Skip to content

Using Chemception to predict the activity of molecules against aid_686978. Taking this further to visualize kernels decision process through Grad-CAM.

Notifications You must be signed in to change notification settings

Wa-lead/Virtual_screnning_using_CNNs

Repository files navigation

Drug Discovery using Virtual Screening

This project aims to develop a computational approach for drug discovery using virtual screening techniques. Virtual screening is a computational method that involves screening large chemical libraries to identify potential drug candidates. It can significantly reduce the time and cost required for the initial stages of drug discovery.

CNN_Animation.mp4

Project Overview

The project consists of several steps:

  • Data Preparation: The project utilizes a dataset of chemical compounds and their corresponding biological activities. The dataset is preprocessed and featurized using the ChemCeption model.

Alt text

  • Model Building: The ChemCeption model, which is based on the InceptionV3 architecture, is used to predict the biological activity of the chemical compounds. The model is trained using the preprocessed data.
  • Model Evaluation: The trained model is evaluated using a separate test dataset. The performance of the model is measured using metrics such as ROC curve and AUC score.

Alt text

  • Kernel Visualization: The project includes techniques for visualizing the model's decision-making process, such as kernel visualization and Grad-CAM.

Alt text

Alt text

Hyperparameter Optimization

We used the Hpbandster package for optimizing the model, which combines hyperband and bayesian optimization.

Alt text

Installation

To run the project, follow these steps:

Clone the repository:

git clone https://github.com/your_username/project.git

Install the required dependencies:

pip install -r requirements.txt

Evaluation

The project achieves an ROC score of 0.69, indicating a good performance in predicting the biological activity of chemical compounds.

Usage

Run the inference.ipynb Capture-2023-05-17-132002

About

Using Chemception to predict the activity of molecules against aid_686978. Taking this further to visualize kernels decision process through Grad-CAM.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published