Integrated <i>In Silico</i>, <i>Ex Vivo</i>, and <i>In Vitro</i> Framework for Early Derisking of Mast Cell Degranulation in Peptide Drug Candidates

Recent studies have shown that certain peptides and small molecules can induce pseudoanaphylaxis reactions by triggering mast cell degranulation (MCD), resulting in the release of vasoactive and proinflammatory mediators. This mechanism can result in severe adverse drug reactions with potentially life-threatening consequences in humans or loss of tolerability in animal studies, representing a considerable challenge in the development of peptide and small-molecule therapeutics. Therefore, early identification of drug candidates with MCD potential is crucial for an efficient Design-Make-Test-Analyze (DMTA) cycle while promoting the 3Rs principle (replacement, reduction, refinement) in animal research. In the present work, we introduce a proactive risk mitigation strategy aimed at evaluating and minimizing the MCD activity of peptide drug candidates. We developed an <i>ex vivo</i> rat peritoneal mast cell degranulation (rMCD) assay to screen and prioritize candidates that do not exhibit rMCD activity during the lead optimization phase. Importantly, structure-activity relationships (SAR) were established by leveraging rMCD data sets which included ∼3000 diverse peptides across 28 internal programs targeting multiple therapeutic areas. Critical physicochemical properties were identified as predictive calculated parameters for rMCD outcomes. Additionally, we developed a directed message passing neural network (D-MPNN) model that combines structural features with calculated and predicted physicochemical properties, demonstrating strong predictive performance for rMCD outcomes. This model facilitates the early prioritization of peptide drug candidates for rMCD assays during the candidate selection phase and accelerates hit-to-lead and lead optimization by identifying peptides within a series that exhibit minimal rMCD liabilities. Notably, the D-MPNN model outperformed traditional <i>in silico</i> property-based calculators in our prospective validation study. Furthermore, to address species-specific SAR, we also established a human MCD (hMCD) assay, revealing an 80% concordance in MCD outcomes between species. This hMCD assay identifies the MCD liabilities of compounds that differ from those in rats, indicating potential risks in humans. This comprehensive <i>in silico</i> and <i>in vitro</i> approach enables drug discovery teams to advance drug candidates that are free from MCD liability in a resource-efficient manner, thereby increasing the likelihood of success in both nonclinical and clinical studies.