Until recently, Pharma has relied on experimentation methods 100 years old.
Pharma now are rapidly adopting methods powered by machine learning such as Active Learning and Bayesian Optimisation that enable up to 80% cost savings.
This symposium gathers academia and industry to make a first step towards an ecoystem of Active Learning.
Meet with academic experts and industry innovators in a focused symposium to hear the latest case studies in the use of Active Learning/Bayesian Optimisation in Pharma.
Stay informed about the latest developments. This programme takes a deep dive into best practice approaches and latest technological advancements; bringing together thought leaders for focused discussions and networking. Discuss and facilitate knowledge exchange about the latest use-cases.
Understand current challenges and limitations. What challenges will I face in the future? What do I have to prepare for? Where are my strengths and weaknesses, compared to other teams and their challenges?
In the spirit of transparency and community building, we’re pleased to share the analytics from the ActiveLearning in Pharma Symposium. It’s been wonderful to see so many attendees engaging with and enjoying the event.
Stats: 180 Unique Attendants, 120 Attendants at 2 hour mark.
We are especially grateful for the more substantive feedback, including topic and speaker suggestions. We also want to acknowledge comments regarding diversity and inclusion. These are areas we’re committed to prioritizing in future events. To learn more about our ongoing efforts, we encourage you to explore our Active Learning Grants initiative here.
Any questions and feedback, reach us here.
SOLVE
Merck KGaA
Novo Nordisk
Bayer
MSD
Evotec
Discussion Questions:
Technical
a. How big should data-sets be before they are useful (especially for transfer learning)
b. How similar do processes need to be in order to benefit from a transfer learning approach?
c. How do we deal with nested or changing search spaces?
d. What is the “way to go” for hybrid optimization?
e. In the context where the objective can evolve at each Active Learning cycle (toward more complexity), how can we show that previously selected compounds brought valuable information in the BO process?
Community
a. What are the advantages and disadvantages of Open Source for Active Learning?
b. What improvement in your organization would provide a 10x efficiency increase in the domain or process optimization?
c. How do you structure active learning in high risk and highly-regulated environments?
Adoption
a. How can we be better at making scientists adopt closed loop experimentation?
b. How to keep engagement of the experimentalist?
c. How to convince people to continue the active learning process when the first cycle (i.e. explorative) lead to disappointing results?
d. When do you use a classical DoE approach (fractional factorial, D-optimal etc.) and when do you decide for BO? Or do you usually combine both approaches?
- Principal Scientist (Solid Formulation) at top 10 Pharma Company
Pharma Companies have built Bayesian Optimisation teams since 2022
Estimated average annual savings from Bayesian Optimisation
Average BO performance improvement with Transfer Learing Bayesian Optimisation
Time to implement BO in established applied maths teams in Pharma
14:00 10-minute Keynote: How to build an ecosystem for Active Learning in Pharma
14:10 5-minute Lightning Talks (Opportunity for teams to introduce themselves and their work)
Novo Nordisk: Rune Christensen
Bayer: Giulio Volpin & Timo Wolf
Merck Germany: Adrian Šošić & Alexander Hopp
Exscientia: Jonathan Harrison
MSD (Merck Sharpe & Dohme): Kevin Stone
TriNova: Thomas Casey
Evonik: Johannes Peter Dürholt on BoFire
Acceleration Consortium: Sterling Baird
SOLVE: Jose Folch
Evotec: Lionel Colliandre
15:00 Discussion Round:
OpenSource for Active Learning Software: Where are we going as a community?
How do you structure active learning in high risk and highly-regulated environments?
How similar do processes need to be in order to benefit from a transfer learning approach?
What improvement in your organization would provide a 10x efficiency increase in the domain or process optimization?
How can we be better at making scientists adopt closed loop experimentation?
How do we deal with nested or changing search spaces?
What is the “way to go” for hybrid optimization?
How to keep engagement of the experimentalist?
How to convince people to continue the active learning process when the first cycle (i.e. explorative) lead to disappointing results?
In the context where the objective can evolve at each Active Learning cycle (toward more complexity), how can we show that previously selected compounds brought valuable information in the BO process?
When do you use a classical DoE approach (fractional factorial, D-optimal etc.) and when do you decide for BO? Or do you usually combine both approaches?
how big should data-sets be before they are useful?
16:00 Break/Buffer
16:20 Talks: Latest Advancements in Algorithms (UCL, ETH)
"Applying Multi-Fidelity Bayesian Optimization in Chemistry: Open Challenges and Major Considerations", Mohammed Azzouzi, EPFL (Paper)
"Sample-efficient Bayesian Optimisation Using Known Invariances", Theo Brown, UCL (Paper)
16:50 Talks: Latest Advancements in Applications
"Automated Discovery of Pairwise Interactions from Unstructured Data", Jason Hartford, Valence Labs/Recursion Pharmaceuticals (Paper)
"Self Driven Media Optimization Using Multi-Objective Bayesian Optimization With Laboratory Automation", Dania Awad, WSSB (Presentation)
17:25 Conclusion and 2025 Symposium Announcement
If you'd like to post an Active Learning role here (Bayesian Optimisation, Lab Automation etc.), please reach out to us.
Two Positions at Crop Science in Frankfurt am Main:
Data Scientist (all genders) - Active Learning and Small Molecules Design: https://lnkd.in/eB3s6DQs
Data Scientist (all genders) - Probabilistic Modelling and Phenotyping https://lnkd.in/ezhFYRi5
Two Positions at Pharma in Berlin (one link/advert for both):
Research Scientist (all genders) - Machine Learning for Small Molecules https://lnkd.in/eST2XEp3
Unified Laboratory Informatics & Management Software made for enterprise R&D teams.
Automata empowers automation teams to transform their labs into data powerhouses with LINQ, a cloud-native, end-to-end automation platform.
TBC
It all started with the New Orleans West Inn hotel manager giving us the penthouse ballroom for free for one night back in 2023!
At NeurIPS 2023, we at Matterhorn Studio hosted an informal gathering of Bayesian Optimisation & Lab Automation enthusiasts. It was an incredible evening of deep discussions, idea-sharing, and cross-industry insights. Here we:
Brought together experts from ML, science, and industry
Bridged AI-driven optimisation with real-world applications
Created a space for innovation and future collaborations
Matterhorn Studio has raised £1000 towards Active Learning Grants, and is hoping to raise more during the Active Learning U.K. symposium.
These initial grants will be given to the most promising African research proposals in Active Learning.
We enable high-performance Machine Learning for high-performance teams of scientists in Pharmaceuticals, Chemistry and Materials.
Based in Oxford, UK, we pride ourselves in Our Values of scientific excellence and social responsibility.