Combining molecular and cell painting image data for mechanism of action prediction

In this work we developed a model capable of predicting mechanism of action (MoA) using both structural information from chemicals and morphological information from cell paining images.

---

Setting up the environment

To create and activate the environment.

conda env create -f environment.yml
conda activate chem-moa
pip install -q git+https://github.com/huggingface/transformers.git

To export the conda environment to jupyter notebook.

python -m ipykernel install --user --name=chem-moa

---

Our work contains three stages.

Stage 1: Predicting MoA using compound structure based model based on molecular data

Folder name: Compound_structure_based_models
The models explored are given below.

• Multi-Layer Perceptron (MLP)
• Graph Convolutional Network (GCN)
• Convolutional Neural Network (CNN)
• Long Short-Term Memory (LSTM) without data augmentation
• LSTM with data augmentation
• Traditional machine learning algorithms
  • We explored random forests, light gradient boosting machines, cat boost, k-nearest neighbors classifiers, logistic regression, bagging, stacking, voting, and adaboost.

Stage 2: Predicting MoA using cell morphology based model based on image data

Folder name: Image_based_model

• Efficient net

Stage 3: Predicting MoA using global model based on the integration of molecular data and image data

Folder name: Cell_morphology_based_model_and_global_model

• Global model (CNN + MLP model)

---

Citation

Please cite:

Guangyan Tian, Philip J Harrison, Akshai P Sreenivasan, Jordi Carreras-Puigvert, Ola Spjuth, Combining molecular and cell painting image data for mechanism of action prediction, Artificial Intelligence in the Life Sciences, Volume 3, 2023, 100060,ISSN 2667-3185, https://doi.org/10.1016/j.ailsci.2023.100060.
Status: Published