Epidemiological studies common analysis is that the correlation analysis.
Correlation analysis can somehow provide us with a number of research ideas, such as the lack elements associated with certain cancers A, then we can reduce cancer rates by supplementary elements A. This conclusion premise is the lack of an element A cause this cancer, which means that elements A and cancer causation.
But in fact, the elements A and cancer correlation, does not mean there is a causal relationship between them . It is also possible that people suffering from cancer at the same time there are other complications, which can lead to complications element A deficiency.
As another example, studies have shown that big chest girls do not love sports related. So, in the end because the big chest women do not love sports, or because they do not love sports led to a large chest (obesity).
If you do not do other analysis, look at this correlation, we can not know whether there is a real causal relationship between these two characterization.
In order to clarify whether there is a causal relationship characterized, we generally consider in the analysis randomized study added Mendel.
1 thought Mendelian randomization analysis
Or the previous example, suppose we consider a large chest of people (because) unwilling movement (fruit).
The idea is that our study found significant genetic loci associated with large chest, and then analyze whether these genetic loci are also associated with not love sports.
If the chest associated with large genetic loci also related to not love sports, then we will think big chest of people just do not love sports.
Conversely, if associated with a large chest loci nothing to do with not love sports, so we think the big chest with not love sports is no causal relationship, there is a correlation may be caused by other factors between them.
2 Why Mendelian randomized study genetic loci as media analysis
The reason is simple, we are born genetic locus of the decision, the latter will not change with cultural, economic and other factors.
If we find sites A large chest relevant, then there is no doubt, is the site of A on the chest size.
On the other hand does not hold, because the big chest, so genetic locus becomes A, such a conclusion may sound very weird.
3 Mendelian randomization principle Calculation
3.1 find SNP loci affecting a large chest
The most popular way is to genome-wide association analyzes.
Get your cup size by different individuals (phenotype) and their genetic loci (genotype), do genome-wide association study .
3.2 Effects signal significantly large chest site as the site for subsequent analysis
Our definition of a site of significant signal is a P-value <5 * 10-8
3.3 calculate the relationship between big breasts significant signal sites and do not love sports
You can also use a method of genome-wide association study .
Then get big chest significantly effect the value of the signal sites do not love sports (BETA) or value at risk (OR)
3.4 draw significant signal scatter plot sites are in effect a large chest and value do not love sports, a scatter plot of the slope of the Congress of the chest do not love sports affect the size of this behavior (causal effect)
4 how do Mendelian randomization analysis
Here is an example of simple small.
4.1 Installation Package R
install.packages("devtools")
library(devtools)
install_github("MRCIEU/TwoSampleMR")
4.2 introducing large chest (Exposure) significant signals site
library(TwoSampleMR)
bmi_file <- system.file("data/bmi.txt", package="TwoSampleMR")
exposure_dat <- read_exposure_data(bmi_file)
Significantly large chest site data signal format is as follows:
4.3 introducing significant signal loci sedentary correlation data
outcome_dat <- read_outcome_data(snps = exposure_dat$SNP,filename = "F:/download/test.csv", sep = ",",snp_col = "SNP",beta_col = "beta",se_col = "se",effect_allele_col = "effect_allele",other_allele_col = "other_allele",eaf_col = "eaf",samplesize_col = "samplesize")
Do not like sports site data signal format test.csv follows:
Introducing significant signal sites associated with sedentary outcome_dat following data:
4.4 Unified big chest and do not love sports value orientation effect
dat <- harmonise_data(exposure_dat, outcome_dat)
4.5 Mendelian randomization analysis
res <- mr(dat)
The results are not significant, indicating that the two are correlated, but no causal relationship.
4.6 Scatter painting
p1 <- mr_scatter_plot(res, dat)
p1[[1]]
4.7 Other supplementary analysis
4.7.1 Heterogeneity statistics
mr_heterogeneity(dat)
4.7.2 Horizontal pleiotropy
mr_pleiotropy_test(dat)
4.7.3 Single SNP analysis
res_single <- mr_singlesnp(dat)
4.7.4 Leave-one-out analysis
res_loo <- mr_leaveoneout(dat)
4.7.5 Forest plot
res_single <- mr_singlesnp(dat)
p2 <- mr_forest_plot(res_single)
p2[[1]]
4.7.6 Leave-one-out plot
res_loo <- mr_leaveoneout(dat)
p3 <- mr_leaveoneout_plot(res_loo)
p3[[1]]
4.7.7 Funnel plot
res_single <- mr_singlesnp(dat)
p4 <- mr_funnel_plot(res_single)
p4[[1]]
Research on Mendelian randomization talked about this, for more details see the contents: https://mrcieu.github.io/TwoSampleMR/ . In addition, the relationship mentioned herein big chest and exercise, not mine had imagined, someone really did a study on this article interesting, please go to the public micro-channel number "bio biological information" backstage reply "big chest" : D