Gx2Mol

A PyTorch implementation of “Gx2Mol: De Novo Generation of Hit-like Molecules from Gene Expression Profiles via Deep Learning“. The paper is under review by Neural Computing and Applications.

Components

• A VAE is used to extract features from gene expression profiles
• and an LSTM is utilized to generate hit-like molecules with the conditioned features.

Objectives

Gx2Mol aim to

• generate hit-like molecules from gene expression profiles.
• generate therapeutic molecules from patients’ disease profiles.

Environment Installation

Execute the following command:

$ conda env create -n gx2mol_env -f gx2mol_env.yml
$ source activate gx2mol_env 

File Description

• datasets
  • LINCS/mcf7.csv: The training and validation datasets, which consist of gene expression profiles of the MCF7 cell line treated with 13,755 molecules, were used.
  • tools floder
• main.py:: Define the main function for training the GeneVAE and SmilesDecoder models.
• GeneVAE.py: Defines a VAE model for extracting features from gene expression profiles.
• train_gene_vae.py: Code for training the GeneVAE model.
• SmilesDecoder.py: Defines a decoder model for generating SMILES strings with extracted gene features.
• train_smiles_vae.py: Code for training the SmilesDecoder model.
• utils.py: Defines other functions used in Gx2Mol.

Experimental Reproduction

• STEP 1: Pretrain GeneVAE:

$ python main.py --train_gene_vae --cell_name 'mcf7'

• STEP 2: Test the trained GeneVAE:

$ python main.py --test_gene_vae --cell_name 'mcf7'

• STEP 3: Train SmilesDecoder:

$ python main.py --train_smiles_decoder 

• STEP 4: Test SmilesDecoder:

$ python main.py --test_smiles_decoder

• STEP 5: Generate molecules for the 10 ligands:

$ python main.py --generation --protein_name 'AKT1'

• STEP 6: Calculate Tanimoto similarity between a source ligand and generated SMILES strings:

$ python main.py --calculate_tanimoto