Graph Generation for Drug Discovery using Python and Keras

Name: Graph Generation for Drug Discovery using Python and Keras
Rating: 4.0 (36 reviews)

Python-based Graph Generation for Molecular Structures using Keras: A Practical Introduction to Neural Network Modeling

Created byKarthik Karunakaran, Ph.D.

Last updated 10/2023

English

What you'll learn

Understand the basics of graph generation and its applications in various fields.
Learn how to manipulate molecular structures using the RDKit library in Python.
Gain proficiency in preprocessing chemical data stored in CSV files.
Develop an understanding of mapping atom symbols and bond types to numerical representations.
Learn to convert SMILES strings into graph representations.
Understand the concepts of Generative Adversarial Networks (GANs) and their application in graph generation.
Implement a Graph Generator using TensorFlow and Keras to generate molecular graphs.
Create a Discriminator model to evaluate the quality of generated graphs.
Learn about the Wasserstein GAN framework for improved GAN training stability.
Gain hands-on experience in training and fine-tuning GAN models for graph generation tasks.
Understand the importance of GPU acceleration and how to configure it for faster computations.
Develop the ability to save and load model weights for future use.
Gain proficiency in generating molecular graphs using the trained GAN model.
Learn to visualize and analyze the generated molecular structures.

Course content

2 sections • 37 lectures • 1h 24m total length

Introduction0:48
Explore molecular structures and the graph generation approach in drug discovery, building generator and discriminator models to craft novel, authentic-appearing molecules while tuning hyperparameters and refining training for better outcomes.
About this Project0:53
Explore how deep learning and graph-based models generate new molecules as small molecular graphs, accelerating drug discovery with the Wgan GP model and a GCN architecture.
Why Should we Learn?2:39
Applications3:21
Explore how generative graph models accelerate drug discovery by exploring chemical space to identify candidate molecules with desired properties, and extend design to materials, catalysts, and personalized medicine.
Python, Keras, and Google Colab1:32

Setup Working Directory1:01
What is qm9.csv file?2:08
Examine the cm9 dataset CSV, where each row represents a molecule with a Smiles string and properties used in quantum chemistry for predicting molecular behavior and graph generation.
What is code.ipynb?0:31
Launch Code0:28
Activate GPU0:36
Enable gpu acceleration to speed up graph generation tasks by switching runtime type to gpu in the hardware accelerator settings, saving changes, and activating a t4 gpu.
Mount Google Drive1:28
Mount Google Drive in Google Colab to access and save files for graph generation projects in drug discovery, after authorizing the drive connection and enabling read-write access.
Installing two Python libraries2:00
Install two python libraries with pip. Use the Rdkit PyPI library for cheminformatics and the pillow library for image processing in the notebook, suppressing output with the -q flag.
Importing several libraries2:17
Import essential libraries for graph generation in drug discovery, including Rdkit chem tools, IPython console and grid image, numpy as NP, and TensorFlow Keras for deep learning.
Disabling the logging functionality1:15
Loading Dataset0:48
Process CSV file2:16
Selects a specific SMILES string0:42
Selects a specific smiles string from a data list by index 1000, prints the result, and highlights using valid indices to avoid errors in Python.
Convert the SMILES string1:47
Mapping atom symbols2:25
Use an atom mapping dictionary to map symbols to integers and back. This enables converting C, N, O, F to and from numeric values for machine learning on molecular data.
Mapping bond types2:34
Map bond types to integers with a bidirectional dictionary, enabling quick lookups from single, double, triple, and aromatic bonds to 0–3 and back for molecular data in computational chemistry.
Constants2:18
Convert a SMILES string to a graph representation4:55
Convert a smiles string to a graph by building an adjacency and features matrix from a molecule object, mapping atom and bond types, and handling disconnected atoms and missing features.
Convert graph representation back into RDKit molecule object5:38
Graph representation0:21
Convert a smile string into a graph representation and reconstruct the chemical structure with Rdkit to visualize the compound for drug discovery.
Converting subset of SMILES data to graph tensors3:13
Convert a subset of smiles data into graph tensors by transforming smiles strings into adjacency and feature matrices, then store as NumPy arrays and inspect shapes.
Defines a generator model4:47
Creates an instance of the GraphGenerator model1:54
Defines a custom graph convolutional layer5:48
Define a custom relational graph conv layer in Keras that performs a relational graph convolution on adjacency and features matrices with configurable units, activation, bias, and initialization for graph generation.
Creates the discriminator model3:55
Build a graph discriminator that distinguishes real from generated graphs using adjacency and features inputs, relational graph convolutional layers, global pooling, dense layers with dropout, and a real-or-generated score.
Creates a discriminator model2:29
Wasserstein Generative Adversarial Network4:12
Sets up a WGAN2:07
Set up a WGAN graph generation model by configuring a generator and discriminator, training steps, and Adam optimizers with a 5e-4 learning rate to improve generated graphs.
Training2:11
Saving and loading the model weights2:35
Save and load the trained graph model weights to a path on Google Drive, with index and data files, enabling graph generation, evaluation, and deployment without retraining.
Sample molecules2:57
Generating molecules1:18
Displaying molecules2:47

Requirements

Basic programming knowledge is recommended, but not mandatory. Familiarity with Python programming will be helpful.
A Google account is required to access Google Drive and Google Colab for practical exercises.
Access to a computer with a stable internet connection is necessary to access online resources and run code in the Google Colab environment.

Description

Are you curious about the world of molecular structures, drug discovery, and generative models? Look no further! This exciting course will take you on a journey through the fascinating field of graph generation and its real-world applications.

In this course, we will start by exploring the basics of molecular representations using SMILES notation and how to convert them into graph structures using the powerful RDKit library. You will learn how to handle and manipulate molecular data efficiently.

Next, we will dive into the realm of generative models, specifically GraphWGAN (Graph Wasserstein Generative Adversarial Network). You will gain an understanding of how GraphWGAN combines the power of generative adversarial networks (GANs) and graph neural networks (GNNs) to create realistic and diverse molecular graphs.

Throughout the course, we will build and train both the generator and discriminator models, learning how they work together to create new molecules that closely resemble real chemical compounds. As we progress, you will discover the art of hyperparameter tuning and optimizing the training process to achieve better results.

But the journey doesn't end there! We will explore various real-world applications of graph generation, particularly in drug discovery and materials science. You will witness how this cutting-edge technology is revolutionizing the pharmaceutical industry, accelerating the process of drug development, and contributing to groundbreaking research.

As we delve into the practical aspects of this course, you will gain hands-on experience working with TensorFlow, Keras, and other essential libraries, honing your skills in machine learning and data manipulation.

By the end of this course, you will be equipped with the knowledge and skills to tackle graph generation tasks independently. You will also have a portfolio of impressive projects that showcase your expertise in this exciting field.

The job prospects in the world of graph generation and artificial intelligence are booming! Industries such as pharmaceuticals, biotechnology, and materials science are actively seeking professionals who can leverage the power of graph generation models for innovative research and product development. So, this course can open doors to exciting job opportunities and career growth.

So, if you are ready to embark on a journey that merges chemistry, artificial intelligence, and real-world impact, join us for this thrilling course on Graph Generation using GraphWGAN. Let's uncover the secrets of molecular structures and unleash the power of generative models together!

Enroll now and let the adventure begin!

Who this course is for:

Beginners in Machine Learning: If you're new to the field of machine learning and want to learn how to generate molecular graphs using advanced techniques, this course will provide a gentle and comprehensive introduction.
Aspiring Data Scientists: If you're aspiring to become a data scientist or work in the domain of chemistry-related data analysis, this course will equip you with valuable skills in graph generation and neural networks.
Chemistry Enthusiasts: If you have a background or interest in chemistry and want to explore how machine learning can be applied to molecular structures and graph generation, this course will bridge the gap between chemistry and AI.
Python Programmers: If you are already familiar with Python programming and want to expand your knowledge into the realm of graph-based machine learning, this course will offer a structured pathway.
Students and Researchers: Whether you're a student working on a project or a researcher looking to integrate graph generation into your work, this course will provide practical skills and knowledge to enhance your capabilities.
Lifelong Learners: If you're simply curious about the intersection of machine learning, chemistry, and graph generation, this course welcomes learners of all backgrounds and experiences.

Graph Generation for Drug Discovery using Python and Keras

What you'll learn

Explore related topics

Course content

Introduction5 lectures • 9min

Model Generation, Training and Prediction32 lectures • 1hr 16min

Requirements

Description

Who this course is for: