Machine Learning Approaches for Protein Engineering: Part 1 banner

Part 1 of the PEGS Europe machine learning program delves into essential strategies and best practices small and large research groups need to employ as they strive to use machine learning tools to accelerate and optimize biologics drug discovery and development. We will explore the pros and cons of different approaches for developing and accessing high quality training data and then consider ways of using methods for “out of set” predictions that present new opportunities for ML-based studies arising out of known antigens, structures and successful campaigns. And to empower smaller companies working to compete with the substantial resources of major research organizations, a session will showcase the workflows, capabilities and successes of a set of emerging biopharma companies structured around the use of ML/AI tools as a primary R&D paradigm.

Recommended Short Course*
Monday, 4 November, 14:00 – 17:00
SC4: In silico and Machine Learning Tools for Antibody Design and Developability Predictions
*Separate registration required. See short courses page for details. All short courses take place in-person only.

Wednesday, 6 November

07:30Registration and Morning Coffee

DATA STRATEGIES

08:25

Chairperson's Remarks

Philip M. Kim, PhD, Professor, Molecular Genetics & Computer Science, University of Toronto

08:30

Scalable Active Learning for Therapeutic Antibody Design

Nathan Frey, PhD, Senior Machine Learning Scientist, Prescient Design, a Genentech Company

We will discuss our approach and general considerations for implementing active learning and design of experiments to iteratively optimise therapeutic antibody candidates. Our active learning framework is underpinned by both algorithmic innovations and robust data pipelines. We achieve improvements across binding affinity, expression yield, and developability properties via orthogonal optimisation approaches, analogous to the multitude of affinity maturation pathways observed in immune responses.

09:00

Expanding Open-Source Structure Prediction with OpenFold

Jennifer Wei, PhD, Machine Learning Software Engineer, OpenFold

The OpenFold Consortium brings together academic and industrial teams to build state-of-the-art protein structure and co-folding prediction models optimised for use on commercial computational hardware. We develop fully open-sourced models and support creation of new experimental datasets, aiming to build more powerful models that can accurately predict complex systems of significance to life sciences. In my presentation, I will present the latest modelling and software developments from the consortium.

09:30

Key Insights from Boehringer Ingelheim’s Digital Transformation Journey

Kausheek Nandy, Digital Transformation-Research, Boehringer Ingelheim Pharmaceuticals Inc.

Boehringer Ingelheim’s digital transformation journey began in 2023, focusing on three key areas. Firstly, empower scientists by liberating them from routine tasks, allowing them to concentrate on high-value work. Second, create a hub for innovation by building an in-house digital portal, a scientist-driven mechanism to formalize and standardize in silico protocols. Lastly, emphasize digital data capture FAIR and APIs, setting the stage for leveraging AI/ML in future.

10:00

AI-Driven De Novo Design of High-Affinity VHH for GPCR Targeting

Per Greisen, President, BioMap

The development of targeted biologics is often hindered by the challenges of identifying and engineering antibodies against specific epitopes. AI-powered de novo antibody design offers a promising solution, enabling precise epitope selection and sequence optimization. Here, we leverage a synergistic combination of structural sampling diffusion models and our proprietary large language model (xTrimo) to design VHH antibodies against a functional GPCR epitope.

10:30Coffee Break in the Exhibit Hall with Poster Viewing

11:15

Pioneering Data-Driven Strategies in de novo Nanobody Design

Roberto Spreafico, PhD, Director, Discovery Data Science, Genmab

AI's potential to create antibodies from scratch is promising but hampered by poor hit rates and binding strengths, rooted in insufficient training data. We have addressed this issue by using computational simulations to determine data requirements such as modality, amount, and diversity. Simulations have been guiding our ongoing experimental data generation work, marking a shift towards a data-centric strategy that complements recent algorithmic progress, aiming to overcome current challenges.

11:45

KEYNOTE PRESENTATION: Generating Data and Labels to Train AI Models for the Design of Better Therapeutic Antibodies

Yanay Ofran, PhD, Founder, CEO, Biolojic Design Ltd.

This presentation focuses on the challenges in obtaining large and well-labeled datasets for training effective AI models. High-throughput data is often not sufficiently labeled to allow for the training of good models. I will review current approaches to coping with this challenge and propose a path to generating and labeling data to train models that design better antibodies that do things that traditionally discovered antibodies are unlikely to do.

12:15 LUNCHEON PRESENTATION: Cutting through the Hype: Real-World Applications of AI in Antibody Discovery and Engineering

Mary Ann Pohl, Director of Alliance Management, Biologics Discovery, XtalPi Inc.

Artificial intelligence (AI) is transforming antibody discovery and engineering. Ailux's platform synergistically combines the best of wet lab and AI. We will explore a series of case studies that exemplify the applications of our AI-driven approach for tackling difficult GPCR targets, designing next-gen display libraries, predicting Ab-Ag complex structures and engineering challenging molecules. This presentation provides a realistic and evidence-based perspective on AI’s impact on the industry.

12:45Luncheon in the Exhibit Hall with Poster Viewing

INVERSE FOLDING MODELS

13:45

Chairperson’s Remarks

Amir P. Shanehsazzadeh, Artificial Intelligence Scientist, Absci Corp.

13:50

Antibody CDR Design by Ensembling Inverse Folding with Protein Language Models

Rahel Frick, PhD, Investigator, GSK

Antibody design is a multi-parameter optimisation problem that integrates data from multiple sources, such as high-resolution structures and sequence libraries. Here we show that predictions from multiple independently-trained machine learning models (ProteinMPNN, ESM, AbLang) can be easily and effectively combined when redesigning antibodies, and that doing so retains the strengths but not the weaknesses of each ML method in isolation.

14:20

Improved Antibody-Antigen Interaction Prediction Using Inverse Folding Latent Representations

Paolo Marcatili, PhD, Head, Antibody Design, Novo Nordisk

Inverse folding (IF) and protein large language models (pLLMs) have become useful tools for antibody variant generation, with generally good performance, but limited ability to find mutations that enhance the binding to the antigen. Here, we show how IF models can be used to predict B cell epitopes, how to extend this approach to estimate antibody-antigen interaction energy and find mutations that increase affinity, and to fine tune this model to increase its predictive power.

14:50

Scalable, Robust and easy to use Generative AI for engineering Novel Biologics

Stef Van Grieken, Co founder & CEO, Cradle

Generative machine learning methods can accelerate development timelines for novel biologics. At Cradle we have developed a software platform for biologics that any scientist without knowledge of machine learning can use across hit-identification, hit-to-lead and lead-optimisation. The platform enables multi-property optimization for various modalities and effectively leverages wet lab data to continuously improve outcomes (i.e. lab-in-the-loop). Using several real-world antibody and vaccine development case studies, this talk will highlight Cradle's approach and discuss which limitations we overcame to develop a robust, scalable and easy-to-use generative system for protein design that is used across dozens of campaigns at companies like Johnson & Johnson, Novo Nordisk and Grifols.

15:05

Featured Poster Presentation: Interpretable Prediction of Antibody Binding Affinity Exploiting Normal Modes and Deep Learning

Kevin Michalewicz, PhD, Research Postgraduate, Mathematics, Imperial College London

The high binding affinity of antibodies towards their cognate targets is key to eliciting effective immune responses. We propose ANTIPASTI, a Machine Learning approach that achieves state-of-the-art performance in the prediction of antibody binding affinity using as input Normal Mode correlation maps. The learnt representations are interpretable: they reveal similarities of binding patterns among antibodies and can be used to quantify the importance of antibody regions contributing to binding affinity.

15:20Transition to Plenary Keynote Session

PLENARY DEEP DIVE

15:30

Chairperson's Remarks

Christian Klein, PhD, CXO in Residence and Drug Hunter, Curie.Bio

15:35

Immunotherapy Highlights 

Taruna Arora, PhD, Formerly Vice President, Biotherapeutics, Bristol Myers Squibb

15:45

Multispecific Antibody Highlights 

Tomoyuki Igawa, PhD, Vice President, Discovery Research Division, Chugai Pharmaceutical Co.,Ltd

15:55

ADC Highlights 

Hironori Matsunaga, PhD, Scientist, Discovery Research Lab I Group II, Daiichi Sankyo Co., Ltd.

PLENARY PANEL

16:05

Shaping the Next Stage of Antibody Development with Complex Modalities and Combinations

PANEL MODERATOR:

Christian Klein, PhD, CXO in Residence and Drug Hunter, Curie.Bio

In the past, the field of therapeutic antibodies was dominated by monoclonal antibodies. Notably, during the past decade, novel antibody based modalities including Fc-engineered antibodies, antibody drug conjugates, bispecific and multispecific antibodies, antibody fusion proteins, immunocytokines and antibody-like scaffolds have emerged and reached clinical trials and patients with increasing speed and numbers in diverse areas including oncology, hematology, immunology, autoimmune diseases, infection, CNS and metabolic disorders, ophthalmology. Similarly, today, antibody combinations have been approved and numerous antibody-based therapies are combined in clinical trials. In the Plenary Fireside Chat "Shaping the Next Stage of Antibody Development with Complex Modalities and Combinations", renowned experts in the field will discuss major breakthroughs and how the field will evolve in the years to come.

PANELISTS:

Taruna Arora, PhD, Formerly Vice President, Biotherapeutics, Bristol Myers Squibb

Tomoyuki Igawa, PhD, Vice President, Discovery Research Division, Chugai Pharmaceutical Co.,Ltd

Hironori Matsunaga, PhD, Scientist, Discovery Research Lab I Group II, Daiichi Sankyo Co., Ltd.

16:35Refreshment Break in the Exhibit Hall with Poster Viewing

PROGRAM UPDATES FROM AI-CENTRIC BIOPHARMAS

17:15

Development of ABS-101: A Potential Best-in-Class Anti-TL1A Antibody for the Treatment of Inflammatory Bowel Disease

Douglas Ganini da Silva, PhD, Director, Purification & Analytics, Absci Corp

We describe the development of ABS-101, a potential best-in-class anti-TL1A antibody for the treatment of inflammatory bowel disease. Generative AI models were leveraged to design antibodies against desired epitopes to achieve reduced immunogenicity risk and binding to both monomeric and trimeric TL1A. AI-guided lead optimization then produced several high-affinity candidates with high cellular potency, desirable developability properties, and cross-reactivity to non-human primate (NHP) species.

17:45

De novo Design of Miniprotein-Based NK Cell Engagers

Mireia Solà Colom, PhD, Investigator and Head, Immunotherapeutics, AI Proteins

Bispecific immune cell engagers have emerged as a highly effective new therapeutic modality for oncology. Current engagers are limited by their reliance on immunoglobulin proteins, which restricts the valency, geometry of binding, developability, and speed of engineering. We solved these challenges by leveraging de novo-designed miniproteins, which enabled us to rapidly create and optimise highly potent NK cell engagers for AML that control tumour growth using xenograft models.

18:15

Method Development and Application of Machine Learning to Rapidly Reduce the Immunogenicity of Bacterial Proteases That Degrade Pathogenic Immunoglobulins

Ryan Peckner, PhD, Director, Machine Learning, Seismic Therapeutic

We develop and apply machine learning models to optimise in parallel multiple drug-like properties of the bacterial enzyme IdeS, to design a therapeutic for chronic autoantibody-mediated diseases, while minimising its immunogenicity and other liabilities. The success of this approach is demonstrated via in vivo and in vitro assays, and we illustrate its generalizability by engineering non-immunogenic bacterial cysteine proteases with a variety of immunoglobulin isotype specificities. Finally, we present a general mathematical framework enabling this process to be applied to virtually any naturally occurring protein in silico, using only the protein sequence as input.  

18:45Close of Machine Learning for Protein Engineering: Part 1 Conference