Missing heritability

These two days, Mr. Huang Sanwen's team brushed the screen （2022 year 6 month 8 Japan , Shenzhen Institute of agricultural genomics, Chinese Academy of Agricultural Sciences （“ The genome ”） Huang Sanwen's team is Nature The journal published two research papers online at the same time ）, Two important research achievements of the team in the field of plant genomics are reported , It provides a new solution for the application of Pan genome in crop breeding . One of them 《Graph pangenome captures missing heritability and empowers tomato breeding》 This paper introduces the pan genome through graph （Graph Pan-genome） It explains the important problems of quantitative genetics ： Lost heritability .

Here is the concept , And give my understanding .

Lost heritability

“ Loss of heritability ”（ Missing heritability） Is an important quantitative genetics problem , That is, heritability estimated by genetic markers and genome-wide association analysis （GWAS） The total heritability of all the relevant genes found was lower than the actual heritability .

The calculation method is ： The heritability obtained from quantitative genetic genetic evaluation is the real heritability , For example, the heritability of height is 0.8, however GWAS Significant in the analysis SNP Only explained 45% Variation of （ The heritability is 0.45）, There are still 0.35 The heritability of is lost , This is called lost heritability .

Concept segmentation

Heritability , More general heritability is not narrow heritability or broad heritability , It is ：

The first one is ：h2 family ：

$h_{family}^2$, Twin studies , By comparing the similarities between identical and fraternal twins , To calculate the h2, Usually the highest of the three .

Brother Fei understands ： This is equivalent to adding the repetition of common environmental effects

The second kind ：h2 SNP

$h_{SNP}^2$,GWAS Used in the study chip On all the variants Jointly explained variance And Ratio of trait variance , Than h2 family low , But it will be significantly higher than h2 GWAS.

Brother Fei understands ： This is equivalent to GBLUP The percentage of the additive variance component in , That is, heritability in a narrow sense .

The third kind of ：h2 GWAS

$h_{GWAS}^2$, Only by GWAS Found to be related to a disease variants Explained variance And Ratio of trait variance , The lowest of the three .

Brother Fei understands ： This is quite remarkable SNP Percentage of ,PVE Value

The relationship between the above three is

$$h_{GWAS}^2<h_{SNP}^2<h_{family}^2$$

What we call vanishing heritability （Missing Heritability） Namely ：

$$h_{Missing}^2=h_{family}^2-h_{GWAS}^2$$

The main reason for the loss of heritability ： Non additive genetic effects , Rare variation with large effect （rare variants）, Or overestimation caused by common environmental factors in twin studies .

Vanishing heritability can be divided into ：

• Still lost heritability （still missing heritability）
• Hidden heritability （hidden heritability）

Still lost heritability （still missing heritability）：
$$h_{stillmissing}^2=h_{family}^2-h_{SNP}^2$$

Hidden heritability （hidden heritability） The calculation method of ：
$$h_{hidden}^2=h_{SNP}^2-h_{GWAS}^2$$

stay GWAS under advisement , Due to the high or low of the significance threshold we selected , Heritability may not be lost (missing) It's hidden （ hidden ） 了 . Another possibility is , The heterogeneity of the population (heterogeneity), because h2 GWAS Most of them come from people with multiple groups meta analysis , The heterogeneity of genetic effects in these populations may also make h2 GWAS Low .

Brother Fei understands ： In breeding , Estimated heritability of the family （ Identical twins 、 Fraternal twins ） Rarely used , We should be interested in animal models （GBLUP） Estimated heritability and GWAS remarkable SNP The difference in estimated heritability , That is ： Hidden heritability （hidden heritability）, Be able to find out what causes the difference , For us to implement molecular marker assisted （MAS）, For example, significant SNP、INDEL, Other variations , And the implementation of genome-wide selection （GS）, Consider these factors , Put it in the model to improve GS The accuracy of the estimate .

Feige's speech

Graph pangenome Proposal and application of , The use of multiomics information to improve breeding efficiency has been put into practice ！ There is machine learning on pure algorithm 、 neural network 、 Reinforcement learning , From a purely biological point of view, there are genomes 、 Transcriptome 、 Proteome , But the proposal and application of the map pan genome , Feeling is the best way out . For the practitioners of breeding data analysis （ For example, I ）, It is necessary to study and master this method ！

Reference resources ：

《An Introduction to Statistical Genetic Data Analysis》
https://zhuanlan.zhihu.com/p/362604272