Geometric Deep Learning for Protein Structure Data with PyTorch Lightning

advanced bioinformatics
live training

Geometric Deep Learning for Protein Structure Data with PyTorch Lightning

Target Audience:
VIB PhD Student
VIB Postdoc
VIB Staff Scientist
VIB Group leader & Expert
VIB Technical support
Flemish Academics & Reseachers
Non-Flemish Academics & Researchers
Industry professionals
Location:

Provinciehuis Leuven

Duration:

6 March 9:30h-17h

FacebookTwitterLinkedin

General context

In this course we will develop a geometric deep learning model that takes a protein structure as input and predicts binding site residues. Through the process, we will cover the basics of data processing, tracking performance, hyperparameter configuration, training and validation of a production-level deep learning model using PyTorch Lightning.

Objectives
  • Construct featurised graphs from protein structures

  • Create a geometric deep learning (GDL) architecture by combining off-the-shelf graph layers

  • Efficiently load and batch graph representations

  • Implement evaluation metrics and tracking specific to the task at hand

  • Train and checkpoint the model, log and visualise validation performance

Approach

We ask the participants to use their own laptop.

Event intended for

Python programmers with some experience in machine learning and/or protein structure data.

Required skills

Python

Trainers

Janani Durairaj

I’m a postdoctoral researcher in Torsten Schwede’s Protein Bioinformatics in 3D research group at the Biozentrum, University of Basel. I work at the interface of protein structural bioinformatics and machine learning, developing efficient algorithms and software to represent, explore and understand protein structures for applications ranging from enzyme engineering to drug discovery.

Program

Graph construction from protein structures with PyTorch-geometric
-
GDL architecture with PyTorch-geometric and LightningModule
-
Loading and batching graph data with LightningDataModule
-
Implementing metrics for tracking performance
-
Hyperparameter configuration, training, and checkpointing
-
Wrap-up
-

Practical info

Location & Venue

06 March 2024

Leuven - Provinciehuis Vlaams-Brabant

Provincieplein 1
3010 Leuven
Belgium

Public transport

06 March 2024

Leuven - Provinciehuis Vlaams-Brabant
Public transport

Leuven is easy to reach by train from all Belgian and European cities. Leuven is a 20-minute train ride from Brussels, 45 minutes from Antwerp, 50 minutes from Liège, 1.5 hours from Paris and 2.5 hours from London, Amsterdam or Cologne.

www.belgianrail.be

Route description

06 March 2024

Leuven - Provinciehuis Vlaams-Brabant
Parking

Leuven is located at the junction of two important highways: the E40 and the E314. The 'P' signposts will guide you to the different car parks. If you are using a GPS, please enter “Tiensevest” and not “Provincieplein 1”.

Car parks in Leuven

Site plan

Venue contact

06 March 2024

Leuven - Provinciehuis Vlaams-Brabant
Location contact

+32 16 26 70 15

provinciehuis@vlaamsbrabant.be