This year at ASHG one of the most fascinating talks was Po-Ru Loh’s, where he reviewed the BOLT-REML method. It’s introduced in the paper, Contrasting genetic architectures of schizophrenia and other complex diseases using fast variance-components analysis. As you likely know many diseases such as schizophrenia manifest as complex trait; that is, they’re basically quantitative in their genetic architecture. Lots of alleles in the population, at varied frequencies (e.g., it might be low frequency and large effect, or higher frequency and smaller effect). In the abstract they state that “We also observe significant enrichment of heritability in GC-rich regions and in higher-frequency SNPs for both schizophrenia and GERA diseases.” In other words, they’re getting toward the holy grail of these sorts of studies, actually fixing upon likely loci which explain the variation.
But the genesis of these methods goes back to the late 2000s, when some statistical geneticists began to synthesis the power of genomics with classical quantitative genetic frameworks and insights. Another paper which sums up this tradition is Genetic variance estimation with imputed variants finds negligible missing heritability for human height and body mass index. That is, the authors have confirmed the classical heritability estimates for height, using inferences such as twin studies, with genomic methods. Many geneticists operating just outside this field are totally unaware of the power, precision, and rapidity in advance of this set of techniques. If so, I suggest you read A Commentary on ‘Common SNPs Explain
a Large Proportion of the Heritability for Human Height’ by Yang et al. (2010) (ungated). Here is the final paragraph:
Why have we encountered so much apparent misunderstanding of the methods and results in the human
genetics community? The core of our method is heavily steeped in the tradition of prediction of random effects and the estimation of variance due to random (latent) effects. While estimation and partitioning of variance has a long history in human genetics, in particular in twin research, the prediction of random effects is alien to many human geneticists and, surprisingly, also to statisticians (Robinson, 1991). Another reason could be the simultaneous use of population genetics and quantitative genetics concepts and theory in our paper, since these are usually applied in different applications, e.g., gene mapping or estimation of heritability. All concepts and methods that we used are extensively described in the textbooks by Falconer and Mackay (1996; chapters 1, 3, 4, 7–10) and Lynch and Walsh (1998; chapters 4, 7, 26, 27).
Please, if you read anything on this blog, read this.