You are absolutely on the spot, always with minimum thickness of the fcap. As per the Oxford study. All the plaques we used have histologies available. Unfortunately we didn't have the time resources to profiles all the plaque tissues at MUC biobank. I'll email you tonight :)

Thank you Luke! @erwinschoof.bsky.social talked highly about you. If possible, may we have a zoom call? It would be good to build some synergies/exhabges between our projects, and get an independent, unbiased perspective on it! :)

I appreciate the detailed reply. I agree & experimentally observed that convergence to the 1st moment (i.e. mean) leads to underestimating 2nd+ moments. Also, MICE was a little better at preserving the higher moments. I used the matrix norm of the covariance matrix to measure the imputation artefact

This is a very informative article. Reducing 37 imputation methods into 5 major groups helps grasp their diversity. May I ask how Neighbor/Regression and NN-based approaches deal with heteroskedasticity in quantitative data (ex. MS-proteomics)? How would it compare in MICE approaches? Thank you!

My general approach to biostats is that the stats problems we just discovered were solved by WH 10 years ago :)

www.nature.com/articles/nme...

Yes, and entirely common. Wolfgang Huber's Independent hypothesis weighting approach is great for adjusting p-vals, and accommodates heteroscedasticity present in proteomics data. Briefly, you adjust p-vals based on bins defined by some covariates (ex. protein median).

The fox is in the henhouse.

Mike's Sage is outstanding. When I first used it, I thought it had errored out. Turns out it was done.

I would frame it another way. Science is a method to answer questions. We have drifted away from asking questions, or at least from asking good, focused questions, framing testable hypotheses and then testing with good experiments. Big datasets can be very useful, but we need good questions.