Songling Li

A germline shortcut in protein language model retrieval of adaptive immune receptors

It is shown that conventional random train-test splits inflate retrieval accuracy by 15–28 percentage points owing to clonal leakage, and clone-aware benchmarking as a practical standard of comparison is established, defining the strengths and limitations of frozen zero-shot PLM embeddings for immune receptor retrieval and establishing clone-aware benchmarking as a practical standard of comparison.