Related Experiment Video
Updated: Jun 26, 2026
Computational Prediction of Amino Acid Preferences of Potentially Multispecific Peptide-Binding Domains Involved in Protein-Protein Interactions
Published on: January 26, 2024
Machine learning algorithms to predict spray dried protein/peptide formulations
Liping Wei1, Jiayin Deng2, Yang Sun1
1Wuya College of Innovation, Shenyang Pharmaceutical University, No. 103, Wenhua Road, 110016 Shenyang, China; Joint International Research Laboratory of Intelligent Drug Delivery Systems, Ministry of Education, 110016 Shenyang, China.
Machine learning models accurately predict properties of spray-dried protein and peptide formulations, accelerating drug development. This approach optimizes process parameters and reduces trial-and-error for stable biomacromolecule delivery.
Area of Science:
- Biopharmaceutical formulation
- Computational chemistry
- Process engineering
Background:
- Proteins and peptides show therapeutic potential but face stability challenges in drug development.
- Spray drying is a method to stabilize these biomacromolecules into solid formulations.
- Current spray drying methods rely on extensive trial-and-error, increasing resource demands.
Purpose of the Study:
- To develop machine learning (ML) models for predicting key properties of spray-dried protein and peptide powders.
- To accelerate formulation development and optimize spray drying process parameters.
- To identify critical factors influencing the properties of spray-dried protein/peptide formulations.
Main Methods:
- Collected data on yield, particle size, residual solvent content, solid-state properties, and aggregation for spray-dried protein/peptide powders.
- Utilized various molecular descriptors for model building.
- Tested seven ML algorithms, including Light Gradient Boosting Machine (LightGBM) and logistic regression.
- Validated model generalizability using alpha-lactalbumin formulations.
Main Results:
- LightGBM demonstrated superior performance in regression tasks, especially for residual solvent content (MAE=0.841).
- Logistic regression excelled in predicting solid-state characteristics and aggregation.
- Feature importance analysis highlighted protein type, excipients, processing, and environmental conditions as critical factors.
- Experimental validation showed good predictive accuracy for yield, particle size, residual solvent content, solid states, and aggregation.
Conclusions:
- Machine learning offers a powerful, data-driven approach to streamline the development of spray-dried protein formulations.
- This methodology provides a material- and time-saving solution for optimizing spray drying processes.
- The developed ML models can serve as a valuable reference for future biopharmaceutical formulation studies.
