Part 1. Fine-tuning biological LLMs for functional classification
Intructor: Ernest Mordret (postdoctoral researcher at the Institut Pasteur)
Large Language Models (LLMs) are revolutionizing biology! They capture intricate patterns within DNA and protein sequences through an extensive and costly pre-training phase. With the right downstream task, valuable information can be efficiently extracted from these “foundation models” using limited resources. Join us to master the art of fine-tuning!

Objective: In this workshop, you will explore efficient techniques for fine-tuning LLMs. We will discuss the challenges and best practices associated with applying deep learning methods to biological sequences. Additionally, we will delve into interpreting your fine-tuned model’s decisions using explainable AI methods.
Workshop overview: Ernest Mordret will provide an overview of Natural Language Processing (NLP) methods applied to biological function determination, focusing on his recent work on the “phage defense” phenotype. You will be introduced to the fine-tuning ecosystem of Hugging Face and guided through fine-tuning your own LLM.
Part 2. Use an AI Foundation model on single-cell RNA-seq data
Instructor: Jérémie Kalfon (doctoral researcher in the Machine Learning for Integrative Genomics team at the Institut Pasteur)
How can you use foundation models (LLMs) for single-cell RNAseq data? From deciding what the right run environment is to classifying cell types and fine-tuning the model on your own data and objectives.

extracted and modified from (Cui et al. 2024)
Workshop overview: Jérémie Kalfon will guide you through installing and applying a Foundation Model to predict novel features from a single-cell RNAseq dataset. Depending on your level of expertise, this can go from understanding how to run the model up to creating a novel fine-tuning task for your own specific application and dataset.
Key words: Fine-tuning, LoRA, Protein Language Models, scRNAseq, Foundation models, classification, embedding, gene networks. Preprequisites: Python, intro to AI.
The workshop is open for all pasteurians. For more information and registration, please, visit this webpage.