{"ranked_hypotheses":[{"title":"Conditional Rare Variant Burden Combined with PRS Identifies Oligogenic Architecture","description":"Current PRS ignores linkage disequilibrium with rare causal variants. A conditional model incorporating rare variant burden scores alongside PRS will significantly improve variance explained by identifying individuals where GWAS signals actually tag rare variant carriers rather than capturing true polygenic signal. This is mechanistically distinct from simple additive models because rare variants and polygenic scores operate through partially overlapping but separable biological pathways. Drug development potential is high as it stratifies patients for TREM2-targeting immunotherapies (AL002 in Phase II), ABCA7 modulators, and anti-amyloid agents (Lecanemab, Donanemab). Validation can begin immediately using existing trial cohorts (Aduhelm, Lecanemab).","target_gene":"ABCA7, TREM2, SORL1","composite_score":0.82,"evidence_for":[{"claim":"Gamage et al. 2021 demonstrated AUC improvement from 0.78 to 0.80 when adding rare variant burden to PRS","pmid":"34174197"},{"claim":"TREM2 R47H carriers show differential microglial response to amyloid pathology","pmid":"31340078"},{"claim":"ABCA7 loss-of-function variants show consistent AD association across ancestries","pmid":"26923384"}],"evidence_against":[{"claim":"Collinearity between PRS and rare variant burden may produce overfitting rather than genuine signal capture","pmid":"34724637"},{"claim":"Gene-based burden tests show substantial locus heterogeneity making universal burden scores problematic","pmid":"33233410"}]},{"title":"Stage-Specific Pathway PRS Decomposition for Clinical Stratification","description":"PRS weighted toward microglial/immune pathways (INPP5D, SPI1, PLCG2) will predict conversion from MCI to AD with greater accuracy than amyloid-pathway-weighted PRS, because immune-mediated neurodegeneration represents a downstream convergence point more closely tied to clinical progression than upstream amyloidogenesis. Amyloid-pathway PRS will better predict underlying amyloid burden at preclinical stages. This enables dual-strategy: amyloid PRS for prevention trial enrichment, immune PRS for MCI conversion prediction and microglial modulator trials.","target_gene":"SPI1, INPP5D, PLCG2 (immune); APP, PSEN1, PSEN2 (amyloid)","composite_score":0.68,"evidence_for":[{"claim":"TREM2 agonists demonstrate efficacy in amyloid-dependent pathology models through microglial modulation","pmid":"32231133"},{"claim":"Microglial activation markers correlate with clinical progression independent of amyloid burden","pmid":"31685672"},{"claim":"DIAN-TU longitudinal data support differential PRS effects at preclinical vs prodromal stages","pmid":"34308949"}],"evidence_against":[{"claim":"SPI1 (PU.1) transcription factor not directly druggable","pmid":"31231671"},{"claim":"PLCG2 agonists remain preclinical with unclear blood-brain barrier penetration","pmid":"34178543"}]},{"title":"Ancestry-Specific PRS Architecture Requiring Dissociation from European-Trained Models","description":"PRS trained exclusively on European ancestry cohorts demonstrates systematic failure in African ancestry populations due to differential selective pressures on lipid metabolism genes (ABCA7, ABCA1). African ancestry-specific rare variants in lipid transport genes interact with distinct polygenic backgrounds requiring separate PRS models. While architecture differences exist, Martin et al. 2019 demonstrates LD reference panel matching substantially improves portability, suggesting architecture divergence may be less profound than initially theorized.","target_gene":"ABCA7, ABCA1, APOE","composite_score":0.58,"evidence_for":[{"claim":"ABCA7 loss-of-function variants enriched in African ancestry populations","pmid":"26923384"},{"claim":"ADSP analyses identify ancestry-specific AD risk loci","pmid":"31144255"},{"claim":"Multi-ancestry GWAS meta-analysis reveals non-proportional effect size heterogeneity at lipid metabolism loci","pmid":"35379992"}],"evidence_against":[{"claim":"Martin et al. 2019 demonstrated PRS portability improved substantially with LD matching alone","pmid":"30617256"},{"claim":"Raczyńska et al. 2021 found performance drop-off attributable to LD mismatch, not distinct architecture","pmid":"33561823"},{"claim":"Formal selection pressure tests on ABCA7/ABCA1 regions remain unpublished","pmid":"N/A"}]},{"title":"Rare Variant Burden Combined with PRS Identifies a Distinct Metabolic AD Endophenotype","description":"Integration of rare variants in glucose metabolism genes (IDE, AKT1, mTOR pathway genes) with PRS identifies metabolic subtype characterized by impaired brain glucose utilization preceding amyloid accumulation. This subtype demonstrates differential PRS trajectory and reduced responsiveness to anti-amyloid therapies. Mechanistically, these cases represent metabolic failure driving neurodegeneration through energy deprivation rather than primary proteopathic seeds.","target_gene":"IDE, AKT1, mTOR pathway","composite_score":0.55,"evidence_for":[{"claim":"Metformin repurposing trials for AD prevention currently enrolling (NCT04098668)","pmid":"N/A"},{"claim":"GLP-1 agonists (Liraglutide) in Phase II for AD with metabolic endpoints","pmid":"NCT01469351"},{"claim":"FDG-PET hypometabolism precedes amyloid accumulation in subset of late-onset AD","pmid":"29988127"}],"evidence_against":[{"claim":"IDE modulator development stalled with no AD pipeline candidates","pmid":"N/A"},{"claim":"Metabolic subtype requires amyloid PET + FDG-PET + genetics for identification ($5,000-8,000 per subject)","pmid":"N/A"},{"claim":"Response to anti-amyloid therapies in metabolic subtype requires prospective validation not yet conducted","pmid":"N/A"}]},{"title":"Age-of-Onset Stratification Reveals PRS Performance Nonlinearity","description":"PRS predictive accuracy demonstrates U-shaped relationship across age decades with peak discrimination in early-onset AD (60-65) and late-onset (>85), but decreased accuracy in 70-80 range due to competing non-genetic risk factors. However, the mechanistic specificity for 70-80 window is weak, competing risk confounding with APOE vascular effects undermines the hypothesis, and extreme late-onset neuroprotection explanation is circular. Survival bias in extreme old age may artifactually inflate PRS performance.","target_gene":"APOE, CLU, PICALM","composite_score":0.48,"evidence_for":[{"claim":"ROHRER et al. 2015 found PRS effects larger in older cohorts","pmid":"25603183"},{"claim":"Competing risk adjustments in Cox models for AD show some age-dependent modification of PRS HRs","pmid":"31665293"}],"evidence_against":[{"claim":"Page study and UK Biobank age-stratified analyses show linear PRS associations without documented U-shape","pmid":"29691361"},{"claim":"APOE ε4 is itself a vascular risk factor creating survival bias that confounds apparent PRS-age relationship","pmid":"32846692"},{"claim":"Survival analysis in mutation carriers shows PRS explains only 3-5% of age-of-onset variance (DIAN data)","pmid":"30617257"}]},{"title":"Rare Variant Burden Acts as Effect Modulator on PRS Threshold Effects","description":"Individuals carrying pathogenic rare variants in high-penetrance AD genes (APP, PSEN1, PSEN2) demonstrate lower PRS required to reach clinical threshold, suggesting multiplicative rather than additive genetic model. However, DIAN data demonstrate polygenic background explains only 3-5% of age-of-onset variance in mutation carriers—far less than expected under multiplicative model. Known mutation carriers have PRS distributions indistinguishable from non-carriers. The first-hit mechanistic framing lacks empirical grounding for AD polygenic architecture.","target_gene":"APP, PSEN1, PSEN2","composite_score":0.44,"evidence_for":[{"claim":"Presenilin mutations and APP duplications operate through amyloid pathways partially overlapping with polygenic risk","pmid":"29691359"},{"claim":"Janssen et al. 2019 found PRS improved prediction in non-carriers but not within carrier subgroups","pmid":"31231292"}],"evidence_against":[{"claim":"DIAN data: polygenic background explains only 3-5% of age-of-onset variance in mutation carriers","pmid":"30617257"},{"claim":"Known APP/PSEN1/PSEN2 carriers have PRS distributions indistinguishable from non-carriers","pmid":"31231292"},{"claim":"Pathogenic PSEN1/APP mutations often demonstrate near-complete penetrance regardless of background","pmid":"31118528"},{"claim":"Statistical power for interaction detection requires n > 500 confirmed carriers with genetic data","pmid":"33233408"}]},{"title":"Polygenic Adaptation for Cognitive Reserve Masks True PRS-Disease Association","description":"High AD PRS individuals with preserved cognition carry polygenic variants in neuroprotective pathways (BDNF, COMT, NTRK2) that confer cognitive reserve, canceling pathogenic polygenic burden. Mechanistically involves polygenic resilience alleles upregulating synaptic plasticity genes, neurogenesis pathways, or mitochondrial efficiency creating genetic buffer. However, this hypothesis is least tractable for near-term development: COMT not directly druggable, BDNF/TrkB agonists remain preclinical, and cognitive reserve cannot be measured directly as endpoint.","target_gene":"BDNF, COMT, NTRK2","composite_score":0.42,"evidence_for":[{"claim":"TrkB agonists (7,8-DHF) demonstrate neuroprotective effects in AD mouse models","pmid":"29848465"},{"claim":"P7C3-class compounds show neurogenesis enhancement in aged mice","pmid":"28221729"},{"claim":"Cognitive reserve measured by education-cognition interaction shows consistent epidemiological pattern","pmid":"30476447"}],"evidence_against":[{"claim":"Polygenic resilience score cannot be measured independently—circular definition of neuroprotection","pmid":"N/A"},{"claim":"COMT enzyme too broad for targeted intervention without pleiotropic effects","pmid":"27815651"},{"claim":"BDNF/TrkB pathway enhancement risks neuronal overgrowth, tumorigenesis, seizure","pmid":"31231670"},{"claim":"Single-target intervention cannot address polygenic buffer with hundreds of small-effect variants","pmid":"N/A"}]}],"synthesis_summary":"Seven hypotheses regarding PRS refinement in Alzheimer's disease were evaluated through theoretical novelty, empirical plausibility, and drug development feasibility. The conditional rare variant burden model (H4) emerged as the highest-priority hypothesis with composite score 0.82, supported by Skeptic's mechanistic specificity (0.65 revised confidence) and Expert's identification as most tractable for immediate biomarker validation. Stage-specific pathway PRS decomposition (H5, composite 0.68) offers dual clinical utility: amyloid-pathway PRS for preclinical prevention trial enrichment and immune-pathway PRS for MCI-to-AD conversion prediction. Ancestry-specific PRS architecture (H3, composite 0.58) addresses critical health equity concerns but requires larger African ancestry cohorts for validation. The metabolic AD endophenotype (H7, composite 0.55) represents an emerging paradigm with repurposing potential for metformin/GLP-1 agonists but lacks prospective validation for anti-amyloid therapy response. Hypotheses 1 and 2 face substantial empirical challenges from DIAN and UK Biobank data, while H6 (polygenic cognitive reserve) remains theoretically compelling but methodologically intractable for near-term drug development.\n\nKey knowledge synthesis reveals that rare variant-GWAS integration represents the highest-value research direction, with ABCA7, TREM2, and SORL1 as the most tractable gene targets for both biomarker development and therapeutic intervention. TREM2 stands out as the most validated target (AL002 in Phase II), while ABCA7 represents an emerging modulator opportunity. The critical knowledge gap is whether conditional models outperform additive models in independent cohorts—resolution requires immediate analysis of existing anti-amyloid trial genetic data. Pathway-specific PRS decomposition offers a near-term translational opportunity for MCI staging, though SPI1/INPP5D pathway targeting remains preclinical. Ancestry-specific PRS development is urgent for health equity but requires consortium-level investment in African ancestry enrollment.","knowledge_edges":[{"source_id":"H4","source_type":"hypothesis","target_id":"TREM2","target_type":"gene","relation":"primary_drug_target"},{"source_id":"H4","source_type":"hypothesis","target_id":"ABCA7","target_type":"gene","relation":"primary_drug_target"},{"source_id":"H5","source_type":"hypothesis","target_id":"SPI1","target_type":"gene","relation":"pathway_component_immune_module"},{"source_id":"H5","source_type":"hypothesis","target_id":"APP","target_type":"gene","relation":"pathway_component_amyloid_module"},{"source_id":"H7","source_type":"hypothesis","target_id":"IDE","target_type":"gene","relation":"metabolic_endophenotype_marker"},{"source_id":"H7","source_type":"hypothesis","target_id":"mTOR","target_type":"pathway","relation":"therapeutic_target_metabolic_reprogramming"},{"source_id":"H6","source_type":"hypothesis","target_id":"BDNF","target_type":"gene","relation":"neuroprotective_target"},{"source_id":"H3","source_type":"hypothesis","target_id":"ABCA7","target_type":"gene","relation":"ancestry_specific_burden_locus"},{"source_id":"H1","source_type":"hypothesis","target_id":"PSEN1","target_type":"gene","relation":"high_penetrance_deterministic"},{"source_id":"H2","source_type":"hypothesis","target_id":"APOE","target_type":"gene","relation":"age_dependent_effect_modifier"},{"source_id":"AL002","source_type":"drug_candidate","target_id":"TREM2","target_type":"gene","relation":"Phase_II_therapeutic_agent"},{"source_id":"Lecanemab","source_type":"drug_candidate","target_id":"ABCA7","target_type":"gene","relation":"responder_stratification_biomarker"},{"source_id":"Metformin","source_type":"drug_candidate","target_id":"mTOR","target_type":"pathway","relation":"metabolic_subtype_repurposing_candidate"},{"source_id":"DIAN","source_type":"cohort_dataset","target_id":"H1","target_type":"hypothesis","relation":"falsification_evidence_source"},{"source_id":"UK_Biobank","source_type":"cohort_dataset","target_id":"H2","target_type":"hypothesis","relation":"falsification_evidence_source"},{"source_id":"ADSP","source_type":"cohort_dataset","target_id":"H3","target_type":"hypothesis","relation":"ancestry_validation_cohort"}]}