67. MIND THE GAP: QUANTIFYING AND MITIGATING THE IMPACT OF VARIANT MISSINGNESS ON POLYGENIC SCORES WITHIN THE GENOPRED PIPELINE

{"Background":[0],"Polygenic":[1],"scores":[2,307],"(PGS)":[3],"are":[4],"powerful":[5],"tools":[6,335],"for":[7,208,298],"disease":[8,159],"risk":[9,194,316],"stratification,":[10],"but":[11,251,275],"their":[12],"translational":[13],"utility":[14],"is":[15,49,278],"challenged":[16],"by":[17,271],"variant":[18,31,81,246],"missingness":[19,98,174,266],"when":[20,196],"applied":[21,197,237],"across":[22,120,165,180,256,308],"diverse":[23],"genotyping":[24,167,323],"arrays":[25],"and":[26,35,42,69,122,284,302],"imputation":[27],"panels.":[28],"Discrepancies":[29],"in":[30,52,115,188],"coverage":[32],"between":[33],"discovery":[34],"target":[36,102],"samples":[37],"can":[38],"degrade":[39],"predictive":[40,124,189],"discrimination":[41],"distort":[43],"clinical":[44,200,338],"calibration,":[45],"yet":[46],"this":[47,276,326],"issue":[48],"often":[50],"ignored":[51],"standard":[53,257],"analytical":[54,107],"workflows.":[55],"We":[56,75,131],"introduce":[57],"a":[58,93,209,229,295],"dual":[59],"framework":[60,327],"integrated":[61],"into":[62,286],"the":[63,78,89,112,129,134,151,162,176,185,239,287,291,299,322,329],"open-source":[64],"GenoPred":[65,288],"pipeline":[66,289],"to":[67,110,137,192,198,225,238,336],"quantify":[68],"recover":[70,139],"lost":[71,140,234],"genetic":[72,315],"signal.":[73],"Methods":[74],"systematically":[76],"evaluated":[77,150],"impact":[79],"of":[80,118,154,178,231,233,282,305,321,331],"loss":[82],"using":[83],"whole-genome":[84],"sequencing":[85],"(WGS)":[86],"data":[87],"from":[88,161,333],"UK":[90],"Biobank":[91],"as":[92],"ground":[94],"truth.":[95],"To":[96],"address":[97],"without":[99],"requiring":[100],"individual-level":[101],"data,":[103],"we":[104,149,243],"developed":[105],"an":[106],"method":[108],"(PGS-check)":[109],"estimate":[111],"relative":[113,260],"decay":[114,123],"expected":[116,152],"variance":[117,177],"PGS":[119,163,179,269],"individuals":[121,181],"accuracy":[125],"(R2)":[126],"associated":[127,186],"with":[128,294],"missingness.":[130],"further":[132],"implemented":[133],"PGS-impute":[135,221,285],"algorithm":[136],"actively":[138],"signal":[141],"via":[142],"linkage":[143],"disequilibrium":[144],"(LD)-based":[145],"weight":[146],"redistribution.":[147],"Finally,":[148],"portability":[153],"78":[155,240],"published":[156,241],"coronary":[157],"artery":[158],"models":[160],"Catalog":[164],"common":[166],"platforms.":[168,311],"Results":[169],"Empirical":[170],"analyses":[171],"demonstrated":[172],"that":[173,214,314],"shrinks":[175],"significantly":[182],"faster":[183],"than":[184],"decrease":[187],"accuracy,":[190],"leading":[191],"miscalibrated":[193],"estimates":[195,317],"pre-defined":[199],"models.":[201],"PGS-check":[202,283],"accurately":[203],"estimated":[204],"these":[205],"losses,":[206],"allowing":[207],"\\"passive\\"":[210],"mathematical":[211],"rescaling":[212],"approach":[213],"restores":[215],"score":[216,273],"calibration.":[217],"For":[218],"active":[219],"recovery,":[220],"successfully":[222],"redistributed":[223],"weights":[224],"high-quality":[226],"proxies,":[227],"recovering":[228],"median":[230],"71.4%":[232],"R2.":[235],"When":[236],"scores,":[242],"identified":[244],"substantial":[245],"attrition":[247],"during":[248],"initial":[249],"mapping,":[250],"remaining":[252],"variants":[253],"appeared":[254],"well-captured":[255],"panels":[258],"(median":[259],"R2":[261],"=":[262],"99.5%).":[263],"Discussion":[264],"Variant":[265],"primarily":[267],"distorts":[268],"calibration":[270],"compressing":[272],"distributions,":[274],"degradation":[277],"correctable.":[279],"The":[280],"integration":[281],"provides":[290],"research":[292,334],"community":[293],"scalable":[296],"solution":[297],"robust,":[300],"well-calibrated,":[301],"equitable":[303],"application":[304],"polygenic":[306],"disparate":[309],"technological":[310],"By":[312],"ensuring":[313],"remain":[318],"accurate":[319],"regardless":[320],"array":[324],"used,":[325],"facilitates":[328],"movement":[330],"genetics":[332],"reliable":[337],"instruments.":[339]}

Authors

Institutions

Publication Details

Journal
European Neuropsychopharmacology
Published
2026-09-21
DOI
https://doi.org/10.1016/j.euroneuro.2026.113094
Primary Topic
Genetic Associations and Epidemiology
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
article

67. MIND THE GAP: QUANTIFYING AND MITIGATING THE IMPACT OF VARIANT MISSINGNESS ON POLYGENIC SCORES WITHIN THE GENOPRED PIPELINE

Oliver Pain, Naomi Wray, Xiaotong Wang, Peter Visscher
European Neuropsychopharmacology
Genetic Associations and Epidemiology
article

67. MIND THE GAP: QUANTIFYING AND MITIGATING THE IMPACT OF VARIANT MISSINGNESS ON POLYGENIC SCORES WITHIN THE GENOPRED PIPELINE

Oliver Pain, Naomi Wray, Xiaotong Wang, Peter Visscher
article en

Abstract

Background Polygenic scores (PGS) are powerful tools for disease risk stratification, but their translational utility is challenged by variant missingness when applied across diverse genotyping arrays and imputation panels. Discrepancies in variant coverage between discovery and target samples can degrade predictive discrimination and distort clinical calibration, yet this issue is often ignored in standard analytical workflows. We introduce a dual framework integrated into the open-source GenoPred pipeline to quantify and recover lost genetic signal. Methods We systematically evaluated the impact of variant loss using whole-genome sequencing (WGS) data from the UK Biobank as a ground truth. To address missingness without requiring individual-level target data, we developed an analytical method (PGS-check) to estimate the relative decay in expected variance of PGS across individuals and decay predictive accuracy (R2) associated with the missingness. We further implemented the PGS-impute algorithm to actively recover lost signal via linkage disequilibrium (LD)-based weight redistribution. Finally, we evaluated the expected portability of 78 published coronary artery disease models from the PGS Catalog across common genotyping platforms. Results Empirical analyses demonstrated that missingness shrinks the variance of PGS across individuals significantly faster than the associated decrease in predictive accuracy, leading to miscalibrated risk estimates when applied to pre-defined clinical models. PGS-check accurately estimated these losses, allowing for a "passive" mathematical rescaling approach that restores score calibration. For active recovery, PGS-impute successfully redistributed weights to high-quality proxies, recovering a median of 71.4% of lost R2. When applied to the 78 published scores, we identified substantial variant attrition during initial mapping, but remaining variants appeared well-captured across standard panels (median relative R2 = 99.5%). Discussion Variant missingness primarily distorts PGS calibration by compressing score distributions, but this degradation is correctable. The integration of PGS-check and PGS-impute into the GenoPred pipeline provides the research community with a scalable solution for the robust, well-calibrated, and equitable application of polygenic scores across disparate technological platforms. By ensuring that genetic risk estimates remain accurate regardless of the genotyping array used, this framework facilitates the movement of genetics from research tools to reliable clinical instruments.

European NeuropsychopharmacologyVol. 111
King's College London (GB), University of Oxford (GB)
Reduced inequalities
Openalex Percentile: Top 11%
Genetic Associations and Epidemiology
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.