Lasse Middendorf

The paper identifies a subset of proteins that causes discordance between structure and disorder predictors, showing that this subset is compositionally distinct and not a general sequence-class artefact.

ProtGPT3 is an open-source family of protein language models that enables controlled protein generation through alignment and prompting, showing effectiveness in generating functional proteins.

The paper identifies and characterizes functional de novo proteins in Drosophila using machine learning tools, finding some adopt known folds and participate in cellular processes.

This study evaluates how structure prediction tools perform on de novo, random, and conserved proteins, finding differences in prediction accuracy and disorder correlations.