Direct comparison of PIP for SuSiE, DAP, CAVIAR and FINEMAP¶

DAP, CAVIAR and FINEMAP is performed on some of the same input data and is compared with SuSiE.

%revisions -s -n 10

Plan¶

Here I consider the following scenarios:

Fix total PVE to 0.2
Scenario with growing difficulty: from 1 causal to 5 causal

For FINEMAP and CAVIAR I only try and report 1~3 causal scenarios.

The plan is to get the PIP for those in susie mappable CS (use definition $r^2$ > 0.25 as in DAP's paper to be more comparable with DAP), and compare these PIP to what DAP, CAVIAR and FINEMAP reports. For susie flavors:

use prior = 0.1 (may not matter)
use estimate_residule = TRUE (FALSE maybe a bit conservative)

This setting of susie should reflect its best performance. Additionally I check both the PIP computed before purity filter, and that after purity filter.

Expected outcome¶

PIP plots, pairwise: susie vs DAP, susie vs CAVIAR, DAP vs CAVIAR, etc

Note: DAP might seem slightly more powerful here because we only considered PIP within susie mappable CS. But as will be shown in ROC DAP also has higher false positives when all SNPs are considered.

Previously I've ran this specific DSC using:

dsc susie.dsc --target run_comparison -o susie_comparison

So here I query from that result.

Major conclusions¶

Little difference in PIP before / after filtering LD
susie with estimated residual variance is a bit anti-conservative but is acceptable. susie with residual variance being var(Y) is too conservative.

Workflow¶

Run the pip, and fdr workflow in this notebook

susie `est_var` vs DAP¶

compare PIP before / after LD filter

%preview ~/GIT/github/mvarbvs/dsc/susie_comparison/PIP_comparison_0622_estvar_true_filter_true_0p5.susie_vs_dap.png

%preview ~/GIT/github/mvarbvs/dsc/susie_comparison/PIP_comparison_0622_estvar_true_filter_false_0p5.susie_vs_dap.png

No big difference filtering LD before PIP computation.

susie `var(Y)` vs DAP¶

%preview ~/GIT/github/mvarbvs/dsc/susie_comparison/PIP_comparison_0622_estvar_false_filter_true_0p5.susie_vs_dap.png

It seems results using estimated variance might be slightly anti-conservative. But using var(Y) is conservative.

susie `est_var` vs CAVIAR¶

%preview ~/GIT/github/mvarbvs/dsc/susie_comparison/PIP_comparison_0622_estvar_true_filter_true_0p5.susie_vs_caviar.png

susie `est_var` vs FINEMAP¶

%preview ~/GIT/github/mvarbvs/dsc/susie_comparison/PIP_comparison_0622_estvar_true_filter_true_0p5.susie_vs_finemap.png

DAP vs CAVIAR¶

%preview ~/GIT/github/mvarbvs/dsc/susie_comparison/PIP_comparison_0622_estvar_true_filter_true_0p5.dap_vs_caviar.png

DAP vs FINEMAP¶

%preview ~/GIT/github/mvarbvs/dsc/susie_comparison/PIP_comparison_0622_estvar_true_filter_true_0p5.dap_vs_finemap.png

CAVIAR vs FINEMAP¶

%preview ~/GIT/github/mvarbvs/dsc/susie_comparison/PIP_comparison_0622_estvar_true_filter_true_0p5.caviar_vs_finemap.png

cd ~/GIT/github/mvarbvs/dsc/susie_comparison
convert -append PIP_comparison_0622_estvar_true_filter_true_0p5.susie_vs_dap.png PIP_comparison_0622_estvar_true_filter_true_0p5.susie_vs_caviar.png PIP_comparison_0622_estvar_true_filter_true_0p5.susie_vs_finemap.png PIP_comparison_0622_susie_vs_others.png
convert -append PIP_comparison_0622_estvar_true_filter_true_0p5.dap_vs_caviar.png PIP_comparison_0622_estvar_true_filter_true_0p5.dap_vs_finemap.png PIP_comparison_0622_estvar_true_filter_true_0p5.caviar_vs_finemap.png PIP_comparison_0622_others.png

Revision	Author	Date	Message
934029f	Gao Wang	2018-06-22	Update documentation
bd74ef0	Gao Wang	2018-06-22	New simulations and add FINEMAP
1085909	Gao Wang	2018-06-06	Update power comparisons to use both replicates per gene
2ad462b	Gao Wang	2018-06-06	Add power comparisons
95d3072	Gao Wang	2018-06-05	Update documentation
00da2df	Gao Wang	2018-06-05	Add more PIP comparisons
d1835be	Gao Wang	2018-05-31	Fix CAVIAR vs susie outlier comparison
7b277dc	Gao Wang	2018-05-31	Explore CAVIAR vs susie PIP for n = 1
4324e0d	Gao Wang	2018-05-31	Add data-set names that created DAP issues
a0a3c0b	Gao Wang	2018-05-31	Fix up PIP plot