Hello everyone, I'm writing to you because I have some questions about some association results I got.
You might skip the background and jump directly to the Query. I tried to be brief with the background, but it took some paragraphs.
I'm trying to detect some potential markers for highland adaptation of maize, based on the pressence of an introgression of an invertion at chromosome 4 from teosinte.
I'm currently working with 10 SNPs affter filtering, 30 CMLs available from highland adaptation , 2 from Highland East Africa, 148 CMLs from Lowlands and 4 teosintes.
The phenotypes being used are Adaptation per se, Days to silking, Plant height, plant color and Sheath pubescence, being Plant height the only phenotype not previously associated with highland adaptation.
When analysing the distribution of Days to silking I noticed that It was a mixture of two normals. CMLs that had >74DtS were mostly Highlands, below that were mostly lowlands. So I converted this scalar phenotype in a control/disease phenotype, plant color and Sheath pubescence are ordinal phenotypes, Never the less i converted them in control/disease phenotypes too.
After PCoA, it was clear that there were two genotypic groups, although 70% of highlands had the favourable genotypes, and practically all lowlands had the unfavourable genotypes. 30% of highlands were closer to the subgroup of Lowlands from Latin american tropics (I'm mentioning this because the group of lowlands splitted in two subgroups).
After performing robust response screening to all phenotypes, the best hits were for Days to silking and Adaptation, Sheath pubescence was ambiguous.
Association tests were performed on PLINK for this phenotypes and got some significative hits.
The issue (I believe) is that the Genomic inflation estimate (lambda) based on median chisq for these individual associations was much higher than 1 (possibly indicating stratification).
In order to smooth lambda, Cochran-Mantel-Haenszel tests for 2x2xK and IxJxK, and Linear and Logistic tests taking the first 2 components of the PCoA as covariates were perfomed. Lambda was succesfully smoothed to 1, but no significant hits were found.
After using STRUCTURE I found that Fst was also high.
Am I overdoing this analysis? What are your suggestions?
I appreciate your time helping me.