An abnormal and isolated signal with PIP>0.5 in MeSuSiE

[Xi-Cao]

Dear Boran,
Thanks for your helpful tool on cross-population fine-mapping.

As shown in the figure, within the genomic region surrounding rs1936824, MeSuSiE identified three credible sets, which appeared reasonable. However, MeSuSiE additionally detected an isolated signal, rs829409, with a PIP of 0.72. This signal was not included in a credible set due to the purity threshold of 0.5. It is strange that rs829409 showed highly insignificant P-values of 0.06 and 0.01 in EUR and EAS populations, respectively. Furthermore, I examined two variants correlated with rs829409 in this region, which had maximum R^2 of 0.51 and 0.50, respectively. These variants were also found to be non-significant in two populations. Examination of the LD mismatch also did not reveal any problems.

I am confused why MeSuSiE assigned such a high PIP to rs829409, which did not appear in the population-specific fine-mappings using SuSiE. Should this be considered an unreliable finding?

Best,
xicao

[borangao]

Dear xicao,

I can not tell the reason without looking into the data. My intuition was LD mismatch issue, while you mentioned that there is no LD mismatch issue being detected. You can send me the summary statistics and LD matrix to me if you wanna me to further look into this.

Best,
Boran

[Xi-Cao]

Dear Boran,

Thank you very much for your willingness to assist.

I have sent the GWAS summary statistics and SNP correlation matrices for this region to you via the attached files. These data have undergone QC steps but have not yet been filtered for LD mismatches.
Before running fine-mappings, I excluded SNPs with LD mismatches as recommended:

eas_example <- read.table("E://fine-mapping/eas_example",header = T)
eur_example <- read.table("E://fine-mapping/eur_example",header = T)
eas_cov <- read.table("E://fine-mapping/eas_cov",header = T,row.names = 1,sep = "\t")
eur_cov <- read.table("E://fine-mapping/eur_cov",header = T,row.names = 1,sep = "\t")

eur_diagnostic <- kriging_rss(eur_example$Z, eur_cov,n=720199)
#eur_diagnostic$plot
u <- eur_diagnostic$conditional_dist
signal_u <- eur_example$SNP[eur_example$P==min(eur_example$P)]
u1 <- which(abs(u$z_std_diff)>3&abs(u$z)<2)
u2 <- which(abs(eur_cov[,signal_u])>0.8&abs(u$z)<2)
u3 <- which(abs(u$z)>2&u$logLR>2)

eas_diagnostic <- kriging_rss(eas_example$Z, eas_cov,n=255171)
#eas_diagnostic$plot
a <- eas_diagnostic$conditional_dist
signal_a <- eas_example$SNP[eas_example$P==min(eas_example$P)]
a1 <- which(abs(a$z_std_diff)>3&abs(a$z)<2)
a2 <- which(abs(eas_cov[,signal_a])>0.8&abs(a$z)<2)
a3 <- which(abs(a$z)>2&a$logLR>2)

mis <- unique(c(a1,a2,a3,u1,u2,u3))
eur_example <- eur_example[-mis,]
eas_example <- eas_example[-mis,]
eur_cov <- eur_cov[-mis,-mis]
eas_cov <- eas_cov[-mis,-mis]

In total, 24 out of 946 SNPs in this region were excluded. Then I conducted the locus zoom plot above within the remaining 922 SNPs, including rs829409.

eas_cov.txt
eas_example.txt
eur_cov.txt
eur_example.txt

Best,
xicao

[borangao]

Dear Xican,

Thank you for providing the example. I checked the example, and I believe the main factor driving the signal is rs829410 in the European population, which is in high LD with rs829409. However, when I checked the correlation in the East Asian population, it appears to be very low. This is quite unusual given that the correlation structure of SNPs between European and East Asian populations is generally similar and the MAFs are not low in both populations.

To confirm, I cross-checked the correlation using LDmatrix (https://ldlink.nih.gov/?tab=ldmatrix), and the result showed a correlation of 1 in the East Asian population. This differs significantly from the file you shared with me. Could you kindly verify your data generation process to ensure everything aligns as expected?

For cases like this, it’s often helpful to examine signals strongly correlated with the detected SNP to determine if there’s a robust marginal signal.

I hope this information is helpful.

[Xi-Cao]

Dear Boran,

Thank you very much for your kind assistance.

After reviewing my results based on your suggestion, I found that rs829410 indeed exhibits a high significance in our EUR GWAS summary statistics, with a P-value of 8.96e-12. However, I would like to clarify that rs829410 does not appear to be in high LD with rs829409 in the EUR population (as illustrated in the following figures). The LDmatrix tool gave an R² of 0.002 between these two SNPs when selecting Finnish and British populations, which aligns with our correlation matrix result of 0.009. Interestingly, we observed a contrasting pattern in the EAS population. The rs829410 is highly correlated with rs829409 here, as indicated by both the LDmatrix tool (R² = 1) and our correlation matrix (R² = 0.83). However, it is also not significant in EAS, with a P-value of 0.01. Upon further lookup of the pip_config for rs829409 (the 550th SNP in this region), we found that it is more likely to be detected as a shared signal rather than a population-specific one.

Could MeSuSiE's detection of rs829409 be driven by this peculiar relationship between the two SNPs and the combined information from both EAS and EUR populations? While rs829410 seems like a more suitable candidate？

Here is the figure of my analysis:

Here are the figures of LDmatrix for EUR and EAS populations, respectively:

Best,
xicao