The Error Rate Control In The Multi-locus Association Study Of Complex Diseases

Though thousands of research institutes are searching for the causative genes ofcomplex diseases with expensive biomedical experiments, only 16%~30% of thepositive reports have been successfully replicated. This replication rate is far belowthe theoretical expectation of 80%. Association studies can detect the causativegenes more powerfully than linkage analyses, but they also bear more dangers offalse positive results. Controlling the error rate accurately is important to improvethe reliability of association studies.The possible pitfalls in the research process are reviewed and their solutions aresummarized. In experiment designs, population heterogeneity and stratification arethe two major sources of errors, but their impact can be weakened by matching thecontrols with the cases. In data analysis, both multiple testing and search strategiesconcern error rate control. The two issues are discussed in details. In results'publication, authors and editors usually prefer positive results. This can misleadreviews and meta-analyses to positive results. We believe that authors should beencouraged to publish negative results.The correlation among hypothesis tests is the bottle neck in controlling theeffect of multiple testing. Cheverud (2001) proposed the idea of adjusting them as ifthey were some independent tests, but his estimation of the number of independencyis overly large. We propose a more accurate estimation and applied it to control errorrates. Tested on both real and simulated data, our new method approximated thepermutation test closely, but is 1000 times faster. In multi-locus association study,the computation burden is usually heavy. Our new method can be an alternative ofthe permutation test.Single locus search, simultaneous search and conditional search are threepopular search strategies, but they either miss the interactions among genes or catchspurious causative gene combinations. Considering the priority and hierarchy oflocus-combinations, we propose a hierarchically iterative search strategy whichoutperforms the three strategies on real data. Applying this new search strategy toreal paranoid schizophrenia data, we discover that COMT-158-NlaIII is regulated byCOMT-136-BclI. The same genotype of COMT-158 may function completelydifferently. This discovery explains the conflicting reports on COMT-158-NlaIII.Finally, to evaluate the errors of various analyses algorithms, we design specialsoftware which can simulate recombination, linkage disequilibrium and epistasis.