This study combines kernel machine design, lab automation, and experimental characterization techniques to develop an Tunable Nanoparticle platform guided by AI (TuNa-AI).
- Introduced the concept of tuning drug-excipient nanoparticles by adjusting stoichiometry during synthesis.
- Developed an automated, high-throughput data generation workflow.
- Constructed a bespoke kernel machine (figure below) to guide the design of tunable nanoparticles.
- Enabled encapsulation of previously inaccessible drugs by rational increase of excipient.
- Computationally guided the reduction of excipient to prepare potent and safer nanoparticles.
Supervised machine learning runs using algorithms from scikit-learn, XGBoost, Chemprop (V 1.5.1) and GraphGPS. The e3fp package facilitates the efficient calculation of Tanimoto similarity. RDKit and DescriptaStorus are chemoinformatics libraries designed for molecular featurization. Additionally, tqdm provides a convenient way to visually monitor job progress.
The available data sources include:
- Historical drug-excipient nanoparticle data
- High-throughput screening data with various drug/excipient molar ratios
- Structure information of investigated chemicals
This folder includes core functions that underlie the analysis pipeline and executable examples for users to run:
- Retrospective evaluation of machine learning and deep learning
- Prospective prediction of new compounds and pairs
- Real-time nanoparticle prediction through a webserver
The copyrights of the software are owned by Duke University. As such, two licenses for this software are offered:
- An open-source license under the GPLv2 license for non-commercial academic use.
- A custom license with Duke University, for commercial use or uses without the GPLv2 license restrictions.
If you find this work or code useful, please cite our ACS Nano paper.