Key Points from the ExPreSo Research Paper

• First ML Tool for Excipient Prediction
  The study introduces ExPreSo, the first machine-learning algorithm designed to predict stabilizing excipients for biopharmaceutical formulations.

• Built on 335 FDA-Approved Formulations
  ExPreSo was trained on one of the largest curated datasets of commercially approved protein/peptide drug formulations, covering diverse drug modalities including mAbs and non-mAbs.

• Strong Predictive Accuracy for 9 Common Excipients
  The model accurately predicts the presence of nine widely used excipients (e.g., histidine, polysorbate 80, sucrose), showing ROC-AUC values >0.7 for most in cross-validation tests.

• Resilient Across Model Variants
  Whether using interpretable features, fast models, or protein-only input data, ExPreSo consistently showed similar predictive performance with minimal overfitting.

• Accelerates Formulation Development
  ExPreSo supports early formulation decision-making by reducing experimental workload, enabling smarter excipient pre-selection, and lowering risks and costs in formulation screening.