1	A New Approach Toward Complex Traits Shaw-Hwa Lo, Tian Zheng Columbia University Thursday, August 05, 2004
2	Talk outline The dilemma and the solution The information-driven screening Illustration using a genome scan on IBD Conclusion Reference and current efforts
3	The dilemma Complex traits —“… are caused by multiple genes interacting with each other and with environmental factors to create a gradient of genetic susceptibility to disease.” (Weeks and Lathrop 1995)
4	Screening problem Difficulties with such a screening Large number of markers Mild individual effects of markers Possible gene-gene interactions Moderate size of patients
5	An information-driven solution A backward selection algorithm that applies an association information measure with respect to the disease on a set of markers under study; reduces the set by deleting an unimportant marker that contributes the least to the association information contained in the set; stops when deleting any of the remaining markers will result in a substantial loss of information. Algorithm is ran on random marker subsets of size that can be handled by the size of patients. Final selection is based on aggregated screening results on a large number of random subsets.
6	Notation for Case-Parent Trio Haplotype Data Data: a random sample of n patients and their parents. 2n parent-patient transmission pairs; Each pair consists of two haplotypes; one transmitted and the other untransmitted. For pair, let be the haplotype transmitted to the diseased child, and be the untransmitted
7	Information Measure: Haplotype transmission Disequilibrium (HTD) Define counts: HTD is defined to measure the amount of association information contained in the set of markers being tested
8	Idea of screening Assume m markers are being screened; The marker is to be evaluated; Consider (the -deleted set) The amount of information contributed by can be evaluated using the HTD difference—the information drop, which can used to measure the importance of to current marker set.
9	Marker Importance Measure Haplotype Transmission Association (HTA) Expectation of HTA reflects the importance of the marker under evaluation: where is the haplotype risk ratio.
10	Properties of HTA Statistic
11	BHTA Screening algorithm Backward Haplotype Transmission Association algorithm: Step0: Start with Step1: Calculate HTA for each retained in Step2: If all HTA are negative, stop; otherwise delete the marker with highest HTA; Step3: If , stop; otherwise continue to step1; Stop: return all markers retained in as selected markers
12	Example 1: a model with no marginal effects (Simulated)
13	Information Flow during BHTA
14	Issue of sparseness If the number of possible haplotypes is significantly larger than the number of observations, markers are deleted randomly. This is because the counts of haplotype transmissions are mostly 0's or 1's, hence the HTA scores based on these counts all cluster around zero, which makes the selection non-informative.
15	Two-step BHTA-based marker selection procedure Step 1: Randomly select k (15, for instance) markers out of the original set of K markers. Run BHTA on the selected markers and record the markers returned. Step 2: Repeat step 1 B times (e.g. B=5000). Selection: important markers are selected based on the distribution of the return frequencies.
16	Relative Chance of Deletion During BHTA
17	Two-Step Selection Procedure Results on Example 1
18	Example using Inflammatory Bowel Disease IBD consists principally two chronic idiopathic inflammatory diseases of the gastrointestinal tract: ulcerative colitis (UC) and Crohn’s disease (CD); Genetic basis: relatives of individual with either CD or UC are at increased risk for developing either form of IBD; There have been several loci with relevance to IBD etiology identified since 1996, through different studies, IBD1-IBD7.
19	Data 235 case-parent trios 402 microsatellite markers on all 23 chromosomes with an average of 12 CM inter-marker distance.
20	Implementation of BHTA Step I: Imputation, haplotype inference and marker dichotomization Step II: BHTA screening on 10 independent imputations (each with 10,000 screenings) Step III: Aggregation of return results and selection of important markers
21	Selecting the important markers
22	Another method to identify the important group (interesting group, Efron (JASA, 04) f (z) = p g (z) + (1- p) h (z), false discovery rate fdr (z)= g(z)/ f(z) < .1 g(z) is a normal density with mean= 1090 and SD= 82.
23	Selecting the important markers
24	Results 48 markers were identified as important markers, spread across many of the 23 chromosomes; Our selected markers overlap with all previously reported IBD loci, except IBD6; The importance of each marker is further evaluated by its return frequencies.
25
26
27	Selected markers at IBD loci
28
29	Conclusion The major weakness of conventional approaches are due in part to the fact that only fractional information from the data is used; Our new approach intends to draw substantially more information from data; More understanding of complex traits can be derived if data already collected could be suitably reanalyzed by this approach; This new approach will be useful in the future when the information of a large number of dense markers becomes available; Information about gene-gene interactions can be derived from joint return patterns in BHTA results.
30	Reference and current efforts Backward haplotype transmission association (BHTA) algorithm—a fast multiple-marker screening method. Hum Hered 53:197-215 Extensions to case-control studies and quantitative traits are currently under study.