Contents

  1. The Problem
  2. Altering GAM models
  3. Results
  4. More mu estimates
  5. Switching alleles problem

1. The problem



2. Altering GAM models


Figure: Claimed and corrected coverage for simulations using logit(covered)~claimed (left) or logit(covered)~logit(claimed) (right) GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. The true mu value was used.

Figure: Claimed and corrected coverage for simulations using logit(covered)~claimed (left) or logit(covered)~logit(claimed) (right) GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. The true mu value was used.


Faceting by OR and maxz0


Figure: Claimed and corrected coverage for simulations using logit(covered)~claimed.cov GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. The true mu value was used. Faceted by OR and maxz0.

Figure: Claimed and corrected coverage for simulations using logit(covered)~claimed.cov GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. The true mu value was used. Faceted by OR and maxz0.

Figure: Claimed and corrected coverage for simulations using logit(covered)~logit(claimed) GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. The true mu value was used. Faceted by OR and maxz0.

Figure: Claimed and corrected coverage for simulations using logit(covered)~logit(claimed) GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. The true mu value was used. Faceted by OR and maxz0.

  • Green is better than red, showing that our method works at all OR and all maxz0.

  • Problem boils down to trying to find a representative value of \(\mu\).


3. Results


Figure: Claimed and corrected coverage for simulations using logit(covered)~claimed (left) or logit(covered)~logit(claimed) (right) GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. mu=sum(pp*abs(z0)) was used.

Figure: Claimed and corrected coverage for simulations using logit(covered)~claimed (left) or logit(covered)~logit(claimed) (right) GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. mu=sum(pp*abs(z0)) was used.

Figure: Claimed and corrected coverage for simulations using logit(covered)~claimed (left) or logit(covered)~logit(claimed) (right) GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. mu=sum(pp*abs(z0)) was used.

Figure: Claimed and corrected coverage for simulations using logit(covered)~claimed (left) or logit(covered)~logit(claimed) (right) GAM models. mean(covered) used if the GAM model cannot be fitted for that SNP considered causal. The number of simulated credible sets for each SNP considered as causal is 10000. mu=sum(pp*abs(z0)) was used.


4. More mu estimates


Key (for some - can explain the rest):

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'




5. Switching alleles problem


LD1 LD2 LD3 LD4
< -0.5 0.13% 0.26% 0.40% 0.80%
> 0.5 0.76% 1.99% 2.37% 2.91%

Questions