What is the objective of the project?

To evaluate whether learning across tasks, data types and partners improves the predictive ability of a model. In other words, is it possible to train a model on a multi-partner dataset without leaking IP-sensitive information and will this model be useful to predict in a more effective way which compounds might be promising in the later stages of drug discovery and development?

Why is this project important?

By virtualizing parts of the drug discovery process through machine learning and better predicting the chances of success that a compound has in the discovery and development process, MELLODDY aims to improve the overall efficiency of bringing a therapeutic to the market, which on average can take about 13 years and cost €1.9 billion[1]. As an IMI2 funded project, MELLODDY contributes to the core mission of IMI2, which is “bringing the right treatment to the right patient at the right time without adverse effects and outcomes.”

[1] DIMasi JA et al., 2016. Innovation in the pharmaceutical industry: new estimates of R&D costs. Journal of Health Economics 47, 20-33.

What kind of efficiencies do we hope to achieve?

Since the project is just kicking off, it is too early to speculate on specifics. Our hypothesis is that the MELLODDY privacy-preserving federated machine learning platform will help pharmaceutical R&D to explore fewer drug candidates that are of a higher overall quality, therefore likely saving time and costs.

How will the platform work? And how does MELLODDY define federated learning?

• MELLODDY aims to train machine learning models across multi-partner datasets while ensuring privacy preservation of both the data and the models by developing a platform using federated learning.
• The MELLODDY platform uses Amazon Web Services technologies in order to execute Machine Learning algorithms from academic partners on a large scale.
• The data never leaves the owner’s infrastructure and only non-sensitive models are exchanged. A central dispatcher allows each partner to share a common model to be consolidated collectively. To provide full traceability of the operations, the platform is based on a private blockchain.
• This means that a ledger will be distributed across all contributing pharma partners in such a way that there is no central authority. The platform guarantees by design that partners keep control and visibility over their own private data. Since there is no central authority, any communication between the dispatcher and a ledger needs to be approved by all partners before one can proceed. Like a bank statement, the ledger holds a log of all activities and can be requested after a federated run.
• The MELLODDY platform is designed to prevent the leaking of proprietary information from one data set to another or through one model to another while at the same time boosting the predictive performance and applicability domain of the models by leveraging all available data.

Can additional partners join the MELLODDY consortium or a federated run?

As MELLODDY is approaching its final project year (starting 1st June 2021), we regret that it is no longer feasible to consider partner accession. Hoping that you are interested in our sustainability efforts we encourage you to sign up to our future newsletter.

Who are the MELLODDY Partners?

Check out our Partners Page to find more information. The consortium is comprised of 17 partners:

• 10 pharma partners: Amgen, Astellas, AstraZeneca, Bayer, Boehringer Ingelheim, GSK, Janssen Pharmaceutica NV, Merck KGaA, Novartis, and Institut de Recherches Servier.
• 2 academic universities: Budapesti Muszaki es Gazdasagtudomanyi Egyetem, KU Leuven.
• 4 small and medium-sized enterprises (SMEs): Iktos, Kubermatic (ex-Loodse), Owkin, Substra Foundation.
• 1 large AI computing company: NVIDIA

The MELLODDY project is under the direction of Project Lead Hugo Ceulemans, Janssen Pharmaceutica NV, and Project Coordinator Mathieu Galtier, Owkin.

How is MELLODDY funded?

• The project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No 831472.
• This Joint Undertaking receives support from the European Union’s Horizon 2020 research and innovation programme and EFPIA.
• The overall budget estimated for the MELLODDY project is €18,4 million, of which the EFPIA partners contribute €10 million in kind, the public partners receive a maximal grant amount of €8 million from the EU, and NVIDIA contributes another €120,000.
• All figures are estimates and can be subject to change.

What do you use for compute infrastructure management?

Kubernetes has become the IT industry standard for cloud application managment and application orchestration. MELLODDY leverages Kubermatic's software and expertise to use Kuberenetes as the secure computing base.