Multivariate Bayesian variable selection regression

• Overview
• Analysis
• Prototype
• Writeup

Power comparison susie vs DAP

Here we compare power of susie and DAP under different number of simulation signals for fixed PVE.

%revisions -s -n 10

Revision	Author	Date	Message
4bbe27e	Gao Wang	2018-06-11	Add median size to power comparison
24e1fbc	Gao Wang	2018-06-07	Update revision table
85bdece	Gao Wang	2018-06-06	Add average size and LD for susie vs dap
1085909	Gao Wang	2018-06-06	Update power comparisons to use both replicates per gene

We set n_causal ranging from 1~5, each simulating 100 data-sets from 50 genes. For each simulated data, we compute both susie 95% CS and DAP 95% cluster, and we evaluate for the CS or clusters reported how many of them capture at least a signal.

We evaluate the following quantities:

• Power: the proportion of signals captured by susie CS or DAP clusters
• False discovery proportion (fdp): the proportion of susie CS or DAP clusters that do not contain any signal
• Average size: average number of variables in susice CS or DAP clusters that contains a signal
• Average LD: average LD between variables in susice CS or DAP clusters that contains a signal

The power workflow in this notebook does the computation.

%cd ~/GIT/github/mvarbvs/dsc

/home/gaow/GIT/github/mvarbvs/dsc

Conclusions

• susie has slightly more power over DAP
• FDP of susie is lower
• Average size of susie CS is greater than DAP cluster, but the average LD within susie CS is smaller than DAP cluster
  • in other words susie CS is more pure

susie var(Y) vs DAP

readRDS('susie_comparison/Power_comparison_0622_cluster_prob_estvar_false.rds')

n_signal	expected_discoveries	susie_discoveries	dap_discoveries	susie_power	dap_power	susie_fdp	dap_fdp	susie_avg_size	dap_avg_size	susie_median_size	dap_median_size	susie_avg_ld	dap_avg_ld
1	200	200	177	0.99000	0.8850000	0.01000000	0.0000000	18.03535	7.570621	4.5	4.0	0.9783407	0.9749153
2	400	275	265	0.66750	0.6350000	0.03272727	0.0490566	19.50376	10.623016	6.5	6.0	0.9487225	0.9363929
3	600	326	325	0.52500	0.4883333	0.05214724	0.1107692	19.65372	11.000000	6.0	7.0	0.9431493	0.9325156
4	800	358	342	0.44875	0.3912500	0.03631285	0.1052632	21.14493	11.568627	9.0	7.5	0.9238565	0.9118464
5	1000	375	383	0.36500	0.3360000	0.09066667	0.1488251	20.53372	12.303681	9.0	10.0	0.9232786	0.8895613

susie est_var vs DAP

readRDS('susie_comparison/Power_comparison_0622_cluster_prob_estvar_true.rds')

n_signal	expected_discoveries	susie_discoveries	dap_discoveries	susie_power	dap_power	susie_fdp	dap_fdp	susie_avg_size	dap_avg_size	susie_median_size	dap_median_size	susie_avg_ld	dap_avg_ld
1	200	200	177	0.9700000	0.8850000	0.03000000	0.0000000	16.25773	7.570621	3	4.0	0.9825322	0.9749153
2	400	289	265	0.6875000	0.6350000	0.05190311	0.0490566	16.42701	10.623016	5	6.0	0.9487389	0.9363929
3	600	345	325	0.5366667	0.4883333	0.07536232	0.1107692	18.25078	11.000000	5	7.0	0.9471269	0.9325156
4	800	389	342	0.4662500	0.3912500	0.06940874	0.1052632	18.20718	11.568627	8	7.5	0.9295411	0.9118464
5	1000	413	383	0.3860000	0.3360000	0.11138015	0.1488251	18.07357	12.303681	8	10.0	0.9268335	0.8895613

---

Copyright © 2016-2020 Gao Wang et al at Stephens Lab, University of Chicago