BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//ox.ac.uk//NONSGML oxford.event//EN
X-WR-TIMEZONE:Europe/London
BEGIN:VTIMEZONE
TZID:Europe/London
X-LIC-LOCATION:Europe/London
BEGIN:DAYLIGHT
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:BST
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:GMT
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
SUMMARY:Personalized polygenic risk prediction and assessment with a Mixtu
 re-of-Experts framework
DTSTART;TZID=Europe/London:20260512T093000
DTEND;TZID=Europe/London:20260512T103000
DTSTAMP:20260512T052608Z
UID:62506112-ab43-f111-bec7-7c1e52046848
CREATED:20260429T090912Z
DESCRIPTION:For our next talk\, in the BDI/CHG (gen)omics Seminar series\,
  we will be hearing from Shadi Zabad\, SMARTbiomed postdoctoral research f
 ellow\, Department of Statistics\, University of Oxford. We’re delighted
  to host Shadi in what promises to be a great talk!\n\nDate: Tuesday 12 Ma
 y\nTime: 9:30 am – 10:30 am\nTalk title: Personalized polygenic risk pre
 diction and assessment with a Mixture-of-Experts framework\nLocation: Big 
 Data Institute\, Seminar Room 0\n\nShort bio: Shadi Zabad is a SMARTbiomed
  postdoctoral research fellow at the Department of Statistics\, University
  of Oxford. His research is focused on developing methods for complex trai
 t analysis and prediction\, incorporating aspects of Bayesian statistics a
 nd computational techniques to scale inference algorithms to modern bioban
 k datasets.\n\nAbstract:\nWith the increasing availability of high-quality
  genomic data from diverse cohorts\, polygenic scores (PRSs) have become a
  mainstay of genetic analyses of complex traits and diseases. Despite thei
 r proliferation in numerous research domains\, a major obstacle to wider a
 doption in clinical settings has been the well-established heterogeneity i
 n prediction accuracy across a variety of demographic variables\, such as 
 age\, sex\, and genetic ancestry. To address this deficiency\, recent rese
 arch efforts aimed to improve representation in genetic studies and develo
 p stratified PRS inference methods that greatly enhanced accuracy in minor
 ity populations. However\, with these stratified scores in hand\, it remai
 ns unclear how to assign the best score\, or mixture of scores\, for a par
 ticular test individual in the clinic. To bridge this gap\, we present MoE
 PRS\, an ensemble learning method based on the Mixture-of-Experts framewor
 k\, that blends the stratified scores using personalized mixing weights to
  predict the target phenotype. In biobank-scale analyses of 23 complex tra
 its and diseases in the UK and CARTaGENE biobanks\, we show that MoEPRS ge
 nerally provides modest improvements in prediction accuracy over single-so
 urce PRS models\, and its predictive performance is maintained across biob
 anks. Furthermore\, we demonstrate practical use cases where the model aut
 omatically identifies and adapts to diverse sources of heterogeneity in th
 e data\, which allows for evaluating the strengths and weaknesses of curre
 nt polygenic scores across various cohort strata.
LAST-MODIFIED:20260429T091157Z
LOCATION:Big Data Institute - Lower Ground Seminar Room 0\, Lower Ground S
 eminar Room 0 Big Data Institute Old Road Campus Oxford Oxfordshire OX3 7L
 F United Kingdom
SPEAKER:Shadi Zabad (Department of Statistics)
END:VEVENT
END:VCALENDAR