Python scripts and jupyter notebooks to accompany the manuscript:
Deep learning models for lipid-nanoparticle-based drug delivery
Authors: Philip J Harrison, Håkan Wieslander, Alan Sabirsh, Johan Karlsson, Victor Malmsjö, Andreas Hellander, Carolina Wählby and Ola Spjuth.
Note: The LNP data used in the scripts and notebooks is not included in this repository, but can be found at https://scilifelab.figshare.com/articles/LNP_drug_delivery_image_data/12482183/1
1. LNP_CNN_data_prep.ipynb
Data preparation to extract the cell-level time-lapse data needed for the CNNs.
2. LNP_CNN_train.ipynb
Training the CNN between time points 1 and 20 in two prediction models (classification and regression) and performing 5-fold cross-validation.
3. LNP_time-series_data_prep.ipynb
Using the trained CNNs create the time-series data required for the LSTM and tsfresh based applications.
4. LNP_LSTM_model_selection.ipynb
200 sample grid search for the best LSTM model arcitecture for each prediction mode and cross-validation fold.
5. LNP_LSTM_train.ipynb
Train the best LSTM models from the model selection and save out predictions on the test set.
6. LNP_tsfresh_efficient_extract_select_PCA.py
Using tsfresh with the "efficient parameters" setting extract and select the relevant time-series features, followed by PCA for dimenion reduction.
7. LNP_tsfresh_efficient_gbm.ipynb
Gradient boosting machine (GBM) grid search based on the time series features derived from (6) and save out predictions on the tests set from the best GBM model.