Summary
Enzymes can be miniaturized with Monte Carlo sampling and protein language model per-residue embedding similarity at key catalytic residues (1). This required a loss function for sequence length and for embedding similarity (cosine). No wet lab validation was carried out to see if the designs actually worked.
1.
Lála J, Agrawal H, Dong F, Wells J, Angioletti-Uberti S. An Energy Landscape Approach to Miniaturizing Enzymes using Protein Language Model Embeddings. openRxiv; 2026. Available from: https://doi.org/10.64898/2026.03.04.709378