0

I have some SNP data, and wanted to investigate the association between mother's SNPs and a phenotype in their children. In my model, the child's phenotype is the response variable; mother's SNP is the predictor variable. I also adjust for multiple confounders, such as mother's nutrition status during pregnancy etc. Most of the children in my study came from different families, but some are siblings. How do I take into sibling effect into this analysis?

kjetil b halvorsen
  • 77,844
user106506
  • 1

1 Answers1

1

A mixed model with a random intercept for family clusters ought to suffice to handle a constant correlation of the residuals between siblings. This would be an exchangeable correlation structure. It can be summarized using the ICC which is a post-estimate from such models using the random intercept variance over the between subject variance. Large ICCs indicate greater heritability, but cautioning of course that if some of the "adjustment variables" are phenotypic of the maternal trait, small ICCs can still arise in highly heritable traits. This also doesn't account for dominant/recessive patterns of inheritance.

AdamO
  • 62,637
  • Thanks for the answer! I'm not sure if I understood it correctly. I have subject ID for the children. For example, they are 1, 2, 3, 4... Should I create a new variable called family? If 3 and 4 are siblings, I make the value of family 1, 2, 3, 3. And then treat family as a random effect in the mixed model? Thanks. – user106506 Jun 06 '17 at 17:16
  • @user106506 it depends on the software... but generally, the family indicator will assign the same value to two children who are related and different values to two children who are not related. So if subject IDs 1 and 2 are siblings, but 3 is not related to 1 or 2, then family ID would be 1, 1, 2. for those three participants. – AdamO Jun 06 '17 at 17:18