Feature reuse and scaling: Understanding transfer learning with protein language models | Alex Lu

Large pretrained protein language models (PLMs) have improved protein property and structure prediction from sequences via transfer learning, in which weights and representations from PLMs are repurposed for downstream tasks.