), 17 Lacoste alleles without a clone, 8 under another SPDI, 3 1% variants no longer measured. New: 22 study alleles newly matched to a clone, 4 1% variants newly measured, 6 under another SPDI. Every variant and reason: variant_changes.csv. DUAL-IPA from the rebuilt analysis (log2 fold change and empirical WT tails): 21 of the shared variants change call. sqY2H edgotypes from the rebuilt pipeline's current edge call on the image scores: 112 change class. Imaging is unchanged. ClinVar from the 2026-10 release on the representative clone; the studies' own ClinVar labels and the 2025-01 lookup are no longer used. Middle and full versions added, with the allele-metadata annotation. Notes The imaging results will be replaced by an updated imaging analysis, and ppi_hit filled once the sqY2H pipeline has a variant-level call; both will come as new versions of this record. Licence: CC BY-NC-SA 4.0, inherited from the allele-metadata table (share-alike). The terms of each source are in data_licences.md and THIRD_PARTY_NOTICES.md.