PepFoundry: A Pipeline for Building Machine-Learning Ready Representations of Non-Standard Peptides Containing Cycles, Non-Natural Residues, Polymer Units, and More 1. PepFoundry introduces a novel approach to convert peptide sequences, including those with non-canonical amino acids and complex structures like cycles and polymer units, into machine-learning ready representations. This innovation significantly expands the scope of peptides that can be analyzed using ML techniques. 2. The pipeline leverages SMILES strings in the CHUCKLES format to generate atom-mapped RDKit molecule objects. This method allows for the extraction of detailed atom-level features such as Morgan fingerprints and graph representations, which are crucial for accurate ML modeling. 3. A key finding is that atomic-level representations consistently outperform traditional sequence-level representations in predicting peptide properties, regardless of the ML model used. This highlights the importance of incorporating detailed structural information in peptide analysis. 4. PepFoundry also enables the visualization of peptide embeddings in latent space, demonstrating its ability to capture relationships between different peptide classes, including L-peptides, D-peptides, and peptoids. This feature enhances our understanding of peptide chemical space. 5. The package includes a customizable database of amino acids and modifications, making it easy to incorporate new chemistries. This flexibility is essential for researchers working with emerging peptide designs and modifications. 6. PepFoundry provides a scalable platform for peptide discovery and optimization, facilitating the rapid construction of customized peptide libraries. This accelerates the application of ML in peptide-based drug design and other related fields. 📜Paper: doi.org/10.26434/chemrxiv-20… #MachineLearning #PeptideScience #DrugDiscovery #ComputationalBiology #Cheminformatics