Merck is expanding access to TEDDY (Transformers for Enabling Drug DiscoverY), its foundation model for drug discovery, by adding it to HuggingFace, the premier platform for open-source machine learning and LLMs. 

TEDDY was created in partnership with the Boston Consulting Group (BCG) AI Science Institute and is trained on more than 116 million single cells from 24,000 donors across 122 diseases. 

It leverages cutting-edge transformer architecture to map gene regulation in a way never seen before, with the goal of targeting the right drug, to the right patient, at the right time, faster. 

It is formed of two models. TEDDY-G is trained on genomic data which has been ranked, allowing the model to predict which genes are most likely to be involved in a certain disease, based on their relationships with other genes. 

TEDDY-X is also trained on genomic data and processes gene expression by grouping the values into discrete categories, to predict which category a gene’s expression falls into. This helps the model to learn factors responsible for variations in gene expression. 

By adding TEDDY to HuggingFace, researchers worldwide will have access to powerful AI tools that Merck hopes will accelerate global innovation through transparency, collaboration, and interoperability.  

Merck says that the move is especially relevant as the White House emphasised the importance of open-source systems like HuggingFace during its announcement of the US Action Plan. 

In a further initiative, Merck and Harvard are working together to create new computer models that can combine and learn from different types of genetic data to better understand diseases that affect the immune system.

Diana Turner, Senior Digital Content Editor, DDW