Seeking Guidance on Standardization Methods for eQTL Analyses with Diverse Preprocessing Techniques

Question

[DS]

Hi all,

I am currently conducting eQTL analyses on different datasets and have observed variations in the gene expression preprocessing methods used for each dataset. Here's a summary:

• Dataset A: TPM -> Trimmed Means -> Inverse Variance Normalization
• Dataset B: TPM -> log10(TPM+1)
• Dataset C: TPM -> log10(TPM+1) -> Quantile Normalize -> Scaling
• Dataset D: TPM -> log10(TPM) -> Median Normalize -> Remove Outliers (three interquartile ranges above or below the upper and lower quartile)

I believe the ideal scenario would involve obtaining TPMs from each dataset and applying the same metrics consistently. However, considering the current transformations, I'm exploring alternatives for standardizing beta estimates, obtained from eQTL analyses with a linear mixed model.

Apart from employing z-scores, are there any other effective approaches for standardizing beta estimates that you would suggest?

Additionally, I'm considering the idea of ensuring a standard normal distribution for each dataset before running eQTL, irrespective of the preprocessing method used. Any insights or recommendations on this approach?

Thank you!