SuSiE paper results based on estimated prior¶

This notebook displays another version of this page -- in the SuSiE paper analysis we focused on fixed scaled prior variance (PVE = 0.1). Here we show the results of using estimated prior variance.

Conclusion¶

Results are largely robust to setting prior to 0.1 vs estimating it.

Results on the ROC and PIP direct comparisons now appear closer to that of DAP-G.
CS power results remained mostly the same. Estimated prior did not help at improving coverage (reducing false positives).

%cd ../dsc

/project/mstephens/SuSiE/mvarbvs/dsc

PIP calibration¶

%preview susie_comparison/PIP_comparison_1008_calibrated_prior_0p0_null_0p0.png

Variable-level power vs false discovery¶

%preview susie_comparison/ROC_1008_prior_0p0.png

CS power comparison¶

%preview susie_comparison/cs_eval/cs_comparison_prior_0p0_null_0p0.png

PIP direct comparison¶

Agreement with DAP is now tighter for the one-causal case.

%preview susie_comparison/PIP_comparison_1008_prior_0p0_null_0p0.susie_vs_dap.png

Reliability of reported CS coverage¶

%preview susie_comparison/Coverage_1008_prior_0p0.png