MMLS-C By : Laurence Bisht References : The Power to Detect Linkage in Complex Diseases Means of Simple LOD-score Analyses. By David A.,Paula Abreu and Susan E. hodge
Overview Introducing the problem. Goals.. Intuition. What is MMLS? What is MMLS-C? Generating Models. Results. Discussion
Introducing The Problem … What is it? Analyzing Complex diseases, i.e. analyzing human linkage data.
Our Goal Is … Finding the disease gene’s locations. Limitations : Complex Disease. MOI (Mode Of Inheritance) is unknown. Using all data available, somehow… Get a Powerful Method, stable and reliable one.
What was in Lecture 8… Affected sib pair (ASP) Affected Pedigree Member (APM) Nonparametric linkage (NPL) Fact 1: We need to Exploit all data we have. But… These method’s use ONLY affected family members. Intuition To MMLS-C
Intuition Cont. Fact 2: Maximum Likelihood analysis via LOD- score, assuming we have the inheritance model is most powerful method for finding linkage.
Solution 1. Use Maximum likelihood analyzes trying all modes of inheritance.. Why not? Is it logical? Suppose given a super machine that can do it… how will this work? problem : 1. How will we compare?
Solution 2 – MMLS Choose several models. Run the Maximum likelihood analyzes for every chosen model- (LOD-score). Take max(Z) as the test statistic for linkage. This is MMLS – Maximizing the Maximum LOD-Score
MMLS – analyzes Negative sides. 1. Using Several parameters (models), Multiple tests … Result: Increase of type I error. 2. Unknown effect on the statistical power. 3. Most important: Is there a reason to believe that the models we used can lead us somewhere close to the true model?
Solving 1 Using Several parameters (models), Multiple tests … Result: Increase of type I error. We will show that: If we perform linkage analyzes twice, once assuming recessive and once assuming dominant, with an arbitrary penetrance of 50% Then : The Z threshold must be increased by at most ~0.3 for Z max <3.
Solution facts Too stringent.. in most cases examined.. Suggestion: Perform the test twice with the two models proposed. take arbitrary penetrance (0.5 is good) take the larger between the two resultant Z max subtract 0.3 to “correct” the result It has been shown that: when there is linkage, Z max relatively modest as the penetrance is varied. (relatively little information is lost assuming a single penetrance).
Points 2 and 3. Simulation study will answer them … Simulation will : Quantify the effect of correction for multiple testing. Examine the power to detect linkage in two cases discussed later…
Generating Models D20,D80- Dominant with 20% and 80% penetrance. R20,R80- Recessive with 20% and 80% penetrance. Int10,Int30,Int50,Int80 – Intermediate (i.e. heterozygote penetrance is 10%, 30%, 50%, 80% while the homozygote will always be 90% and 0%). note that when f2=0 (homozygote penetrance) its simple recessive
Generating Models Cont. The MMLS power is expressed when f2=5-15%.. A hard case… Additive-3, additive-2: Two loci models. when it is required at least 3, 2 (accordingly) disease alleles at the two loci.
Generating Models Cont. Always one disease locus linked to the marker with recombination fraction (theta = 0.01). for the additive model the other one is linked and for the other we will examine 3 recombination fractions: 0.1, 0.05, 0.01
Simulation Parameters They examined 14 Generated Models one of each. On modified version of the Two-locus simulation program for Greenberg (their program) 1000 datasets of 20 families.
Simple MMLS-C Running MMLS for R50 and D50, as previously described. Correction factor was varied* ~0.24 when Z max < 0.59 ~0.3 when Z max < 0.59 *(according to hodge (1997))
Results Table Notice that: Max[D50,R50] < E[raw MMLS] =~ E[MMLS-C]+0.30 < E[True]
Power Vs. LOD-Score
Power Table
Discussion … Our main goal was : Examine the power to detect linkage using MMLS-C After we passed over the results we can see the following: MMLS-C doesn’t substantially decrease the power compared with the True MOI. The range of the MMLS-C – TRUE was [0.3,0.7] (except for three case )
ASP Vs. MMLS-C
Conclusions Pro : MMLS-C is a simple method for analyzing complex diseases. Exploits all data available. Reliable. The assumption that the linkage at the locus being tested is critical. Against : Was tested on small data set. not always the best method.
The End!