Predicting-Drug-Side-Effects

Code for the paper "Enhancing Drug Side-Effect Prediction With Word Embeddings And Label Taxonomy". We enhance an MLP with supplementary information from word embeddings and label taxonomy to further exploit label correlation in the high-dimensional, multi-label classification problem of in silico drug side-effect prediction.

How To Use

1. Configure hyperparameters in code/hyperparameters.yaml.

2. Run the respective function in main.py
   2a: Training and validation scripts for the various models from the paper:

   • run_mlp_baseline(): For Baseline MLP, from Section 2.5.1
   • run_mlp_concat(): For MLP Concat, from Section 2.5.2
   • run_mlp_mega_concat(): For MLP Mega Concat, from Section 2.5.2
   • run_mlp_concat_w_se_embeds(): For MLP Concat With Side-effect Embeddings, from Section 2.5.2
   • run_mlp_concat_w_soc_labels(): For MLP Concat With SOC Labels, from Section 2.5.2
   • run_mlp_mega_concat_cheat(): For MLP Mega Concat (Cheated Model), from Appendix B.2

   2b: Generate the dataset to use
   (NOTE: Default datasets are already in the repo. You do not have to run these to run the scripts above.):

   • generate_soc_data(type, aggregate, randomise): See documentation for how to use
   • generate_se_embedding(vec_path): See documentation for how to use

Model Architectures

MLP Baseline Model Architecture	MLP Concat Model Architecture
MLP Mega Concat Model Architecture	MLP Mega Concat (Cheated) Model Architecture
MLP Concat With Side-effect Embeddings Model Architecture	MLP Concat With SOC Labels Model Architecture