Panel Discussion
summary
Panelists argue that the missingness‑indicator approach fails under observational unconfoundedness and under conditioning on missingness patterns, with epidemiology simulations and theory showing bias unless extra modeling assumptions hold. Genomic analyses use hidden‑Markov models for genotype knockoffs so that knockoffs control FDR but often cut discoveries when requiring invariance across populations. Protein design pipelines combine variational autoencoders or simple marginal “control‑knob” distributions, blockwise mutation heuristics informed by structure, and emerging tools like conformal prediction and deep knockoffs for uncertainty and sampling.
AI-written summary · about