AntiBioGen: AI-Driven Antibiotic Discovery

Overview

This project aims to discover and design novel antibiotic compounds targeting specific proteins. It features tools for predicting toxicity, generating antibiotic candidates, and obtaining 3D structures of the designed molecules. Leveraging neural networks for accurate toxicity prediction and efficient antibiotic generation, along with genetic algorithms for design optimization, this project enhances the potential for developing effective and safe antibiotic solutions.

Usage

Preparing Data

1. FASTA Sequence: Ensure your target protein is in FASTA format (amino acid sequence).

2. Neural Networks: You need two neural networks:

   • Toxicity Prediction: Use cnn_affinity.py to train or utilize the pre-trained model.
   • Antibiotic Generation: Use generate_rnn.py to train or utilize the pre-trained model.

   Alternatively, use the pre-trained models located in the definitive_models folder.

3. Data: Use data from databases such as Chembl, PubChem, or other relevant sources.

CNN Usage (Affinity Prediction)

To predict toxicity using the CNN model, use check_affinity.py. You need to specify the path to the model, the path to the data, and the path to the target protein. The program will return the toxicity of the designed bioinsecticides using the calculate_affinity function.

RNN Usage (Generation)

To generate antibiotic compounds using the RNN model, use pretrained_rnn.py. You need to specify the path to the model, the path to the data, and the path to the target protein. The program will return the designed bioinsecticides using the generate function.

Combination of Models

For combining both models (generation and toxicity prediction), use affinity_with_target_and_generator.py. You need to specify the path to the model, the path to the data, and the path to the target protein. The program will generate antibiotic compounds and filter out those exceeding the specified toxicity limit. You can also specify a path to check generated molecules.

Installation

Via Git Clone:

git clone https://github.com/RubenVG02/AntibioticsDiscovery.git

Via Lastest Release:

https://github.com/RubenVG02/AntibioticsDiscovery/releases

To use this project, you need to have Python 3.7 or higher installed. Then, you need to install the following libraries:

• Keras
• Tensorflow
• Numpy
• Pandas
• Matplotlib

To install the required libraries, use:

pip install requirements.txt

Authors

• @RubenVG02

Features

• Design of new antibiotic compounds based on the target protein
• Predicting the toxicity of the designed antibiotics
• Obtaining CSV files and screenshots of the results
• Fast and easy to use

Future Improvements

• Add more databases to the CNN
• Add more databases to the RNN
• Use GA to improve the efficiency of the generation
• Directly obtain the 3D structure of the designed compounds

License

MIT

Acknowledgements

• Keras
• Tensorflow
• Numpy
• Pandas