GPepT: A Foundation Language Model for Peptidomimetics Incorporating Noncanonical Amino Acids

Language models have been increasingly popular in therapeutic peptide generation, but molecular diversity remains limited due to reliance on the 20 canonical amino acids. We propose a language model that generates peptidomimetics incorporating noncanonical elements like noncanonical amino acids and terminal modifications. To accomplish this, we created a vocabulary of over 17,000 noncanonical elements by extracting them from chemical formulas stored in the ChEMBL database. Our pretrained language model, GPepT, showed improved diversity in molecular structures and chemical properties. To demonstrate its real-world application, we fine-tuned the model for antimicrobial peptides. Experimental validation revealed that one of the generated peptidomimetics exhibited effective antimicrobial activity, marking a successful case of AI-driven peptide development. GPepT is fully accessible on HuggingFace: https://huggingface.co/Playingyoyo/GPepT.