More hereditarians need to take articles like these seriously imo. I’ve seen lots of sneering but no attempts at a constructive “rebuttal” (not that this should be a goal but you understand) and this is coming from someone who is biased against an env model.
"The 'genome-wide complex trait analysis (GCTA)' method (Yang, Lee, Goddard, Visscher) has appeared to partially resolve the 'missing heritability' problem: that the sum of GWA-identified SNPs explain only a small fraction of heritability. It estimates the variance explained by a constellation of common SNPs from the whole genome for a complex trait, rather than testing the association of any particular SNP to the trait. Using the PGC sample, it was estimated that SNPs account for 23% and 25% of variation in liability to SZ (Lee et al., 2012b) and BD (Cross-Disorder Group of the Psychiatric Genomics, 2013), respectively. They also estimate that 1) this is mainly due to common causal alleles, 2) they must be evenly spread across chromosomes since the variance explained by each chromosome is linearly related to its length, 3) the genetic basis of SZ is the same in males and females and 4) as expected, a disproportionate amount of variation in liability is attributable to a set of 2725 genes expressed in the CNS. Furthermore, using only unrelated subjects and the same SNP genotypes, a 68% genetic correlation between these disorders was found. Although most of the SNPs responsible for the variance explained are not yet identified, the rationale is that they will be, as GWAS sample sizes increase and more accurate estimation of the effect size of each SNP is achieved."
Prata, D. P., Costa-Neves, B., Cosme, G., & Vassos, E. (2019). Unravelling the genetic basis of schizophrenia and bipolar disorder with GWAS: A systematic review. Journal of psychiatric research, 114, 178–207. https://doi.org/10.1016/j.jpsychires.2019.04.007
This is referring to a different "missing heritability problem", which is the gap between the variance explained by the specific significant SNPs that have been identified (something like 8% for SCZ) and the variance explained by all common SNPs using GCTA-like methods (23-25%). And the statement is entirely correct, the total variance explained by all SNPs in a GCTA-like approach is much larger than what can be explained by the individually significant associations in a given study. But this doesn't get at the gap between GCTA-like estimates and estimates from twins.
Another very informative post Sasha, but I have a question about your Figure 1. It is not clear to me what "Unmeasured Cause" represents. Only genotype and environment can affect phenotype, and so it is either unmeasured genotype, unmeasured environment, or unmeasured genotype-environment interaction.
Thanks Tau-Mu! And that's right, the unmeasured cause can be some genotype or some environment that is not explicitly being considered in the model. For example, if we are looking at a GxE interaction with BMI (environment) on lung cancer (outcome), but there is an unmeasured cause in smoking where: (1) smoking causes lower BMI, (2) smoking causes lung cancer, (3) BMI does not impact lung cancer. Then (if we did not put smoking in our model) we would actually be estimating the effect moderation of BMI on lung cancer (e.g. acting on BMI would not change lung cancer).
More hereditarians need to take articles like these seriously imo. I’ve seen lots of sneering but no attempts at a constructive “rebuttal” (not that this should be a goal but you understand) and this is coming from someone who is biased against an env model.
What do you think of this?
"The 'genome-wide complex trait analysis (GCTA)' method (Yang, Lee, Goddard, Visscher) has appeared to partially resolve the 'missing heritability' problem: that the sum of GWA-identified SNPs explain only a small fraction of heritability. It estimates the variance explained by a constellation of common SNPs from the whole genome for a complex trait, rather than testing the association of any particular SNP to the trait. Using the PGC sample, it was estimated that SNPs account for 23% and 25% of variation in liability to SZ (Lee et al., 2012b) and BD (Cross-Disorder Group of the Psychiatric Genomics, 2013), respectively. They also estimate that 1) this is mainly due to common causal alleles, 2) they must be evenly spread across chromosomes since the variance explained by each chromosome is linearly related to its length, 3) the genetic basis of SZ is the same in males and females and 4) as expected, a disproportionate amount of variation in liability is attributable to a set of 2725 genes expressed in the CNS. Furthermore, using only unrelated subjects and the same SNP genotypes, a 68% genetic correlation between these disorders was found. Although most of the SNPs responsible for the variance explained are not yet identified, the rationale is that they will be, as GWAS sample sizes increase and more accurate estimation of the effect size of each SNP is achieved."
Prata, D. P., Costa-Neves, B., Cosme, G., & Vassos, E. (2019). Unravelling the genetic basis of schizophrenia and bipolar disorder with GWAS: A systematic review. Journal of psychiatric research, 114, 178–207. https://doi.org/10.1016/j.jpsychires.2019.04.007
This is referring to a different "missing heritability problem", which is the gap between the variance explained by the specific significant SNPs that have been identified (something like 8% for SCZ) and the variance explained by all common SNPs using GCTA-like methods (23-25%). And the statement is entirely correct, the total variance explained by all SNPs in a GCTA-like approach is much larger than what can be explained by the individually significant associations in a given study. But this doesn't get at the gap between GCTA-like estimates and estimates from twins.
Another lesson learned!
Another very informative post Sasha, but I have a question about your Figure 1. It is not clear to me what "Unmeasured Cause" represents. Only genotype and environment can affect phenotype, and so it is either unmeasured genotype, unmeasured environment, or unmeasured genotype-environment interaction.
Thanks Tau-Mu! And that's right, the unmeasured cause can be some genotype or some environment that is not explicitly being considered in the model. For example, if we are looking at a GxE interaction with BMI (environment) on lung cancer (outcome), but there is an unmeasured cause in smoking where: (1) smoking causes lower BMI, (2) smoking causes lung cancer, (3) BMI does not impact lung cancer. Then (if we did not put smoking in our model) we would actually be estimating the effect moderation of BMI on lung cancer (e.g. acting on BMI would not change lung cancer).