Leave-one-batch-out cross-validation reveals strong batch effects in Raman spectroscopy of yeast cultures

MSA-based pLMs encode evolutionary distance but don’t reliably exploit it

From black box to glass box\: Making UMAP interpretable with exact feature contributions

Do protein language models understand evolution? Mixed evidence from ancestral sequences and ESM2

Cross-trait learning with a canonical transformer tops custom attention in genotype–phenotype mapping

Paired residue prediction dependencies in ESM2

Closing the divide between analysis and publication\: The notebook pub