The Challenge
Enabling Collaboration in Drug Discovery While Protecting Proprietary Data
The pharmaceutical industry faced a critical challenge: improving machine learning (ML) models for drug discovery while maintaining the confidentiality of proprietary datasets. Traditionally, pharma companies developed ML models in isolation due to concerns over data privacy and intellectual property protection. However, this approach resulted in suboptimal models, as ML models improve with larger and more diverse datasets. In other words, pharmaceutical companies could achieve better results by training ML models on combined data. Yet, companies feared exposing sensitive information to their competitors - preventing collaboration and slowing down scientific progress.
The MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) project aimed to break this impasse by enabling 10 leading pharmaceutical companies - including Bayer, GSK, and Novartis - to collaboratively train ML models without sharing raw data.
The Solution
"Coopetition"
The MELLODDY team turned to federated learning, an approach that allows ML models to learn from multiple datasets without ever transferring raw data. Instead of pooling sensitive information in a central database, each company kept its data within its own secure environment. In this project, models were trained on distributed data across companies without centralizing it, creating a new form of “coopetition”. The Kubermatic Kubernetes Platform (KKP) was used to build the scalable Kubernetes infrastructure for each pharma partner. This infrastructure enabled partners to register and use their proprietary datasets locally, allowing private models to learn from the combined knowledge without sharing sensitive data. For the first time, competing pharmaceutical companies could collaborate on their research without compromising security.
The Impact
Paving the way for data-sharing in Pharma
MELLODDY proved that pharma companies don’t have to choose between privacy and progress. By working together under a federated learning framework, they built better predictive models that outperformed every single partner model. Therefore, the MELLODDY project has created a viable solution for pharma companies to cooperate in developing better machine-learning models.
This project set the stage for a future where AI-driven breakthroughs can happen faster - resulting in more accurate models and faster drug development.
MELLODDY was a European Innovative Medicines Initiative (IMI) that gathered 10 pharmaceutical companies, academic research labs, large industrial companies, and startups, including Bayer, GSK, Novartis and Kubermatic. The MELLODDY (Machine Learning Ledger Orchestration for Drug Discovery) platform was the first industry-scale platform to enable the creation of a global federated model for drug discovery without sharing the confidential data sets of individual partners. It was a groundbreaking project that allowed Machine Learning models to be trained with data from several pharmaceutical partners, maintaining the confidentiality of the data.
The MELLODDY project is a groundbreaking collaboration that has the potential to accelerate drug discovery and improve patient outcomes by enabling, for the first time, research to be conducted across the consortium's decentralised and highly proprietary databases of annotated chemical libraries. This project allows the pharma partners for the first time to collaborate in their core competitive space, invigorating discovery efforts through efficiency gains.
