Code and data to study and predict drug-excipient co-aggregation for nanoparticle formation

For more information, please refer to the associated publication: https://www.nature.com/articles/s41565-021-00870-y

If you use this data or code, please kindly cite Reker et al. Nature Nanotechnology 2021 https://doi.org/10.1038/s41565-021-00870-y

Dependencies

The machine learning and the molecular simulations have been enabled through a broad range of available software packages. We have generated two distinct conda environments to enable the implementation of these two pipelines.

Machine Learning

Runs in Python 2.7 using the scikit-learn library as well as RDKit. Optional libraries from OpenPyXL and Joblib enable reading of self-aggregation data from the Shoichet lab as well as parallelization of code for more efficient processing (not implemented in this version of the repository). A fresh conda environment can be set up to run these experiments using

conda create -n python2 python=2.7 
conda activate python2 
conda install scikit-learn
conda install -c rdkit rdkit
conda install -c anaconda openpyxl
conda install -c anaconda joblib

Simulations

Dependent on Python libraries OpenMM, openmoltools, Open Babel, ambermini as a stripped down version of Amber, MDTraj, ParmEd, and RDKit. Also requires an installation of PACKMOL A fresh installation of Ubuntu 18 and anaconda can be set up to run these experiments using:

conda create --name simulations
conda activate simulations
conda install -c omnia -c conda-forge openmm
python -m simtk.testInstallation
conda install -c openbabel openbabel
conda install -c kyleabeauchamp ambermini
conda install -c conda-forge mdtraj
conda install -c omnia parmed
conda install -c omnia openmoltools
conda install -c rdkit rdkit

sudo apt-get install gfortran
wget https://github.com/m3g/packmol/archive/20.010.tar.gz
tar -zxvf 20.010.tar.gz 
cd packmol-20.010 
./configure
make
export PATH="/home/user/packmol-20.010:$PATH"

Descriptions of folders

data

• Structural information of utilized molecules and screening results.

machine learning

• Python code to run predictions of self-aggregation and co-aggregation. All scripts also contain evaluations and bechmarks as well as adversarial control calculations.

simulations

• Code to run and automatically analyze short MD simulations to assess interaction potential of drug molecules and excipients.