George I. Austin

The paper shows that sparse microbiome features can become phenotype-associated after compositional transformations like CLR, challenging claims that this indicates information leakage.

DEBIAS-M corrects processing bias in microbiome data, improving cross-study generalization of prediction models by learning interpretable bias-correction factors.

This study predicts preeclampsia using first trimester vaginal microbiome data with machine learning, achieving an auROC of 0.78 in the training set and 0.80 in an independent cohort.