CrossFeat

CrossFeat: A transformer-based cross-feature learning model for predicting drug side effects frequency DOI :

Requirements

• Python == 3.6.12
• PyTorch == 1.2
• Numpy == 1.19.2
• Scikit-learn == 0.24.0

Files

• data

  • sample_5fold_idx.pkl: This file includes drug and side effect indices in the PS1 and PS2 datasets for training, validation, and test set per fold. Additionally, it incorporates frequency values for drug-side effect pairs.
  • PS3.pkl: This file contains drug and side effect indices of PS3 dataset, along with corresponding frequency values for drug-side effect pairs.
  • drug_id.csv: .csv file with drug names
  • side_id.csv: .csv file with side effect names

  The four input files below were preprocessed following the method by Zhao et al. (DOI: https://doi.org/10.1093/bib/bbab449)

  • drug_word.pkl: The drug Mol2vec word vector matrix.
  • drug_fingerprint_similarity.pkl: The Jaccard score similarity matrix of drug fingerprint.
  • side_sem.pkl: The side effect semantic similarity matrix.
  • side_word.pkl: The side effect GloVe word vector matrix.

• main

  • main.py: This Python file allows for training and testing the CrossFeat using a five-fold cross-validation setup.
  • network_main.py: This file contains each network comprising CrossFeat.
  • utils.py: This file contains the utilities for CrossFeat.

• FAERS dataset: Collectively, these files provide diverse input data for training and evaluating models using the FAERS dataset from the 4th quarter of 2012 to the 2nd quarter of 2023.

  • SEname.tsv: Contains the names of side effects.
  • age_sample_5fold_idx_VL.zip: File with sample indices divided into training, validation, and test folds for 5-fold cross-validation, specifically related to patient age.
  • drug_fingerprint1024_jaccard.pkl: Holds Jaccard similarity values for drug fingerprints.
  • drug_mol2vec300d.csv: Includes 300-dimensional mol2vec vectors for drugs.
  • drugname.csv: Contains the names of drugs.
  • gender_age_sample_5fold_idx_VL.zip: Similar to the age file, it also includes sex information for 5-fold cross-validation.
  • gender_sample_5fold_idx_VL.zip: Sample index file considering sex for 5-fold cross-validation.
  • se_DAG_sim.csv: File representing the similarity between side effects using a Directed Acyclic Graph (DAG) approach.
  • se_glove_wordvector.csv: Holds word vectors for side effects obtained using the GloVe algorithm.

Contact

If you have any questions or suggestions regarding CrossFeat, please feel free to contact Bin Baek at [email protected].