I have a highly-related set of samples that I am attempting to run GENESIS on for association analysis of a phenotype. When I initially ran pcair on the genodata with my kinship matrix my initial pcair result had only 3 PC's. This appears to be due to the related set containing nearly all individuals and only 3 individuals assorting into the unrelated set. After relaxing my kinship threshold to kin.thresh = 0.077 I was able to generate a pcair structure with 20 PC's, our general goal for analysis. Using pcair in ScanAnnotationDataFrame and attempting to run fitNullMM with resulting ScanAnnot data structure, I receive an error. Here is the full output of running fitNullMM:
nullmod <- fitNullMM(scanData = scanAnnot, outcome = p, covars = c("pc1"), covMatList = covMatList, family=varType) Reading in Phenotype and Covariate Data... Fitting Model with 410 Samples Computing Variance Component Estimates using AIREML Procedure... Sigma^2_Kin Sigma^2_E logLik RSS Error in chol.default(Sigma) : the leading minor of order 402 is not positive definite
I have several questions. 1) Is reducing relaxing the kin.thresh during pcair an appropriate way to deal with the relatedness in these data? 2) Is there a more appropriate way to analyze highly-related individuals for GWAS using GENESIS? -- I would prefer to stick with GENESIS as it is our labs current standard for analysis. 3) What does this error message mean and how do I go about correcting the problem? -- I would appreciate any insight into what is going on here. I'm not even certain what data-structure is causing the problem or where in the process.
Thank you for any help you can provide.
Note: I am using version GENESIS_2.4.0 on R 3.3.1 and am currently unable to update to the current version. (I'm in process of upgrading but I need to run with the current tool-chain right now).