VarChAMP sqY2H protein-protein interaction screen: edge- and variant-level scores

Edge- and variant-level scores of the sqY2H (semi-quantitative yeast two-hybrid) screen of VarChAMP, from the scoring pipeline varchamp-ppi-sqy2h (GitHub, commit 4965eb0), computed from the input record doi:10.5281/zenodo.23924656 (version 1.1, which adds the day-4 plate photographs scored here) and the VarChAMP allele-metadata table (doi:10.5281/zenodo.23376506). The screen tests disease-gene baits, wild type and missense variants from the VarChAMP 1% allele collection, against a panel of interactor preys. Version 2.0 scores the colonies directly from the plate photographs instead of using the CellProfiler spot scores of versions 1.x. The scores, thresholds and calls all change; see Notes. Files File Content edges_llr_intensity.parquet, edges_llr_intensity.tsv Edge level. One row per variant edge (variant bait × prey): 1,876 edges on 500 variants, 75 baits and 275 preys. Growth scores, log2fc against the WT bait, the current perturbation call and the original one. Columns below variants.parquet, variants.tsv Variant level. One row per variant (500): norm_dist_rms, the profile distance to the WT, num_edges, the counts of the variant's edges by call, and its edgotype 2.03-mt-ppi-edge-llr-intensity.summary.json, 2.02-mt-ppi-variant-distance.summary.json Provenance of each table: pipeline commit, SHA-256 of every input, code file and output, parameters, counts Each table is given as .parquet (keeps the column types) and .tsv (the same table, tab-separated; missing values are empty). Spot scoring Each registered day-4 photograph is fitted with the 48 × 48 colony lattice, subplate by subplate, with a check that each subplate's indices agree with its neighbours'. The growth score of a spot is its background-subtracted integrated grey intensity. The first lattice column, against the plate wall, is not used. Neither are failed pins or subplates whose spots do not match their labels. The consensus growth of a bait-prey pair on one medium is the median of its replicate spots. Edge scoring log2fc = log2((c + 1) / (c_WT + 1)) compares the consensus growth of the variant bait with that of its WT bait, for the same prey, medium and batch. The medium (best_condition) is the most stringent of LWH1, LWH10 and LWH25 on which the WT pair grew. The WT-growth thresholds are fixed at 21,700 / 20,000 / 21,300: the no-growth/growth valley of the score distribution on LWH1, and on LWH10 and LWH25 the score at which 90% of groups carry a manual score of at least 2. A pair is kept only if it mated, meaning an LW consensus of at least 15,000. For 3 edges, the variant had no consensus on that medium. They are scored on the next less stringent medium where the WT grew (condition_name), against the WT on that medium. The call is a log-likelihood ratio (LLR) from a two-component mixture fitted to the log2fc of all variant edges. Unperturbed edges are noise around a zero. Perturbed edges carry a loss of at least 2-fold: a latent loss truncated at −1, plus the same noise. The loss is centred at −2.1 on LWH1 and −3.6 on LWH10 and LWH25, the peak of each medium's loss. Perturbed: LLR ≤ −2; unperturbed: LLR ≥ 2; uncertain in between. Current call: the noise and the zero depend on the WT's growth, as dim WTs are noisier (median noise SD 0.59 for a WT consensus below 28,100, 0.50 above 40,900). 570 perturbed, 141 uncertain and 1,165 unperturbed edges. Original call: the same model with one noise level for every edge, as in the original analysis. 572 perturbed, 154 uncertain and 1,150 unperturbed edges. Edge table columns Columns Content gene_variant, symbol, aa_change, nt_change, ccsb_mutation_id, db_orf_id, db_mut_id Bait variant ad_symbol, ad_orf_id Prey (interaction partner) retest_batch, best_condition, condition_name, assay_id, scoring_pla Where the edge was measured. condition_name is the medium scored; it differs from best_condition only for the 3 fallback edges consensus_score, consensus_mad, wt_consensus_score, log2fc Consensus growth (image score) of the variant and the WT, the variant's replicate spread (median absolute deviation), and their log2fc consensus_manual_score, wt_consensus_manual_score, log2fc_manual The same from the manual scores (0–4); not used for the calls perturbation_LLR_intensity, perturbation_status_intensity, hit_intensity Current call (hit: True perturbed, False unperturbed, empty uncertain) mu_wt, sigma_e, log2_wt, p_loss Zero and noise of an unperturbed edge at this WT growth; p_loss: probability of a log2fc this low for an unperturbed edge perturbation_LLR, perturbation_status, hit Original call status_changed The two calls differ log2fc_lw, variant_lw_score, wt_lw_score The same contrast on LW, a medium on which growth does not depend on the interaction; used as a null check of the noise model, not for the calls perturbation_LLR_uniform, perturbation_status_uniform An alternative model of the perturbed edges, documented, not used n_spots, n_size_scores, n_manual_scores, n_contaminated, manual_consensus_failed, aa_score Consensus and QC details of the variant's spots; n_size_scores: spots with an image score; aa_score: growth of the empty-AD control seq_ad, seq_db, db_seq_final Sequence confirmation of the prey and bait clinvar_clnsig_clean, clinvar_clnsig_clean_pp_strict ClinVar class from the allele-metadata table; annotation only is_wt_bait, is_variant_bait, is_aa_control, is_rrs_control, first_unique_id, ppi_group_clean Pipeline bookkeeping Variant level norm_dist_rms is the original profile distance: the root-mean-square difference between the variant's and the WT's consensus scores (divided by 74,300, the top of the image-score scale) over the preys measured for both; num_edges is the number of preys in the bait's profile. edgotype summarizes the variant's edges under the current call, as in the original analysis. n_ppi and n_measured_ppi count its edges, and n_perturbed, n_unperturbed and n_uncertain count them by call; uncertain edges are measured but neither perturbed nor unperturbed. edgotype Measured edges Rule Variants Quasi-wild-type 2+ 2+ unperturbed, none perturbed 134 Edgetic 2+ at least one perturbed and one unperturbed 79 Quasi-null 2+ 2+ perturbed, none unperturbed 49 Uncertain 2+ none of the above 13 Perturbed-degree-1 1 perturbed 87 Unperturbed-degree-1 1 unperturbed 124 Uncertain-degree-1 1 uncertain 14 No variant-level hit call is included in this version. Notes Version 2.0 replaces the CellProfiler spot scores of versions 1.x with scores computed from the plate photographs. CellProfiler's wells sat at least 0.75 pitch off the colonies on 124 of 798 subplates. Its score (integrated intensity × colony area) also compressed faint growth to near zero, which turned small growth drops into strong losses. The thresholds and the loss centre are now set from the image scores. Columns outlier_flag and manual_correction_applied are removed, since image scoring uses neither outlier masking nor manual correction, and consensus_mad is added. No clinical label was used for scoring or calling. The full method, with every step and its difference from the original analysis, is docs/ppi-methodology.md in the repository. Credits CellProfiler colony quantification workflow of versions 1.x: Ahmad Kamal Hamid, Cimini Lab, Imaging Platform, Broad Institute of MIT and Harvard.