Rare variant tests

Summary of discussion from ESP rare variant working group

The rare variant working group within ESP has discussed the issue of rare variant tests on several conference calls. The end result is that we recommend selecting one test from each of these three categories;

1. Aggregate tests (typically with 1% threshold, nonsynonymous SNPs only, with meta-analysis across different ethnic groups)

2. Tests that allow for risk and/or protective variants (again, probably 1% threshold, nonsynonymous SNPs only, with meta-analysis across different ethnic groups)

3. Weighted tests that allow incorporation of more common variants (possibly apply 5% threshold?, nonsynonymous only, etc.)

A brief summary of the RV discussion;

- Permutations (where we permute phenotype while maintaining ethnic group) will likely be required to get empirical p-values. These RV tests typically provide conservative p-values (deflated QQ plot), but not always. Thus, a computationally intensive test will not be practical for performing large numbers of permutations (at least 1000).

- Using too many tests will decrease the power overall because of correction for family-wise error.

- Although we'd like to evaluate power and type I error rates of these tests under a variety of genetic models, the reality is that we have so few known positive examples it would be difficult to assess them all in a fair way at this time. Instead, we expect to re-convene this discussion group at a later date once some true positive associations are identified.

- Shamil Sunyaev is performing a bake-off with some of these tests, and we look forward to seeing his results in the future.

- PLINKSeq is on its way, but is likely a month away from release (end Feb 2011)

Summary of rare variant tests for sequence data

Compiled by Cristen Willer and Suzanne Leal for the ESP Feb 1, 2011

* indicates applicability to quantitative data

1) Aggregate tests using a cut off e.g. 1 % analyzing nonsynonymous variants to detect detrimental variants

Test Name	Reference	Software	Notes
CMC/T1 test*	Li & Leal, 2008
KBAC	Liu & Leal, 2010
VT*	Price et al., 2010	http://genetics.bwh.harvard.edu/rare_variants/
WSS	Madsen & Browning, 2009
CMAT	Zawistowski et al. 2010
ANRV/GRANVIL*	Morris & Zeggini
RARECOVER	Bhati et al. 2010
CCRaVAT and QuTie*	Lawrence et al. 2010	http://www.sanger.ac.uk/resources/software/rarevariant/
RVE (rare variant exclusive)	Cohen & Hobb

2) Aggregate tests for protective and detrimental variants (recommend 1% cutoff)

Test Name	Reference	Software	Notes
C-alpha	[Neale et al., submitted]
Ionita-Laza & Lange	Ionita-Laza & Lange, 2011
DASH*	Han & Pan
SKAT*	Wu et al., 2010	http://www.hsph.harvard.edu/~xlin/software.html
WHaIT	Li et al. 2010	http://www.sph.umich.edu/csg/yli/whait/
EMMPAT*	King et al. 2010	http://home.uchicago.edu/~crk8e/papersup.html

3) Analyzing common and rare variants together (could down-weight or threshold common variants)

Test Name	Reference	Software	Notes
WSS	Madsen & Browning, 2009
RARECOVER	Bhati et al. 2010
Step-Up Collapsing*	Hoffman et al. 2010
CMC/T5 test*	Li & Leal, 2008
MENDEL*	Zhou et al. 2011	http://www.genetics.ucla.edu/software/download?package=1

4.) Analyze higher frequency rare variants >1% individually

                  Use same regression frame work which has been used for common variants*
                  Use meta analysis to combine results from sequence data and imputed genotypes to increase power*

Additional tests

Test Name	Reference	Software	Notes
Logic regression*	Kooperberg et al. 2001
Sequence diversity	Anderson et al. 2006
Sequence dissimilarity*	Schork et al. 2008, Wessel et al. 2006
Ridge regression *	Malo et al. 2008