TorchDrug: A powerful and flexible machine learning platform for drug discovery

Description

The practice of building machine learning models for various tasks in drug discovery with our platform, TorchDrug.

Recently, there is a growing interest in developing machine learning techniques for drug discovery problems, such as property prediction, de novo molecule design and biomedical knowledge graph reasoning. While many algorithms have been proposed to solve these problems, implementing these algorithms requires a lot of domain knowledge and most existing projects are implemented from scratch. As a result, it is often challenging to either reproduce, compare, integrate or scale up these algorithms for drug discovery.

In this Masterclass, we present a platform, TorchDrug, to promote reproducibility and accelerate development of machine learning techniques for drug discovery. TorchDrug covers techniques from graph neural networks, geometric deep learning, knowledge graphs, deep generative models to reinforcement learning. It provides solutions to 5 drug discovery tasks, backed by a large number of machine learning models. It also exports a hierarchical interface from high-level tasks, mid-level models and layers, to low-level data structures, which can be easily customized and extended by users.

After this Masterclass, participants will know how to quickly build machine learning models for various tasks in drug discovery using TorchDrug, such as property prediction, molecular graph generation and biomedical knowledge reasoning. They will also know how to customize layers, models, and tasks provided in the platform, which we believe will greatly simplify the code work when designing new models for drug discovery. We will also cover some promising directions in the field of drug discovery, which serve as great starting materials for beginners in this field.

Key Topics

• Motivation & Inspiration of TorchDrug
• Data Structures & Operations in TorchDrug
• Basic layers and models covered in TorchDrug
• Hands-on codes on drug discovery tasks
  • Property Prediction
  • Pretraining & Finetuning
  • Graph Generation and Synthesis Planning
  • Biomedical knowledge graph reasoning

Target Audience

• Anyone interested in drug discovery
• Researchers coming from the background of computer science
• Researchers coming from the background of computational biology
• Machine learning beginners who want to find an exciting field to dive in and learn some machine learning algorithms

Goals

Get hands-on experience using Torchdrug to build machine learning models for drug discovery

Session outline:

• Motivation & Inspiration of TorchDrug
• Basic usage of TorchDrug
• Hands on tutorials on writing Torchdrug codes for several drugdiscovery tasks
  • Property Prediction
  • Pretraining & Finetuning
  • Graph Generation and Synthesis Planning
  • Biomedical knowledge graph reasoning
• Closing remarks

Format

• The tutorial will be interactive.
• Speakers will cover the basic usage of TorchDrug as well as the example codes for several drug discovery tasks using Google Colab.
• Participants will be asked to answer some simple questions in Google Colab and write toy codes to solve some questions using the API of TorchDrug.
• Participants can report the bugs and unexpected results they meet in the Google Colab and we will help them fix the issues.

Level

Intermediate - Advanced

Prerequisite Knowledge

• Basic machine learning knowledge
• Google Colab will be used

You need an access pass to attend this session: Diversity Access Pass or Full Access Pass apply