Programme

15th & 16th October – Hackathon & Workshops

Join us for an exciting pre-conference hackathon and tutorial sessions. These events are designed to equip you with the essential skills and hands-on experience to revolutionise genomic research with AI.

AI Hackathon: Innovate and Collaborate

Challenge yourself with real-world questions or bring your own idea.

The Hackathon will run in-parallel with the Tutorials on day 1 and 2 of the conference. Here's a peek at some of the topics and question that will be explored.

"To streamline the curation process, we aim to leverage AI models to predict the correct order and orientation of shuffled genomic fragments in a Hi-C map. Specifically, the challenge is to train a graph neural network (GNN) that can learn from a set of training samples, where both the shuffled and correct orders are provided, and use this model to predict the correct order for new shuffled samples."

Working with a large dataset of MAVE studies and a range of short ML project possibilities including:

Impute missing scores to reach full saturation.
Train gene embeddings (e.g. train/finetune a VAE/transformer on specific genes using MAVE data).
Train a MAVE based variant effect predictor.

15th October – Tutorial Day 1

8:45 - 9:30 Registration + Refreshments

9:30 - 12:30 Jack Fraser-Govil, Wellcome Sanger Institute

Jack Fraser-Govil will take attendees through implementing neural networks from first principles, recreating basic deep learning techniques in their own words. He will discuss what we mean by Explainable AI, the limitations and opportunities for use in your own work.

Identify the key building blocks of, and processes within common neural network implementations
Understand how information is used and transformed by deep learning techniques
Understand different definitions of Explainable AI, and their limitations

12:30 - 13:30 Lunch

13:30 - 16:30 Simon Mathis and Maximilian Gantz, Cambridge Computer Lab

Simon Mathis and Maximilian Gantz will deliver a workshop focused on protein engineering, particularly how wet lab and dry lab practitioners can collaborate well together. They will discuss which data modalities are needed to improve current predictive models, which data formats are suitable for AI-based interpretation and extrapolation, and the potential pitfalls to avoid.

Understand the limitations of data generation workflows and identify how to improve them
Understand which data modalities are needed to improve current models
Understand the needs of wet lab and dry lab scientists in collaborations
Understand common pitfalls in collaborations between wet lab and dry lab biologists and how to avoid them

16th October – Tutorial Day 2

9:30 - 12:30 Moises Sotelo, University of Guadalajara

Building on the attendees' experience in deep learning, Moises Sotelo will be giving a workshop focusing on genomic sequence classification using PyTorch, and publicaly available models on HuggingFace. Moises will lead attendees through implementing these models and discuss evaluation criteria and interpretation of results.

Understand the implementation workflow of deep neural networks to genetic research questions
Understand how to build, train or fine-tune, and evaluate deep learning models