On 80 WGS samples, I'm dissecting SNP signatures linked to milk production in a scarcely studied animal. Post-variant calling and QC association analysis have been tricky. I'm here to tap into our collective wisdom and refine my approach. Study Snapshot:

Objective: Identify SNPs that differentiate between high (>5 L/day) and low (<5 L/day) milk producers. Data: 80 WGS samples. Methodology:

Phenotypes were normalized (high = 1, low = 0). PCs from PCA are used to correct for population stratification, with phenotype information as a covariate. Initial approach: linear models, then FDR and Bonferroni for multiple comparison correction. A significant SNP threshold was set at <1e-5, yielding four significant SNPs. Expanded to Linear Mixed Models (LMM), we found six significant SNPs at 1e-5 and 99 at 1e-4, focusing on positive associations with high milk production.

Seeking Guidance:

Feasibility: Can 80 WGS samples provide a reliable genetic signature for milk production? Methodology Recommendations: Given my data and initial results, what methodologies or modifications could improve the analysis?


1 Answer 1


Visualise your problem and solution.

6 SNPs is not a particularly large search space (64 possible combinations assuming a dominant or recessive model with dimorphisms, 729 if heterozygous variants are taken into account). I would graph all those haplotypes out in a heatmap (but only the ones present in your samples, so there'd be a maximum of 80 different ones) and make sure that there is a clear pattern.

If population stratification is a concern, then you should also show the PCA, with separate colouring for the different phenotypes. If the two groups have similar spread in the PCA (which would be expected for a trait that is only affected by a small number of genes), then PCA correction doesn't make sense (and may lead to false associations).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.