calculate p value and associated z score for snp-gene pair

Question

[rheab1230]

Hello everyone, I have genotype file and gene expression file. I want to see whether three of my rsid in genotype file is associated with gene expression. I used matrixeqtl R package to generate p value and see whether the selected snp is associated with the gene expression. I used this code:

me = Matrix_eQTL_engine(
  snps = snps,
  gene = gene,
  cvrt = cvrt,
  output_file_name = output_file_name,
  pvOutputThreshold = pvOutputThreshold,
  useModel = useModel,
  errorCovariance = errorCovariance,
  verbose = TRUE,
  pvalue.hist = TRUE,
  min.pv.by.genesnp = FALSE,
  noFDRsaveMemory = FALSE)


unlink(output_file_name)
show(me$all$eqtls)
output: 
Processing covariates
Task finished in 0.004 seconds
Processing gene expression data (imputation, residualization)
Task finished in 0.004 seconds
Creating output file(s)
Task finished in 0.022 seconds
Performing eQTL analysis
100.00% done, 0 eQTLs
No significant associations were found.

I looked at the snps and they have effect for this gene in gtex models but during matrixeqtl analysis it doesn't show any association.

I don't want to focus on eqtl but rather just generate pvalue and associated z score for the snp-gene pair. Is there any software or tool to do that?

Answer 1

[LChart]

The simplest thing to do would be to 'force' matrixeQTL to output everything by setting pvOutputThreshold to 1; this should give you all association statistics between all snps in snp and all genes in gene by saving them to the file $output_file_name

[rheab1230]

Thank you so much for this. Yes, in this case I get all the SNPs. I also have one more question:

SNP gene    beta    t-stat  p-value FDR
rs12160750  ENSG00000100116.16  0.191086848260097   2.78574185437024    0.00555828233127971 0.286529082199489
rs2285177   ENSG00000100116.16  0.187323375911886   2.72992752554224    0.00657473357007616 0.286529082199489

In this results: a lower p value less than 0.05 means the snp-gene pair is significant. What does the rest of the score implies? like t-stat,FDR value? Thank you.

[LChart]

Beta is the effect size (for rs12160750: going from 0 alleles to 1 allele raises the expression by 0.19108 on average). t-stat is the "z-statistic" you were looking for in your OP, but it really is a t-stat rather than a z-stat, though for high degrees of freedom the distributions are indistinguishable. FDR is the multiple-testing adjusted p-value using the false discovery rate method -- and this is the column you should use to determine significance, rather than p-value.

[rheab1230]

Okay, I got it. Thank you so much for the response. Can you tell me what should be the range of FDR value to consider for significance?

[LChart]

I think researchers would agree that FDR > 0.05 is not significant, though some would argue FDR between 0.05 and 0.1 is OK for an initial finding.

An informal survey I performed of bioinformatics post-docs had about 40% saying FDR < 0.05 is significant, 40% saying FDR < 0.01 is significant, and 20% were convinced by a then-recent article arguing for FDR < 0.005. All of these positions are defensible.

[rheab1230]

Thank you so much for the information.