A Graphormer approach for predicting peptide half-life, secondary structure, solvent accessibility, and post-translational modifications

Peptides represent a promising class of therapeutic agents due to their high target specificity, favorable safety profiles, and efficient tissue penetration. However, their clinical development remains limited by poor in vivo stability, rapid enzymatic degradation, and short plasma half-lives. The structural flexibility of peptides and their high solvent accessibility further increase their susceptibility to proteolysis, resulting in rapid clearance and reduced therapeutic efficacy. Although strategies such as cyclization and incorporation of non-natural amino acids can improve stability, it remains challenging at early stages to identify which peptides would benefit from such modifications based on their pharmacokinetic and physicochemical properties. Here, we present a Graphormer-based machine learning framework for predicting four key peptide properties: half-life, relative solvent accessibility, secondary structure, and post-translational modifications. The framework supports both cyclic peptides and non-natural amino acids, extending beyond the scope of many currently available tools. To evaluate performance, we benchmarked our predictive models against existing methods for each property. Rather than focusing on surpassing current approaches, this study emphasizes the practical applicability of the proposed framework for peptide property prediction. Additionally, the half-life prediction model was validated using an experimental dataset generated by a pharmaceutical company, demonstrating the flexibility of the approach for training on user-generated experimental data.