Table 3 Probabilities describing the relationship between flashfm ranks of causal variants when the trait correlation is mis-specified.

From: The flashfm approach for fine-mapping multiple quantitative traits

 

Trait 1 (E+G)

 

Pr(matched ranks)

Pr(matched or improved ranks)

Trait correlation shift

rs1980422/E

rs3087243/G

rs1980422/E

rs3087243/G

−0.2

0.870

0.923

0.960

0.950

−0.1

0.903

0.950

0.970

0.967

    0.1

0.897

0.950

0.933

0.983

    0.2

0.793

0.900

0.857

0.947

 

Trait 2 (E+H)

 

Pr(ranks match)

Pr(matched or improved ranks)

Trait correlation shift

rs1980422/E

rs231775/H

rs1980422/E

rs231775/H

−0.2

0.850

0.913

0.940

0.977

−0.1

0.893

0.937

0.953

0.987

    0.1

0.870

0.960

0.920

0.973

    0.2

0.760

0.897

0.837

0.927

  1. Two traits were simulated to have causal variants E+G and E+H and trait correlation 0.4; sample size is N = 3000. Comparisons are made between flashfm results using the estimated trait correlation as input and flashfm results with this trait correlation estimate shifted upwards/downwards by 0.1 or 0.2. The region has 1231 SNPS and was simulated to mimic the LD structure of the CTLA4 region, 2q-204446258-204816382 (GRCh37/hg19). Results are based on 300 replications.
Back to article page