AI-Powered Prediction of Nanoparticle Pharmacokinetics: A Multi-View Learning Approach
Amirhossein Khakpour, Lucia Florescu, Richard Tilley, Haibo Jiang, K. Swaminathan Iyer, Gustavo Carneiro · 2025
Read AI-Powered Prediction of Nanoparticle Pharmacokinetics: A Multi-View Learning Approach by Amirhossein Khakpour, Lucia Florescu, Richard Tilley, Haibo Jiang, K. Swaminathan Iyer, Gustavo Carneiro (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-13798.txt.
Introduction
Amirhossein Khakpour and co-authors build in 2025 a multi-view learning tool to predict how nanoparticles travel in the body. Clinical use of nano drugs stalls because body travel is hard to foresee. Small data sets and complex body interplay block plain learning tools.
Their frame feeds size and charge facts into an attention step for context-aware picks. An ensemble joins deep nets with two tree tools for firmer grades. The blend beats past models in the reported tests. Trait study shows which particle traits drive spread, and links to body-based models point ahead.
Catalog record
| CREATOR | Amirhossein Khakpour, Lucia Florescu, Richard Tilley, Haibo Jiang, K. Swaminathan Iyer, Gustavo Carneiro |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-13798
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
AI-Powered Prediction of Nanoparticle Pharmacokinetics: A Multi-View Learning Approach Amirhossein Khakpour, Lucia Florescu, Richard Tilley, Haibo Jiang, K. Swaminathan Iyer, Gustavo Carneiro Abstract The clinical translation of nanoparticle-based treatments remains limited due to the unpredictability of (nanoparticle) NP pharmacokinetics—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. keywords Nanoparticle, Cross-Attention, Ensemble learning, SMOTE data augmentation 1 Introduction NP-based drug delivery systems have revolutionized modern medicine, offering highly targeted therapies with reduced side effects Peer et al. 2020 ; Brigger et al. 2012 ; Gavas et al. 2021 . NP imaging agents, ranging from magnetic NP contrast agents to light-emitting quantum dots, have the potential to transform how we medically diagnose and monitor serious diseases within the body Wang and Thanou 2010 ; Sztandera et al. 2018 . However, the clinical translation of NP-based treatments and imaging remains limited due to the unpredictability of NP pharmacokinetics—how NPs distribute, accumulate, and clear from the body Davis et al. 2008 ; Yuan et al. 2023 . Traditional PBPK models attempt to predict NP behavior, but they rely on predefined equations and assumptions that fail to fully capture the complexity of biological variability, tumor microenvironments, and organ interactions Chen et al. 2023 ; Wang et al. 2013 . This gap in predictive accuracy has hindered the development of optimized NP formulations for drug delivery Yuan et al. 2023 . Machine learning has emerged as a powerful tool to improve pharmacokinetic predictions by leveraging large datasets to learn complex relationships between NP properties and their biodistribution He et al. 2024 ; Chou et al. 2023 . However, existing AI-based approaches face two major challenges: limited data availability, as pharmacokinetic studies involve resource-intensive in vivo experiments, and lack of prior knowledge integration, meaning that traditional ML models treat each dataset independently without leveraging fundamental insights from nanomedicine Jyakhwo et al. 2024 ; Rezvantalab et al. 2024 . As a result, existing ML models struggle to generalize beyond specific experimental conditions, limiting their applicability in real-world clinical scenarios Chou et al. 2023 ; Serretiello et al. 2024 . 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 To address these challenges, we introduce a multi-view AI framework that combines prior knowledge of NP properties with cross-attention deep learning to enhance pharmacokinetic predictions, even in data-limited scenarios Lin et al. 2022 ; Du et al. 2018 –see Fig. 1 . This multi-view framework approach enables the model to leverage nanomedical domain-specific insights such as the importance of NP size, charge, shape, and organ biodistribution patterns to refine its predictions beyond purely data-driven learning Choi et al. 2007 ; Sonavane et al. 2008 . The cross-attention mechanism assigns importance to key biological and physicochemical properties, improving interpretability and accuracy Vaswani 2017 . The model also integrates an ensemble learning approach, combining deep learning with classical machine learning models such as RF and XGB to enhance robustness and predictive performance Ganaie et al. 2022 . Building on this framework, the proposed model predicts four key NP pharmacokinetic parameters— {KTRESmax} , {KTRESn} , {KTRES50} , and {KTRESrelease} , which describe NP biodistribution across organs, including the maximum uptake rate constant, Hill coefficient, time to reach half-max uptake, and release rate constant in tumour cells Chou et al. 2023 . By linking machine learning with PBPK modelling, this approach not only improves pharmacokinetic predictions but also lays the foundation for AI-assisted drug formulation, NP optimization, and precision nanomedicine Noorain et al. 2023 ; Lin et al. 2022 . This study presents the first AI-powered pharmacokinetic model that systematically integrates prior knowledge with deep learning, setting a new benchmark for NP biodistribution modelling He et al. 2024 ; Chen et al. 2023 ; Lin et al. 2022 . 1.1 Background PBPK models simulate NP absorption, metabolism, and elimination in various organs Yuan et al. 2023 ; Wang et al. 2013 ; Chou et al. 2023 ; Li et al. 2017 . However, their reliance on simplified assumptions limits their accuracy in complex environments Chen et al. 2023 ; Yuan et al. 2023 ; Wang et al. 2013 . PBPK models struggle to incorporate dynamic changes in NP behaviour, such as real-time cellular uptake or targeting influenced by surface modifications Yuan et al. 2023 ; Cheng et al. 2020 . These challenges have led to growing interest in AI-driven approaches He et al. 2024 ; Chou et al. 2023 . AI models have increasingly been integrated with PBPK modelling to reduce reliance on in vivo animal data Chou et al. 2023 . Despite these advances, challenges remain due to variability in NP behaviour across biological environments Chen et al. 2023 ; Yuan et al. 2023 . The unpredictable uptake and elimination of NPs add complexity to PBPK modelling Davis et al. 2008 ; Zhang et al. 2016 ; Wang et al. 2013 . While AI models leveraging meta-analysis data have improved NP distribution predictions Chen et al. 2023 , meta-analysis alone does not account for critical NP properties such as size and charge, which significantly influence organ-specific biodistribution. Feature engineering improves model performance through polynomial, logarithmic, and exponential transformations Brambor et al. 2006 ; Kohli and Mahajan 1991 ; Fama and French 1992 ; Huber 1985 ; Box and Cox 1964 . By integrating domain-specific NP properties with AI-driven approaches, models can better capture NP pharmacokinetics, addressing limitations of both PBPK and traditional machine learning methods. 2 Results and Discussion The proposed multi-view AI framework was evaluated against baseline models, including standard deep learning architectures and traditional ML methods such as RF and XGB, using performance metrics like mean squared error and R^{2} values to assess prediction accuracy. In this section, we first describe the dataset and experimental setup, followed by details of the ensemble method. We then present the results, conduct an ablation study, and explore the impact of multi-task learning on training the four pharmacokinetic parameters. 2.1 Dataset Our study utilizes a publicly available pharmacokinetic dataset consisting of 378 samples in mice Chou et al. 2023 . Each sample includes NP physicochemical properties such as size,