AARD

Summary: This single entry stands for all 15 rows in the GOA file: they carry an identical term, qualifier, evidence code and reference, and differ only in the WITH/FROM partner, so they are one annotation supported by 15 interactors, all from the HuRI binary interactome. It is also the entirety of AARD's GO record - the gene has no molecular function beyond binding, no cellular component and no biological process. Rather than discount the set on the usual grounds (single publication, no follow-up), a specific alternative explanation was tested against a null model: coiled-coil bias, the known tendency of sticky or self-activating yeast two-hybrid preys to recover coiled-coil proteins through promiscuous heptad-repeat association. Seven of the fifteen partners carry a coiled-coil FEATURE. Against a background of 2059 of 20431 reviewed human proteins with that same feature (10.1%), fetched from UniProt for this purpose, that is a 4.6-fold enrichment with a binomial P(X>=7 | n=15, p=0.101) of 3.3e-04. Numerator and denominator use the same criterion deliberately: two further partners (KRT24, KRT27) carry only the Coiled coil KEYWORD, which UniProt assigns more liberally than the feature, and scoring them in against a feature-only background would overstate the enrichment. On the consistent feature-or-keyword pair it is 9 of 15 against a 10.6% keyword background, 5.6-fold, P = 4.8e-06 - the same conclusion either way. With n=15 this remains a descriptive statistic rather than a controlled test. The partners also come from irreconcilable compartments. Collapsing UniProt's hierarchical location strings to their leading term gives 14 top-level compartments across 15 proteins. Three partners are syntaxins (STX1A, STX2, STX5) - SNAREs from three different membranes that share a coiled-coil SNARE motif - and two are keratins (KRT24, KRT27). The remainder include centrosomal and centromeric coiled-coil proteins (CEP57, KIAA0753, CENPQ), a nuclear splicing factor (TFIP11), an endosomal protein (VPS37C), a secreted protein and a transcription factor (LMO4). There is no compartment in which these could plausibly meet, and what they share is architecture rather than biology. Marked over-annotated rather than removed: the interactions may genuinely have occurred in the assay, and there is no positive evidence that any individual pair is false. But the set as a whole should not be read as AARD's interactome, and bare protein binding conveys nothing about its function in any case.

Reason: A single high-throughput screen contributing every annotation the gene has, with a partner set that matches the coiled-coil Y2H artefact signature (7/15 with a coiled-coil feature, a 4.6-fold enrichment over the 10.1% human background; 14 top-level compartments; three syntaxins and two keratins) and no functional follow-up for any pair, recorded under an uninformative term.