In the light of mean heritability estimates of 49% across 17,804 traits that derive from 14,558,903 twin pairs of different cultures (Polderman et al., Reference Polderman, Benyamin, de Leeuw, Sullivan, van Bochoven, Visscher and Posthuma2015), we argue that the question is not whether but rather how to integrate genetic data to advance our understanding of human psychology. Owing to the unprecedented advances of massive parallel sequencing, large-scale genome-wide association studies (GWASs) have become increasingly accessible and affordable in social sciences. Accordingly, the predictive power of polygenic scores (PGSs) is steadily rising proportional to the GWAS sample size (Mitchell et al., Reference Mitchell, Thorp, Wu, Campos, Nyholt, Gordon and Byrne2021) and can already explain a substantial amount of variance in complex phenotypes such as educational attainment (~12–16%, Okbay et al., Reference Okbay, Wu, Wang, Jayashankar, Bennett, Nehzati and LifeLines Cohort2022) or externalizing traits (~10%, Karlsson Linnér et al., Reference Karlsson Linnér, Mallard, Barr, Sanchez-Roige, Madole and Driver2021). Moreover, studies applying a multi-PGS approach suggest that predictive accuracy for a given outcome can be further improved by combining PGSs of different traits (Allegrini et al., Reference Allegrini, Selzam, Rimfeld, von Stumm, Pingault and Plomin2019; Krapohl et al., Reference Krapohl, Patel, Newhouse, Curtis, von Stumm, Dale and Plomin2018). Effect sizes of some single, well-performing PGSs are already comparable to those achieved by conventional measures used in social sciences such as family characteristics (Derzon, Reference Derzon2010) and neighborhood disadvantage (Winslow & Shaw, Reference Winslow and Shaw2007). This upward trend is expected to continue because of the steady progress in discovering rare genetic variants underlying complex trait heritability that are still insufficiently tagged by current GWASs (Dou et al., Reference Dou, Wu, Ding, Wang, Jiang, Chai and Wang2021). Estimates from large whole-genome sequencing data sets identified rare variants as a major source of the discrepancy between single-nucleotide polymorphism (SNP)-based and pedigree estimates of heritability for complex, polygenic traits such as height (Wainschtein et al., Reference Wainschtein, Jain, Zheng, Aslibekyan, Becker and Bi2022). In contrast, the frequently discussed concern that PGSs of complex traits are doomed to miss a substantial amount of non-additive variance is currently not well supported. Instead, average estimates from large samples of unrelated individuals suggest that dominance effects explain at most a very small amount of variance in complex traits (Hivert et al., Reference Hivert, Sidorenko, Rohart, Goddard, Yang, Wray and Visscher2021; Okbay et al., Reference Okbay, Wu, Wang, Jayashankar, Bennett, Nehzati and LifeLines Cohort2022). Consequently, it is only legitimate to assume that PGSs are just about to unfold their full predictive potential.

Burt is further concerned that PGSs are inevitably compromised by environmental confounding, whereas others argue that traditional environmental measures, for example childhood maltreatment, are also confounded by substantial heritable components (Dalvie et al., Reference Dalvie, Maihofer, Coleman, Bradley, Breen, Brick and Nievergelt2020; Hart, Little, & van Bergen, Reference Hart, Little and van Bergen2021; Warrier et al., Reference Warrier, Kwong, Luo, Dalvie, Croft, Sallis and Cecil2021). As Burt rightly cautions, quantifying the extent by which the predictive power of PGSs results from genotype–environment correlation (rGE) is challenging, but indeed essential for their adequate interpretation. Large within-family studies (e.g., using parent–offspring trios) have significantly contributed to more precise estimates of rGE (Chen et al., Reference Chen, Lu, Lundström, Larsson, Lichtenstein and Pettersson2022) and could be further advanced through developmental approaches starting from infancy when environmental variance is still reduced (Falck-Ytter et al., Reference Falck-Ytter, Hamrefors, Siqueiros Sanches, Portugal, Taylor, Li and Ronald2021). Importantly, however, disentangling direct from indirect genetic effect of PGS is less relevant whenever the primary goal is to improve (risk) prediction accuracy, given that rGE does not undermine a PGS's predictive capacities (see Plomin & von Stumm, Reference Plomin and von Stumm2022). Moreover, even those PGSs where a relatively large amount of predictive power is not derived from direct genetic effects (e.g., educational attainment, Okbay et al., Reference Okbay, Wu, Wang, Jayashankar, Bennett, Nehzati and LifeLines Cohort2022) still capture variance that is substantially independent of and thus incremental to the effects of traditional environmental measures, like socioeconomic status (Judd et al., Reference Judd, Sauce, Wiedenhoeft, Tromp, Chaarani, Schliep and Klingberg2020). Consequently, algorithms that jointly model the effects of PGSs and environmental measures are performing significantly better in predicting, for example, health outcomes (Adeyemo et al., Reference Adeyemo, Balaconis, Darnes, Fatumo, Granados Moreno and Hodonsky2021; Martikainen et al., Reference Martikainen, Korhonen, Jelenkovic, Lahtinen, Havulinna, Ripatti and Silventoinen2021; Østergaard et al., Reference Østergaard, Trabjerg, Als, Climent, Privé, Vilhjálmsson and Agerbo2020) and cognitive functioning (von Stumm et al., Reference von Stumm, Smith-Woolley, Ayorech, McMillan, Rimfeld, Dale and Plomin2020) compared to those that include traditional non-genetic measures only. The potential to improve prediction by combining genetic and environmental data even translates down to epigenetic modifications, which are increasingly recognized in social science studies as a potential mechanism of how life events get under the skin. Epigenome-wide analyses across independent cohorts revealed that variation in DNA methylation is best explained by additive effects and the interaction of genes and environmental forces, but almost never by environmental adversity alone (Czamara et al., Reference Czamara, Tissink, Tuhkanen, Martins, Awaloff, Drake and Binder2021).

Beyond our defense of the immediate practical utility of PGSs for maximizing trait prediction, we also do not share Burt's skepticism regarding the limited potential of PGSs for advancing our etiological understanding of complex traits. The growing number of studies combining PGSs with neuroimaging, proteomic, or other multi-omic data have already provided unique insights into specific mechanisms through which polygenic predispositions exert their effects on complex phenotypes. Exemplary findings from neuroimaging studies include the identification of structural brain changes associated with PGSs for neuroticism (Opel et al., Reference Opel, Amare, Redlich, Repple, Kaehler, Grotegerd and Dannlowski2020) and educational attainment (Elliott et al., Reference Elliott, Belsky, Anderson, Corcoran, Ge, Knodt and Hariri2019), that, in the latter example, partly mediated the association between participants' PGS and their cognitive test performance. Moreover, PGSs have already been successfully applied to study the causal biology of complex traits, for example, in terms of identifying specific proteins underlying cardiometabolic diseases (Ritchie et al., Reference Ritchie, Lambert, Arnold, Teo, Lim, Scepanovic and Inouye2021). Another promising new method to advance etiological understanding of complex traits is to construct PGSs based on gene transcription profiles targeting specific biological systems, including PGSs capturing neurotransmitter signaling pathways (Miguel et al., Reference Miguel, Pereira, Barth, de Mendonça Filho, Pokhvisneva, Nguyen and Silveira2019; Restrepo-Lozano et al., Reference Restrepo-Lozano, Pokhvisneva, Wang, Patel, Meaney, Silveira and Flores2022), immuno-metabolic markers (Kappelmann et al., Reference Kappelmann, Czamara, Rost, Moser, Schmoll, Trastulla and Arloth2021), or cellular stress responses (Arloth et al., Reference Arloth, Bogdan, Weber, Frishman, Menke, Wagner and Binder2015). For example, a recent study reported that a PGS based on corticolimbic-specific DCC gene co-expression, which modulates maturation of dopamine networks, is a better predictor of impulsivity-related phenotypes than conventional PGSs (Restrepo-Lozano et al., Reference Restrepo-Lozano, Pokhvisneva, Wang, Patel, Meaney, Silveira and Flores2022).

To conclude, we argue that despite their indisputable limitations, PGSs hold great potential for both better prediction and understanding of complex traits in social science. Raw SNP data from genome-wide arrays can now be generated for only ~US$35 per individual test with an excellent accuracy that outcompetes those of most environmental measures (genotype concordance >98%, Hong et al., Reference Hong, Xu, Liu, Jones, Su, Ning and Shi2012). Once obtained, SNP data allow for an automated generation and flexible adaption of multiple PGSs at any time in life because of their inherent intraindividual stability. The initial struggle of identifying causal genetic variants for complex traits should not discourage us from embracing the remarkable achievements recently made in molecular behavior genetics.