King's College London

Research portal

Testing the impact of trait prevalence priors in Bayesian-based genetic prediction modeling of human appearance traits

Research output: Contribution to journalArticlepeer-review

Maria Alexandra Katsara, Wojciech Branicki, Ewelina Pośpiech, Pirro Hysi, Susan Walsh, Manfred Kayser, Michael Nothnagel

Original languageEnglish
Article number102412
JournalForensic Science International: Genetics
Volume50
DOIs
PublishedJan 2021

Bibliographical note

Funding Information: This study received support from the European Union’s Horizon 2020 Research and Innovation programme under grant agreement No 740580 within the framework of the Visible Attributes through Genomics (VISAGE) Project and Consortium. TwinsUK is funded by the Wellcome Trust, Medical Research Council, European Union, Chronic Disease Research Foundation (CDRF) , Zoe Global Ltd and the National Institute for Health Research (NIHR) funded BioResource, Clinical Research Facility and Biomedical Research Centre based at Guy’s and St Thomas’ NHS Foundation Trust in partnership with King’s College London. The funding organization had no influence on the design, conduct or conclusions of the study. Funding Information: This study received support from the European Union's Horizon 2020 Research and Innovation programme under grant agreement No 740580 within the framework of the Visible Attributes through Genomics (VISAGE) Project and Consortium. TwinsUK is funded by the Wellcome Trust, Medical Research Council, European Union, Chronic Disease Research Foundation (CDRF), Zoe Global Ltd and the National Institute for Health Research (NIHR) funded BioResource, Clinical Research Facility and Biomedical Research Centre based at Guy's and St Thomas? NHS Foundation Trust in partnership with King's College London. The funding organization had no influence on the design, conduct or conclusions of the study. Publisher Copyright: © 2020 The Author(s) Copyright: Copyright 2021 Elsevier B.V., All rights reserved.

King's Authors

Abstract

The prediction of appearance traits by use of solely genetic information has become an established approach and a number of statistical prediction models have already been developed for this purpose. However, given limited knowledge on appearance genetics, currently available models are incomplete and do not include all causal genetic variants as predictors. Therefore such prediction models may benefit from the inclusion of additional information that acts as a proxy for this unknown genetic background. Use of priors, possibly informed by trait category prevalence values in biogeographic ancestry groups, in a Bayesian framework may thus improve the prediction accuracy of previously predicted externally visible characteristics, but has not been investigated as of yet. In this study, we assessed the impact of using trait prevalence-informed priors on the prediction performance in Bayesian models for eye, hair and skin color as well as hair structure and freckles in comparison to the respective prior-free models. Those prior-free models were either similarly defined either very close to the already established ones by using a reduced predictive marker set. However, these differences in the number of the predictive markers should not affect significantly our main outcomes. We observed that such priors often had a strong effect on the prediction performance, but to varying degrees between different traits and also different trait categories, with some categories barely showing an effect. While we found potential for improving the prediction accuracy of many of the appearance trait categories tested by using priors, our analyses also showed that misspecification of those prior values often severely diminished the accuracy compared to the respective prior-free approach. This emphasizes the importance of accurate specification of prevalence-informed priors in Bayesian prediction modeling of appearance traits. However, the existing literature knowledge on spatial prevalence is sparse for most appearance traits, including those investigated here. Due to the limitations in appearance trait prevalence knowledge, our results render the use of trait prevalence-informed priors in DNA-based appearance trait prediction currently infeasible.

View graph of relations

© 2020 King's College London | Strand | London WC2R 2LS | England | United Kingdom | Tel +44 (0)20 7836 5454