Data-Driven Machine Learning Framework for the Regulation of Protein Adsorption on Surfaces

Area of Science:

• Surface Science
• Biomaterials Engineering
• Computational Chemistry

Background:

• Protein adsorption is complex, influenced by protein, surface, and environmental factors.
• Predicting and controlling protein adsorption remains a significant challenge in surface science.

Purpose of the Study:

• To develop a data-driven machine learning framework for predicting protein adsorption.
• To systematically evaluate determinants of protein adsorption and their impact on adsorption behavior.

Main Methods:

• Utilized the AutoGluon framework with seven critical descriptors.
• Constructed three optimal models, with WeightedEnsemble_L2 (WE_L2) showing superior performance.
• Employed SHAP analysis to quantify descriptor contributions and their effects on protein adsorption.

Main Results:

• The WE_L2 model accurately predicted Bovine Serum Albumin (BSA) adsorption on self-assembled monolayer (SAM) surfaces.
• The model captured complex Human Serum Albumin (HSA) adsorption behaviors, including bilayer formation and electrostatic effects.
• SHAP analysis revealed promoting and inhibiting effects of descriptors on protein adsorption.

Conclusions:

• Interpretable machine learning models enable precise prediction of protein adsorption amounts.
• This framework offers a powerful tool for regulating interfacial protein adsorption in biomaterial applications.