Prot2Prop
Structure-aware fine-tuning of protein language models for joint prediction of multiple developability properties from protein structure inputs.
Prot2Prop trains a shared adapter with lightweight task-specific residual adapters and heads on top of a structure-aware protein language model, jointly predicting developability properties such as thermal stability, aggregation propensity, solubility, and oligomerization state. It targets a gap in the current ecosystem, where most tools handle these properties with separate, older, task-specific models rather than a unified, modern, structure-aware one.