Timing: 0900h Mon 1st Sept -  to 1400h Tues 2nd Sept 2025.

Location: University College Dublin main campus, Ireland.

The workshop will bring together experts working in AI and peptide research. The workshop will highlight current state of the art machine learning techniques.

Session 1 (Monday morning): AI and simulation approaches for Peptide discovery.

Session 2 (Monday afternoon): Peptide creation with low data or in uncertain environments. 

 Session 3 (Tuesday 0900-1400h, UCD): Practical workshop (peptide data partitioning and representation, predictive and generative approaches) followed by discussion and Keynote Lecture (bring your laptop). 

Confirmed Speakers:

Martin Zacharias: “Modelling Cyclic Peptides”

Patrick Timmons, Nuritas Ltd, Dublin: “Peptide Design for Protein Binding”

Pierre Tuffery: “Building scoring functions for peptide de-orphaning using Alphafold”

Raul Fernandez-Diaz: “Partitioning, representation and automation in peptide modelling”

Chandra Verma: “Development of stapled peptides“
    
Raquel Rodriguez Perez: “Uncertainty Estimation“
    
Wolfgang Banzhaf: “Evolutionary Computation based Novel Peptide Design“

BACKGROUND

The peptide global market is rapidly expanding with its value estimated at $14.4 billion, accounting for 1.5% of the total worldwide pharmaceutical market. These peptides are usually between 5 and 40 amino acids and have a diverse range of applications and many advantages, including their ability to be produced at a large scale. Significant advancements in synthetic and recombination technologies have been a driving force in bringing bio-active peptides back to center stage as therapeutic and diagnostic tools. However, similar advances in artificial intelligence (AI) and data analytics methods for peptides are needed.

Recent advances in structural protein prediction have enabled rapid design of proteins with desired structural properties. However, understanding how to design proteins with specific functional or multi-functional properties is still challenging. This is critically important for smaller proteins – peptides – where there is often less secondary or tertiary structure to predict, and the functionality has a great dependence on extrinsically bound factors.

It is clear that new computational technology is needed to meet the requirements of the community that leverages the AI advances seen in protein modelling. However, peptide datasets are generally much smaller than protein training datasets, making it harder to build relevant AI representations, especially for non-canonical amino acids. To successfully scale AI in peptide studies, collaboration is necessary that integrates activities of data collection, software development, transfer learning and implementation.

Much of the critical data during drug development of peptide leads comprises datasets of relatively few compounds in a particular class with relevant key assays. AI approaches can support iterative experimentation and computation approaches that accelerate the design of peptides of interest. 

We welcome attendees interested in these topics, and who may be interested to present a talk or poster on related topics, including but not limited to:

• Discovering peptides with target structures or functions.

• Discovering linear, cyclic or other rigid structure peptides.

• Creating peptides with unusual amino acids.

• Discovering regulatable peptides or multi-functional peptides.

• Using large language models (LLMs), protein language models. (PLMs) or chemical language models (CLMs) to improve peptide machine learning.

• Generative AI.

• Explainable AI applied to peptide analysis and prediction.

• Open-source software.

• Best practices & legal issues for peptide AI drug development.

• AI with uncertainty in assay measurement

• Active learning for selection of important peptide training cases in low data experiments.

• Human in the loop AI for fitness evaluation.

• Dataset augmentation using generative AI.

• Approximate/partial peptide discovery.

• Data curation.

• Collaboration methods.

• Visualization methods.

• GUI based tools.

• Human computer interaction design principles needed for peptide design.

• Tools to visualize/inform trade-offs in peptide functions.

• Peptide search space visualization methods.

• AI methods to ensure peptide safety (e.g. lowering toxicity, ensuring stability).

• Collaboration to overcome the hurdle of small dataset sizes, including federated learning across distributed proprietary datasets.