AI-Powered Prediction of Nanoparticle Pharmacokinetics: A Multi-View Learning Approach

TL;DR

The paper tackles the unpredictability of nanoparticle pharmacokinetics under data scarcity by introducing a multi-view deep learning framework that injects domain priors (size and charge) into a cross-attention model and augments it with ensemble learners RF and XGBoost. By combining primary NP data with engineered priors and extracted features, the approach achieves superior predictive accuracy for four PK endpoints and yields interpretable insights into biodistribution drivers, while bridging ML with PBPK modelling for data-efficient, precision nanomedicine. Extensive benchmarking on a mouse PK dataset demonstrates statistically significant improvements over baselines, and ablation studies underscore the value of priors and multi-model ensembles. The work suggests a practical pathway for AI-assisted NP design and pre-screening that could reduce in vivo experimentation and accelerate translational nanomedicine research.

Abstract

The clinical translation of nanoparticle-based treatments remains limited due to the unpredictability of (nanoparticle) NP pharmacokinetics$\unicode{x2014}$how they distribute, accumulate, and clear from the body. Predicting these behaviours is challenging due to complex biological interactions and the difficulty of obtaining high-quality experimental datasets. Existing AI-driven approaches rely heavily on data-driven learning but fail to integrate crucial knowledge about NP properties and biodistribution mechanisms. We introduce a multi-view deep learning framework that enhances pharmacokinetic predictions by incorporating prior knowledge of key NP properties such as size and charge into a cross-attention mechanism, enabling context-aware feature selection and improving generalization despite small datasets. To further enhance prediction robustness, we employ an ensemble learning approach, combining deep learning with XGBoost (XGB) and Random Forest (RF), which significantly outperforms existing AI models. Our interpretability analysis reveals key physicochemical properties driving NP biodistribution, providing biologically meaningful insights into possible mechanisms governing NP behaviour in vivo rather than a black-box model. Furthermore, by bridging machine learning with physiologically based pharmacokinetic (PBPK) modelling, this work lays the foundation for data-efficient AI-driven drug discovery and precision nanomedicine.

Figures (3)

• Figure 1: Overview of the multi-view AI framework. The primary dataset consists of NP physicochemical properties and pharmacokinetic measurements, while prior knowledge is incorporated through engineered features. A cross-attention deep learning model integrates these two data sources, and ensemble learning improves overall prediction robustness. Created in BioRender. Khakpour, A. (2025) https://BioRender.com/c18t496
• Figure 2: Saliency gradient graphs to estimate the importance of each primary (left column) and secondary (right column) features on model predictions (KTRESmax on first row, KTRESn on second row, KTRES50 on third row, KTRESrelease on fourth row) by measuring the sensitivity of outputs to changes in inputs.
• Figure 3: Overview of the proposed multi-task multi-view model architecture, illustrating the cross-attention and MLP branches that are then concatenated to produce a final prediction. In this figure, the primary dataset represents $\mathcal{D}$, and the secondary dataset $\mathcal{D}_{\text{P+FE}}$. Created in BioRender. Khakpour, A. (2025) https://BioRender.com/r59r002