Unveiling protein corona composition: predicting with resampling embedding and machine learning.

Area of Science:

• Biomaterials Science
• Nanotechnology
• Computational Biology

Background:

• Nanoparticles (NPs) interact with biological fluids, forming a protein corona (PC).
• Accurate PC prediction is vital for assessing biomaterial osteoinductivity and guiding NP design.
• Existing machine learning models struggle with imbalanced PC data and extreme values, limiting prediction accuracy.

Purpose of the Study:

• To develop an improved machine learning approach for predicting protein corona composition.
• To address data imbalance issues in protein corona prediction models.
• To enhance the accuracy of predicting nanoparticle-protein interactions for biomaterial applications.

Main Methods:

• Introduction of resampling embedding techniques to handle imbalanced protein corona data.
• Evaluation of various machine learning models, with a focus on the Random Forest (RF) model.
• Ablation experiments and verification using label-free quantification of four different NPs (HA, TiO2, SiO2, Ag).

Main Results:

• The proposed resampling embedding method improved prediction accuracy, achieving an R² of 0.68 (approx. 10% improvement) and an RMSE of 0.90 (approx. 10% reduction).
• Random Oversampling further enhanced prediction performance for specific NPs, yielding an R² value >0.70.
• Feature analysis identified incubation plasma concentration, PDI, and surface modification as key factors influencing PC composition.

Conclusions:

• Resampling embedding effectively resolves data imbalance issues in protein corona prediction.
• The enhanced RF model provides accurate predictions of protein corona composition.
• This approach facilitates the rational design of nanomaterials with tailored biological interactions.