When the Simons Foundation convened a panel of distinguished scientists at the Buttermilk Falls Inn in upstate New York in June 2012, the goal was to explore the merit of supporting large-scale collaborations in different areas of science. Most panelists described research directions in their own disciplines that seemed ripe for such collaborations. But one attendee — mathematician Ingrid Daubechies of Duke University — proposed something entirely different: an institute dedicated to developing mathematical tools and software to help scientists extract meaning from the gigantic datasets that have become a prominent feature of modern science.
The idea was outside the scope of what Jim and Marilyn Simons had been looking for — yet it instantly resonated with them. “What was beautiful was that she just came with her best idea,” Marilyn Simons says. “It wasn’t what we had asked for, but what’s great about being a foundation is that we didn’t have to stick to what we had asked for.”
Jim Simons had built his career developing mathematical models to handle large datasets, first as a cryptanalyst at the Institute for Defense Analyses and later at the hedge fund firm he founded, Renaissance Technologies. So Daubechies’ proposal “was right up my alley,” he says.
Rather than establishing such an institute externally at a university, Jim Simons decided to build it in-house. Doing so, he felt, would allow him to contribute to the exchange of ideas. In 2013, the foundation inaugurated the Simons Center for Data Analysis (SCDA) with a mission to tackle problems in computational biology. Three years later the center was reborn as the much larger Flatiron Institute, which occupies an entire building across the street from Simons Foundation headquarters in New York City.
From the start, it was clear that the institute could play a vital role in the advancement of science, one that most academic institutions were not well equipped to carry out themselves. While university researchers increasingly generate large datasets, the structure of academia is not conducive to developing the tools needed to analyze them. Creating such tools can take years of sustained effort and requires expertise in both programming and big-data analysis. Few research scientists possess these skills, and their grants are typically not set up to support this kind of project. As a result, scientists often delegate programming tasks to graduate students who are not necessarily expert programmers, and then once the graduate student moves on, there’s no one left who knows how to maintain and improve the software.
“The timescales of the problems that need to be addressed and the kinds of people needed to solve these problems are not well aligned with the academic credit system,” says Leslie Greengard, SCDA’s inaugural director, who now directs the Flatiron Institute’s Center for Computational Mathematics (CCM).
SCDA set out to fill this void. Six years later, what started as a relatively modest operation has grown into an institute that employs more than 150 computational scientists and software engineers, with further growth in the works. The institute’s mandate has broadened beyond data analysis to encompass simulations and other algorithms for advancing the frontiers of science.
The institute now comprises four centers that focus on different areas of science. Jim Simons took an active role recruiting the leaders of each center. “My experience in building an organization is to find great leadership and let them carry the ball,” he says.
The Center for Computational Biology (CCB), led by Michael Shelley, develops modeling tools and theoretical methods for examining biological data whose scale and complexity have historically resisted analysis. “The vast amounts of data being generated in biology are overwhelming people’s ability to make sense of it,” Greengard says.
The Center for Computational Astrophysics (CCA), led by David Spergel, is developing computational frameworks to help researchers understand complex astrophysical systems ranging in scale from planetary systems to the entire early universe. The center hosts many conferences and workshops, to which it invites the broader astrophysics community.
The Center for Computational Quantum Physics (CCQ), led by Antoine Georges and Andrew Millis, focuses on the quantum many-body problem, which concerns microscopic systems with many interacting particles. One ultimate goal is to design new materials with desirable properties, such as superconductors.
“Any little piece of material has about 1021 electrons, so we’re talking about humongously large quantum systems,” Georges says. “It’s a very challenging problem that has been around for many decades, but computational methods have really changed the game in this field.”
CCM, the most recent addition, has a twofold mission: to keep the institute at the leading edge of theoretical advances in data analysis and to build bridges among the other three centers.
“For the Center for Computational Mathematics, the driving force is the methods themselves,” Greengard says. “That makes it the natural glue for this enterprise.”
In addition to the four centers, the institute hosts a scientific computing core that currently consists of nine software engineers and data scientists, led by Nick Carriero and Ian Fisk. This group helps the institute’s researchers develop algorithms, store data, and access the institute’s basement supercomputer and allied supercomputers at Brookhaven National Laboratory and the University of California, San Diego.
“Typically, university scientists trying to figure out how to get their stuff running on supercomputers don’t have a lot of support,” Greengard says. “It’s somewhat painful in most environments, but it’s done extremely well here.”
Flatiron scientists don’t have to go chasing after grants, and this freedom enables them to pursue risky, long-term goals that might make a huge impact. “It allows me to really get into projects, without worrying about these barriers,” says Olga Troyanskaya, who leads the genomics group in the CCB.
In her case, that has meant creating (together with Aaron Wong and other colleagues) a vast data repository called HumanBase, which merges thousands of genomics and gene expression datasets and uses machine-learning algorithms to discover how molecular circuitry functions in every human tissue. “I’ve wanted to build this ever since I became a professor,” Troyanskaya says.
Even though the institute is still quite young, HumanBase is far from the only Flatiron-built tool that research scientists are starting to find indispensable. For example, one CCB team has developed software platforms to make sense of the cacophony of neuronal signals picked up by electrodes implanted into animal brains. Another team that started at the CCB and moved to the CCM is developing software to analyze movies from specialized microscopes that capture the activity of large brain regions at single-neuron resolution. Meanwhile, the CCQ is developing a wide range of different attacks on the many-body problem and making the code available to the public via open-source libraries. And CCA researchers recently presented a new project that uses machine learning to generate accurate and lightning-fast simulations of the universe, one of the major objectives of modern theoretical astrophysics.
The variety of disciplines represented at the institute makes for a vibrant atmosphere, Georges says. “It’s really unique to have scientists under the same roof who share a passion for computational methods but who come from very different fields of science,” he says. “They would probably not meet in any other place.”
This dynamism spills over into the Simons Foundation at large, Marilyn Simons says. Before the institute arrived at the foundation, “we had an office job,” she says. “But with science here, you feel allied to the mission. Everybody’s enthusiasm and originality and joy in their work is infectious.”
Plans are in the works for a fifth center dedicated to computational neuroscience. With that, the Flatiron Institute will be at its planned capacity. “Nothing that I have been instrumental in creating has pleased me more than has the Flatiron Institute,” Jim Simons says. “It is a unique organization, and I have every reason to believe it will last far into the future doing outstanding and impactful science.”