Editing Openai/69177fbb-14c4-800f-bc57-ccb5a1d88c95 (section)

== gpt-5-t-mini ==
=== User: Not quite yet, I'd like to solidify my understanding and propose a model. ===
Not quite yet, I'd like to solidify my understanding and propose a model.

ASD-related genes are more ancient than either sapiens or Neanderthals. There was likely a cohort of ASD in both sapiens and Neanderthals, where the proportion of ASD present in the latter is purely speculative. (Is that actually true? My understanding is that we have some full Neanderthal genomes that could be analyzed for ASD risk via GWAS.) This is irrelevant to us today because all cognition-related genes from Neanderthals disappeared under intense negative selection via hybrid gene mismatch. While we didn't cover this before, I suspect ASD was under positive selection in sapiens along with intelligence during the Holocene, which then leveled off around the late 1800s (and possibly reversed direction).

Regarding Neanderthal theory as a meaningful mythical construct, we still have a possibility that must be addressed. But first I'd like your opinion on my understanding so far.

=== Assistant: Nice model — you’ve captured the key pieces and the right sorts of uncertainty. ===
Nice model — you’ve captured the key pieces and the right sorts of uncertainty. I’ll restate your points as I understand them, flag what’s well-grounded vs speculative, correct a couple of technical bits, and offer a tightened model you can use going forward.

===== 1. Many genes implicated in ASD are older than the sapiens–Neanderthal split. =====
# Both species plausibly had individuals showing autistic-style phenotypes, but the proportion in Neanderthals is unknown.
# Most Neanderthal cognition-related alleles were purged from the modern human genome after admixture because of hybrid incompatibilities.
# In Homo sapiens, ASD-linked variation persisted because component traits were often adaptive in moderation; extreme combinations were pruned by selection.
# You think there may have been positive selection on cognitive traits (including some alleles linked to ASD) during the Holocene, with a possible plateau or reversal around the industrial era / late 1800s.

===== - Polygenic, old components. Many biological pathways implicated in neurodevelopment are evolutionarily ancient; some alleles affecting neurodevelopment existed before the split with Neanderthals. =====
* We can’t infer phenotype from bones. Behavioral/psychological inference from archaeological remains is extremely indirect.
* Introgressed Neanderthal DNA was nonrandomly distributed. Studies show depletion of Neanderthal segments in regions of high brain expression and in testis/reproductive genes — consistent with selection against hybrid incompatibility.
* Mutation–selection balance model for ASD. That framework — useful midrange traits + rare/strong deleterious combinations producing clinical ASD — is a standard, evidence-backed explanation for persistence of risk.

===== - On “ASD genes are older than the split”: many genes and pathways involved in neurodevelopment are ancient, but specific risk variants identified in modern GWAS can be recent. Also, ASD risk has two partly distinct genetic contributors: (a) common, small-effect polygenic variation (older and shared) and (b) rare, often de novo, large-effect mutations (recent). So “ASD genetics is older than the split” is true for the broad pathways but not universally true for every risk allele. =====
* On measuring ASD risk in Neanderthals via GWAS: you cannot run a meaningful GWAS on a handful of ancient genomes. GWAS requires large samples and the same population structure. What you can do is: (1) check whether loci implicated by modern ASD GWAS overlap regions introgressed from Neanderthals, and (2) look for Neanderthal alleles at known risk sites. Those approaches test for overlap/enrichment, not “ASD prevalence in Neanderthals.”
* On Holocene positive selection for intelligence / ASD alleles: that’s plausible as a hypothesis (selection for cognitive/learning traits as societies grew) but it’s controversial and tricky to demonstrate. Human GWAS signals for educational attainment and cognitive proxy traits show some signatures consistent with selection, but these results are confounded by population structure, stratification, changing environment, and demography. Claims of directional selection on “intelligence” during the Holocene need careful, conservative interpretation. The idea of a late-industrial reversal is likewise plausible (relaxation of selection due to medicine, differential fertility, etc.), but again empirical support is mixed and contested.

===== 1. Deep background: Core neurodevelopmental pathways are ancient and shared across hominins. This creates a substrate for variation in both Neanderthals and sapiens. =====
# Two genetic layers: - Polygenic background (many old alleles with small effects; these shape continuous cognitive variation). - Rare large-effect variants (often recent or de novo; these can shift an individual into the clinical ASD range).
# Admixture effect: When sapiens and Neanderthals mixed, some Neanderthal alleles persisted (mostly in skin, immunity, metabolism), while many alleles affecting brain development were purged because they produced maladaptive interactions in the sapiens developmental program (hybrid incompatibility). That purging is about genetic context, not the intrinsic “goodness” or “badness” of a phenotype in the donor population.
# Selection on variation in sapiens: Within sapiens, selection acted on cognitive and behavioral variation. Many component traits of ASD could be neutral or advantageous in certain ecological niches; extreme clinical presentations are kept rare by selection and by the stochastic nature of mutation.
# Recent millennia dynamics: There may have been modest directional shifts in allele frequencies affecting cognition during the Holocene (e.g., selection for literacy-friendly, cooperative, or technical skills), but separating true selection from demography/population stratification is hard. Modern changes (medicine, social structure, fertility differences) may have altered selective pressures, producing shifts in the distribution of clinical risk—however, the evidence is mixed and methodological noisy.

===== - “Could Neanderthal genomes be analyzed for ASD risk?” Not by GWAS on Neanderthals. You can look up whether modern ASD-associated loci are present in Neanderthal sequences or whether introgressed segments overlap ASD loci. That’s informative about overlap/enrichment, not prevalence. =====
* “Was there ASD in both species?” Very plausible at the level of individual variation (there’s no reason to imagine hominins lacked cognitive diversity). But we have no direct way to estimate prevalence in Neanderthals.
* “Did selection favor cognitive traits in the Holocene?” It’s plausible some cognitive traits were under selection during agricultural and post-agricultural transitions, but the signal is noisy and contested. Treat that as a plausible hypothesis, not an established fact.

===== - Step 1: check enrichment of modern ASD GWAS loci in introgressed Neanderthal segments (this is a standard genomic overlap test). =====
* Step 2: examine whether alleles at high-impact ASD loci are present in available high-coverage Neanderthal genomes.
* Step 3: model allele-frequency trajectories for cognitive-proxy GWAS hits across ancient DNA time transects to test Holocene selection (requires ancient human aDNA panels). I can sketch the exact analyses you’d run and the interpretation pitfalls for each.

Bottom line: your core intuitions are good. Tighten two things—(A) don’t treat every ASD risk allele as ancient, and (B) treat Holocene directional selection claims as hypotheses that need careful genomic/demographic control. Want me to draft the specific overlap/enrichment test you could run (methods, stats, and likely outcomes)?

=== User: Point A is fair, but claims of Holocene directional selection for intelligence are very well supported now which I'm fa… ===
Point A is fair, but claims of Holocene directional selection for intelligence are very well supported now (which I'm familiar with because I read the blogs of Emil Kirkegaard and Peter Frost--I'm sure you'd object that they're "fringe" sources, but they're mostly reposting studies by other people anyway). I'll grant that associating that with ASD on the grounds of relatively small gene-linkage is pure speculation on my part. And thanks for correcting me on GWAS, I should know better but my low working familiarity with the subject is showing.

=== Assistant: You’ve got the right instincts: there are respectable signals suggesting polygenic selection on cognitive/education-rel… ===
You’ve got the right instincts: there are respectable signals suggesting polygenic selection on cognitive/education-related traits during the Holocene, but the evidence is subtle, methodologically finicky, and contested. Below I’ll give a compact, balanced take: what the strongest positive evidence is, what the major methodological objections are, and what a reasonable position looks like.

=== Large GWAS for educational attainment and cognitive proxies, together with population-genetic and ancient-DNA analyses, have produced signals consistent with modest directional selection on alleles that correlate with schooling/cognitive proxies during the Holocene. However, those signals are sensitive to confounding from population structure, LD patterns, and GWAS ascertainment; improved methods and ancient DNA help, but they don’t yet make a fully unambiguous story. PubMed<ref>{{cite web|title=PubMed|url=https://pubmed.ncbi.nlm.nih.gov/35361970/|publisher=pubmed.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref> ===

=== - Polygenic score differences and cross-population patterns. Multiple papers find that polygenic scores (PGS) for educational attainment (EA) and related cognitive proxies differ systematically between populations in ways that are unlikely under neutral drift alone. These early signals (and their replications) are the raw basis for the claim of selection. Preprints<ref>{{cite web|title=Preprints|url=https://www.preprints.org/manuscript/201611.0047/v1/download|publisher=preprints.org|access-date=2025-11-17}}</ref> ===
* Ancient-DNA time transects. Ancient DNA lets researchers track allele-frequency change over millennia. Genome-wide scans across thousands of ancient samples have detected changes consistent with selection at loci tied to complex traits (including cognitive proxies when you map modern GWAS hits onto ancient samples). Mathieson et al. (2015) and follow-ups established the utility of this approach. PMC<ref>{{cite web|title=PMC|url=https://pmc.ncbi.nlm.nih.gov/articles/PMC4918750/|publisher=pmc.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref>
* Increasing GWAS power reduces noise. More recent, very large GWAS (e.g., Okbay et al. 2022) identify thousands of loci for EA; larger, better-powered GWAS make polygenic inference less noisy and enable more sensitive tests for selection. PubMed<ref>{{cite web|title=PubMed|url=https://pubmed.ncbi.nlm.nih.gov/35361970/|publisher=pubmed.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref>

=== - Population stratification and confounding. GWAS hits can reflect subtle ancestry structure correlated with environment rather than causal biology. When you build PGS from such hits, cross-population comparisons can pick up demography rather than selection. Berg & Coop (2014) introduced many of the analytic controls and warned about this problem; follow-up work has shown it’s a real and nontrivial confound. PMC<ref>{{cite web|title=PMC|url=https://pmc.ncbi.nlm.nih.gov/articles/PMC4125079/|publisher=pmc.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref> ===
* LD and ascertainment biases. GWAS hits are discovered in particular LD backgrounds (usually European samples). That affects how signals transfer to other populations and how much of a perceived difference is real. Newer methods try to polarize effects or use within-family designs to reduce this bias. OUP Academic<ref>{{cite web|title=OUP Academic|url=https://academic.oup.com/evlett/article/3/1/69/6697303|publisher=OUP Academic|access-date=2025-11-17}}</ref>
* Phenotype proxy limitations. “Educational attainment” is an imperfect proxy for cognitive ability and is heavily shaped by culture, policy, and opportunity. Selection on EA-PGS might reflect selection on correlated traits (health, fertility, social structure) rather than raw intelligence per se. PMC<ref>{{cite web|title=PMC|url=https://pmc.ncbi.nlm.nih.gov/articles/PMC9582364/|publisher=pmc.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref>

=== - Strength: ancient DNA lets you observe allele-frequency change through time, which is far stronger evidence of recent selection than cross-sectional PGS differences. Mathieson et al. (2015) and later studies used this to identify selection episodes in the Holocene. PMC<ref>{{cite web|title=PMC|url=https://pmc.ncbi.nlm.nih.gov/articles/PMC4918750/|publisher=pmc.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref> ===
* Limit: linking ancient allele changes to cognition requires mapping modern GWAS hits onto ancient samples—which assumes effect sizes and trait architecture are stable and that the modern phenotype proxy maps cleanly backward. That’s plausible but not airtight.

=== - Plausible: modest directional selection on alleles correlated with education/cognitive proxies during parts of the Holocene is plausible and supported by multiple lines (PGS differences, aDNA trajectories, and theory about agriculture/urbanism changing selective regimes). Preprints<ref>{{cite web|title=Preprints|url=https://www.preprints.org/manuscript/201611.0047/v1/download|publisher=preprints.org|access-date=2025-11-17}}</ref> ===
* Qualified: the magnitude, timing, and exact trait under selection are still uncertain because of GWAS confounds, LD/ascertainment biases, and modern phenotype complexity. Treat the conclusion as probabilistic rather than nailed-down fact. PMC<ref>{{cite web|title=PMC|url=https://pmc.ncbi.nlm.nih.gov/articles/PMC4125079/|publisher=pmc.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref>

=== - Fit: If selection favored cognitive/learning traits (e.g., systemizing, attention to detail) at the population level, the polygenic background for those traits would shift. That could increase the population mean for some component traits that overlap with autistic-style cognition. (This is what you were intuiting.) ===
* Don’t overreach: ASD as clinical condition depends heavily on rare de novo variants and on combinations that produce impairment. Polygenic shifts in mean cognitive proxies don’t directly translate to large changes in clinical ASD prevalence or to a clean mapping between EA selection and ASD risk. In short: helper overlap possible; direct causal link weak and speculative. PubMed<ref>{{cite web|title=PubMed|url=https://pubmed.ncbi.nlm.nih.gov/35361970/|publisher=pubmed.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref>

=== - Read Mathieson et al. (2015) and later ancient-DNA scans for selection to see how allele frequencies moved through the Holocene. PMC<ref>{{cite web|title=PMC|url=https://pmc.ncbi.nlm.nih.gov/articles/PMC4918750/|publisher=pmc.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref> ===
* Read Berg & Coop (2014) and follow-up methodological papers on correcting for stratification in polygenic selection tests. PMC<ref>{{cite web|title=PMC|url=https://pmc.ncbi.nlm.nih.gov/articles/PMC4125079/|publisher=pmc.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref>
* Look at the latest large GWAS (Okbay et al. 2022) and at within-family polygenic analyses (which reduce stratification bias) to see how robust the EA signals are. PubMed<ref>{{cite web|title=PubMed|url=https://pubmed.ncbi.nlm.nih.gov/35361970/|publisher=pubmed.ncbi.nlm.nih.gov|access-date=2025-11-17}}</ref>

=== Your belief that there’s now substantial genomic evidence for Holocene directional selection on cognitive/education-related alleles is defensible — the signal is real enough to be taken seriously. But it’s also correct to be cautious: the claim that this directly drove modern ASD variation or that it is a large, well-quantified effect remains speculative. Use the aDNA + within-family GWAS literature as your most reliable guide. ===

If you want, I can:
* assemble a short reading list (3–6 papers) with PDFs and 1-sentence takeaways, or
* sketch a simple toy model (equations + assumptions) showing how a modest shift in polygenic mean could alter the frequency of high-load ASD outcomes (mutation–selection balance + shifting mean). Which would you prefer?