How the Novo Nordisk Foundation Center at Broad is using scalable technologies and AI to find the root cause of diabetes

Since the earliest days of her scientific career, Melina Claussnitzer, a metabolic disease researcher at the Novo Nordisk Foundation Center for Genomic Mechanisms of Disease (NNFC) at the Broad Institute, has been thinking about how to accelerate the discovery of the biological mechanisms driving diabetes and obesity. She knows first-hand how slow and painstaking it can be to translate genetic variants linked to increased disease risk into biological insight. She spent several years in the lab as a graduate student figuring out the function of just one variant associated with obesity. 

But there are thousands  of such variants linked to obesity and type 2 diabetes alone, and understanding what all these variants do in cells and tissues to increase disease risk requires a new, more scalable, and systematic approach. 

Now Claussnitzer, through her efforts at the NNFC and collaborations with Denmark-based scientists, is implementing her “variant-to-function” framework, which is already scaling up the work of moving from genes to disease mechanisms. An institute member at Broad, director of the Broad’s Diabetes Initiative, and associate director of scientific strategy at the NNFC, Claussnitzer and her collaborators are using this approach to systematically uncover how certain genetic risk variants affect cellular processes that lead to type 2 diabetes. 

To analyze the enormous amounts of complex biological data, they are developing new AI tools and partnering with their NNFC colleague, Simon Rasmussen, a professor at the University of Copenhagen. He and his lab specialize in building AI algorithms that efficiently integrate and analyze big, multifaceted datasets. 

For more than two years, Claussnitzer and Rasmussen have been meeting regularly to brainstorm ideas and projects; their groups freely share scientific tools, data, and information; and Rasmussen lab members routinely visit the Broad for training and scientific exchange. 

“The Novo Nordisk Foundation Center is about bridge-building and it is bringing together scientists from very different fields,” said Claussnitzer, who also leads the Diabetes Initiative at the Broad. “Working with Simon has been wonderful, and together we’re figuring out exactly how genetic risk converges on cellular programs in metabolic disease.”

“This collaboration would not have been possible without the Novo Nordisk Foundation Center,” noted Rasmussen. “You get a level 10 access to the Broad. You get to work with new types of data and talk to different researchers. It expands one's mind in terms of scientific questions to ask.”

A key question that Claussnitzer and Rasmussen are collaboratively addressing at the NNFC is how human genetic variation impacts type 2 diabetes, which affects more than 463 million people worldwide. This heterogeneous disorder affects patients in a variety of ways, and answers will not only expand our knowledge of disease biology but also bring us closer to more effective and customized treatments. 

A key first step in Claussnitzer’s variant-to-function framework is to develop disease-relevant models that researchers use in the lab to probe the function of genetic variants associated with disease. To study diabetes, the NNFC scientists in Claussnitzer’s lab have created fat cell-based model systems. Fat cells, or adipocytes, are a powerful model for studying type 2 diabetes because the amount and distribution of fat tissue a person has can greatly influence their risk of developing type 2 diabetes. 

To create these models, Claussnitzer, in collaboration with the Massachusetts General Hospital Weight Center, is building and leveraging a unique resource—CellGenBank, a population-scale biobank containing adipose (body fat)-derived stem cells from thousands of individuals with and without metabolic disease. 

The team is now studying how their fat cell models move from a healthy state to an intermediate, at-risk one, and then finally to a diseased state. They are exposing hundreds of cells to a range of stimulatory conditions, such as high levels of glucose or insulin, to push these cells into the diseased state. The scientists then conduct large-scale genetic and cellular profiling on these individual cells to uncover which gene regulation and cellular programs are at play during disease progression. 

At the same time, they are also carrying out large-scale multi-omic profiling on thousands of fat cells from CellGenBank that have been derived from patients with metabolic disease and high-risk individuals to find the key pathways driving disease. 

While the work is still ongoing, early results have been encouraging. “We are identifying fundamental biology in cells that matter in disease, which is particularly relevant for finding genetically anchored therapeutic targets,” said Claussnitzer. “This is a massive project with different layers of information and a unique dataset.” 

*****

By making multiple measurements on thousands of single cells, the NNFC researchers are generating huge datasets that require novel AI algorithms for meaningful analysis, including those built by the Rasmussen Lab. Recently, the University of Copenhagen group created a powerful AI algorithm, called MOVE, that can robustly integrate a wide range of different kinds of data to infer clinically relevant information. 

MOVE, a deep learning generative model, works by initially learning to recognize patterns within different datasets and then compressing them into a large “cocktail.” Using what it’s learned, the model then generates new data as answers to questions posed by researchers. Rasmussen says the answers would be impossible to obtain by analyzing the individual datasets separately. 

“We let the networks figure out connections between the different datasets and learn what's important across the data,” Rasmussen said. “And now, we can modify the inputs and see how the model thinks it will affect the output.”

His team has used MOVE to consolidate and analyze data from almost 800 individuals with type 2 diabetes, spanning 12 different types of clinical data related to genomics, metabolomics, proteomics, questionnaire surveys, organ scans, medication usage, and more.  

A graduate student from Rasmussen’s group, Marc Pielies Avelli, is currently visiting the Claussnitzer lab, where he is applying MOVE to an initial tranche of type 2 diabetes data from the Broad-based team, including large-scale transcriptomic data from adipocytes, polygenic risk scores, and disease progression, to predict how genetic variants modulate fundamental biological processes in disease. 

Specifically, they are asking this data-trained MOVE model to answer crucial questions, such as which cellular programs might be affected if a variant is altered? And how can changes in adipocyte distribution influence disease progression?  

The teams are now gleaning novel insights into disease risk as well as cellular programs that have not been previously associated with metabolic disorders. As Claussnitzer’s group advances and scales up their efforts, Rasmussen and his team are refining their algorithms to analyze larger and more complicated datasets. 

The aim of both the Claussnitzer and Rasmussen groups is to use their AI model predictions to guide experiments in the lab and to feed new models and drive the development of even better algorithms that more accurately predict causal mechanisms of diabetes.

“In our first phase of work, we have put capabilities in place to systematically traverse the path from metabolic disease genetics to biology. We now need to move in a translational direction—scale up, build better models, make sense out of the data, and plug it into the clinic,” explained Claussnitzer. “We believe that the NNF Center is the future of genomic medicine, and both Simon and I are excited to be a part of it. This collaboration is just the start for us. We envision accomplishing much more moving forward.”