. . . . "fetch.py downloads go-basic.obo and the MGI GAF. analysis.py drops NOT-qualified and obsolete-term lines, builds each gene's set of experimental BP terms and their is_a/part_of ancestors, removes 612 genes annotated to ion channel, gap junction or neurotransmission terms, and tests overlap of GO:0007611 and GO:0009653 in the remaining 10,243 genes with a one-sided Fisher exact test. Annotation depth is controlled by a Mantel-Haenszel odds ratio over ten equal-sized strata of genes ranked by their number of distinct experimental BP terms. Specificity is the percentile of the odds ratio among the 327 BP terms of 128-286 genes that are neither ancestors nor descendants of either term." . . "Measured: 2.6573 odds ratio, learning-or-memory vs anatomical-structure-morphogenesis gene sets (mouse, experimental BP annotations, channel/synaptic genes excluded) (n = 10243, one-sided Fisher exact p = 1.140e-09). Interpretation: TAME's prediction is phrased across animals; the test uses the mouse because MGI curation gives the deepest experimental annotation of both developmental and behavioural phenotypes in a single species. 'Genes involved in morphogenesis' is GO:0009653 anatomical structure morphogenesis and 'memory and learning categories' is GO:0007611 learning or memory, each with is_a and part_of descendants. Only experimental evidence codes (EXP, IDA, IPI, IMP, IGI, IEP, HTP, HDA, HMP, HGI, HEP) are used, so electronic or sequence-similarity annotations cannot create the overlap. The universe is every gene with at least one such BP annotation, after removing NOT-qualified lines. 'Beyond the known shared ion channels, connexins and neurotransmitter machinery' is implemented by excluding from the universe every gene annotated (any evidence) to GO:0005216 ion channel activity, GO:0005243 gap junction channel activity, GO:0005921 gap junction, GO:0007268 chemical synaptic transmission, GO:0001505 regulation of neurotransmitter levels or GO:0030594 neurotransmitter receptor activity, with descendants. Only the first part of the claim is tested; its second part (memory genes appearing in cancer, regeneration and embryogenesis screens) is out of scope here." . "Claim test of levin-2022-tame-19: refutes, confidence 0.7" . . . "Planned datasets: go-basic-obo, mgi-gaf" . "claim-test/0.1.0" . "biological" . "Pre-registered: unstratified odds ratio <= 1.2 or p >= 0.05, or Mantel-Haenszel odds ratio <= 1.2, or the odds ratio at or below the median of the size-matched control terms. Outcomes meeting neither criterion are pre-registered as 'mixed' (overlap present but not specific, or explained by annotation depth)." . "GO annotations reflect what has been studied: genes are annotated to learning or memory largely because they were knocked out and tested behaviourally, which happens mostly to genes already well studied in development. Annotation-depth stratification addresses this only approximately." . "Mouse only and experimental evidence only. TAME frames the prediction across animals; other species or inclusion of inferred annotations could differ." . "Only the first half of the claim is tested. Whether memory genes appear in cancer, regeneration and embryogenesis screens is not addressed." . "Size-matched control terms include developmental processes adjacent to morphogenesis (axon, limb, cardiac chamber, neural tube development) with very high odds ratios, which raises the 75th-percentile bar; the depth-stratified result does not depend on this choice." . "The GO release is 'current' at retrieval time and is not versioned in the URL; re-running later will fetch a newer release, so exact numbers will drift. The SHA-256 above identifies the files used." . "The crude overlap is real: 69 genes against 34 expected (odds ratio 2.66, p = 1.1e-9), and removing channel, gap-junction and neurotransmission genes barely changes it (2.64 before exclusion). It does not survive control for annotation depth: learning-or-memory genes sit almost entirely in the most-annotated strata, and within the top decile (13-167 BP terms) 43 of 63 of them (68%) are morphogenesis genes against 621 of 961 other genes (65%). The pre-registered fail rule is met by the Mantel-Haenszel odds ratio (0.94 <= 1.2). The control comparison (64.5th percentile) fails pass criterion (c) but does not by itself meet the fail rule, since 2.66 is above the control median of 1.87." . "Pre-registered: after exclusions, (a) the one-sided Fisher exact test of overlap between the learning-or-memory set (GO:0007611 and is_a/part_of descendants) and the morphogenesis set (GO:0009653 and descendants) gives odds ratio > 1.5 and p < 0.001; (b) the Mantel-Haenszel odds ratio stratified by deciles of each gene's number of experimental BP annotations is > 1.5; (c) the unstratified odds ratio is above the 75th percentile of odds ratios between the morphogenesis set and every control BP term whose post-exclusion gene count is within a factor of 1.5 of the learning-or-memory set's and which is neither an ancestor nor a descendant of GO:0007611 or GO:0009653." . "Pre-registered: unstratified odds ratio <= 1.2 or p >= 0.05, or Mantel-Haenszel odds ratio <= 1.2, or the odds ratio at or below the median of the size-matched control terms. Outcomes meeting neither criterion are pre-registered as 'mixed' (overlap present but not specific, or explained by annotation depth)." . "0.7"^^ . . . "refutes" . "Among mouse genes with experimental Gene Ontology biological-process annotations, genes annotated to learning or memory are over-represented among genes annotated to anatomical structure morphogenesis, even after removing ion channels, gap-junction channels and neurotransmission genes, and this overlap is larger than the overlap of morphogenesis with other biological-process terms of similar size." . "fetch.py downloads go-basic.obo and the MGI GAF. analysis.py drops NOT-qualified and obsolete-term lines, builds each gene's set of experimental BP terms and their is_a/part_of ancestors, removes 612 genes annotated to ion channel, gap junction or neurotransmission terms, and tests overlap of GO:0007611 and GO:0009653 in the remaining 10,243 genes with a one-sided Fisher exact test. Annotation depth is controlled by a Mantel-Haenszel odds ratio over ten equal-sized strata of genes ranked by their number of distinct experimental BP terms. Specificity is the percentile of the odds ratio among the 327 BP terms of 128-286 genes that are neither ancestors nor descendants of either term." . "Pre-registered: after exclusions, (a) the one-sided Fisher exact test of overlap between the learning-or-memory set (GO:0007611 and is_a/part_of descendants) and the morphogenesis set (GO:0009653 and descendants) gives odds ratio > 1.5 and p < 0.001; (b) the Mantel-Haenszel odds ratio stratified by deciles of each gene's number of experimental BP annotations is > 1.5; (c) the unstratified odds ratio is above the 75th percentile of odds ratios between the morphogenesis set and every control BP term whose post-exclusion gene count is within a factor of 1.5 of the learning-or-memory set's and which is neither an ancestor nor a descendant of GO:0007611 or GO:0009653." . "false"^^ . . "term hierarchy (is_a and part_of) used to propagate annotations and define the gene sets" . "CC BY 4.0" . . "Dataset: Gene Ontology, basic edition (go-basic.obo)" . . . "sha256:b08d45b268b8c24ccb2513dbbbc7d4df9f6521c099b413f79eb31e06e0fa3bcc" . "experimental biological-process annotations of mouse genes (gene sets and universe) and any-evidence annotations (exclusions)" . "CC BY 4.0" . . "Dataset: Mouse Genome Informatics GO annotations (GAF 2.2), via the GO Consortium" . . . "sha256:9b7613f786f9f805531a8f6768f310220d28c82095f938e1212c86f39596ce30" . . "delphisci claim tester (claude-opus-5.5; cursor local session)" . . . "claude-opus-5.5" . . . . . . . . . "2026-11-08T16:24:31.877Z"^^ . . . . . "Claim test of levin-2022-tame-19: refutes, confidence 0.7" . . "RSA" . "MIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKCAQEAyFHtPB6p0mCjDozxlOY9CmO3nBm7Te3jFbXyLmM+G1tGe47Da0jYIbhkl9Oo1ZjxmCa3LCwlEqK8gj6hEIcL4ahW30JXh9HycfKVk1bugkzMEV15tJdTzAU3sGMFnbNsn43xY2LFLy45tApkRm5byEiM7oRRrooIdwoQKnP4Wzq/bfAoJ9qvW+MGkelyOTd1jbSt+KfmT7xKmDfWzAHddUlWwMX2QJOLxDaffOteTop6ZRgPiNmPJKA/S62Gac9XtSYT3oDkZ6/nr5cxxNLpOyxTf5FPISiCe11xYO/unYjY+mwnOfA8yhJIyx1l0O9I0vpOpn+E5TvyHJV+oSTrvwIDAQAB" . "N/psZIhmH/DWHFCwWZlBrRStwApXrSWIAe0dSYfWOVKA2ggs8ldH2RtxYd+6RAiFkQi8VDfPQFpsfJA7byPrt++SozPUdZbT+SpaHF7uZf4+AlSf+QcYJi40PngJKOFTQkkl5wvzLa+JmnnNqPC88nVFQ6CNczpFg4POdZ2v+3wXTHW1F1d71ZhNgM+S3GRp3XvUWuj/32m8f+qfklNkdEP/CjRcKS/HS79Wpum3UmkY5iSZYR1tNdJ9ePUuR1V9tbGffrVvBMKDh/V0yt5Jsfb1KY3mH4S3ucUUrI/JrrsbRcAPmomcptkB5n4eQlAnTXz6jxiBGOOW/qjib7k8Ow==" . . .