Can AI Spot a Peptide Hormone From Sequence Alone?

Summary. A research team has described pLM-HP, a deep-learning framework that predicts whether a given peptide sequence functions as a hormone. The model combines representations from the pre-trained protein language model ESM2 with a bidirectional long short-term memory (BiLSTM) network. On an independent test set, the authors report a balanced accuracy of 95.64%, sensitivity of 94.48%, specificity of 96.79% and an area under the ROC curve of 0.991. They say this outperforms existing tools. Identifying which genome-encoded sequences act as hormones has traditionally required slow, costly experimental work.

Research context (RUO). Computational screening is becoming a routine first filter in peptide discovery. It helps narrow very large sequence spaces before synthesis and wet-lab validation. Prediction is not confirmation, though. Candidate sequences still need synthesis, analytical characterisation (identity, purity, counter-ion and water content) and controlled in-vitro or animal-model work before any functional conclusion can be drawn. Tools like this make well-characterised reference material more valuable, not less.